data_exploration.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Exploring the annotated data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "f276c3d2dbc5432eb07f1f363404bfc7",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Resolving data files:   0%|          | 0/95 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "f09865809ad74657ac1504a0eb32ec85",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Downloading data:   0%|          | 0/95 [00:00<?, ?files/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "a59200f458d24ae4bd2d01dc75307d72",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Generating train split:   0%|          | 0/7939 [00:00<?, ? examples/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "ename": "DatasetGenerationCastError",
     "evalue": "An error occurred while generating the dataset\n\nAll the data files must have the same columns, but at some point there are 30 new columns ({'image_code', 'image_inputs', 'buffer_b_code', 'cube_a_inputs', 'author', 'license', 'date', 'buffer_d_code', 'buffer_d_inputs', 'sound_code', 'published', 'likes', 'buffer_a_code', 'buffer_a_inputs', 'buffer_b_inputs', 'tags', 'thumbnail', 'functions', 'sound_inputs', 'test', 'buffer_c_inputs', 'cube_a_code', 'description', 'common_code', 'buffer_c_code', 'id', 'name', 'viewed', 'time_retrieved', 'access'}) and 1 missing columns ({'text'}).\n\nThis happened while the json dataset builder was generating data using\n\nD:/Dokumente/Uni_OFFLINE/SoSe2024/shadertoys-dataset/./data/annotated/20k_2014-02.jsonl\n\nPlease either edit the data files to have matching columns, or separate them into different configurations (see docs at https://hf.co/docs/hub/datasets-manual-configuration#multiple-configurations)",
     "output_type": "error",
     "traceback": [
      "\u001b[1;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[1;31mCastError\u001b[0m                                 Traceback (most recent call last)",
      "File \u001b[1;32mc:\\Users\\Jan\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\datasets\\builder.py:1869\u001b[0m, in \u001b[0;36mArrowBasedBuilder._prepare_split_single\u001b[1;34m(self, gen_kwargs, fpath, file_format, max_shard_size, job_id)\u001b[0m\n\u001b[0;32m   1868\u001b[0m \u001b[38;5;28;01mtry\u001b[39;00m:\n\u001b[1;32m-> 1869\u001b[0m     \u001b[43mwriter\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mwrite_table\u001b[49m\u001b[43m(\u001b[49m\u001b[43mtable\u001b[49m\u001b[43m)\u001b[49m\n\u001b[0;32m   1870\u001b[0m \u001b[38;5;28;01mexcept\u001b[39;00m CastError \u001b[38;5;28;01mas\u001b[39;00m cast_error:\n",
      "File \u001b[1;32mc:\\Users\\Jan\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\datasets\\arrow_writer.py:580\u001b[0m, in \u001b[0;36mArrowWriter.write_table\u001b[1;34m(self, pa_table, writer_batch_size)\u001b[0m\n\u001b[0;32m    579\u001b[0m pa_table \u001b[38;5;241m=\u001b[39m pa_table\u001b[38;5;241m.\u001b[39mcombine_chunks()\n\u001b[1;32m--> 580\u001b[0m pa_table \u001b[38;5;241m=\u001b[39m \u001b[43mtable_cast\u001b[49m\u001b[43m(\u001b[49m\u001b[43mpa_table\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_schema\u001b[49m\u001b[43m)\u001b[49m\n\u001b[0;32m    581\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39membed_local_files:\n",
      "File \u001b[1;32mc:\\Users\\Jan\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\datasets\\table.py:2292\u001b[0m, in \u001b[0;36mtable_cast\u001b[1;34m(table, schema)\u001b[0m\n\u001b[0;32m   2291\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m table\u001b[38;5;241m.\u001b[39mschema \u001b[38;5;241m!=\u001b[39m schema:\n\u001b[1;32m-> 2292\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mcast_table_to_schema\u001b[49m\u001b[43m(\u001b[49m\u001b[43mtable\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mschema\u001b[49m\u001b[43m)\u001b[49m\n\u001b[0;32m   2293\u001b[0m \u001b[38;5;28;01melif\u001b[39;00m table\u001b[38;5;241m.\u001b[39mschema\u001b[38;5;241m.\u001b[39mmetadata \u001b[38;5;241m!=\u001b[39m schema\u001b[38;5;241m.\u001b[39mmetadata:\n",
      "File \u001b[1;32mc:\\Users\\Jan\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\datasets\\table.py:2240\u001b[0m, in \u001b[0;36mcast_table_to_schema\u001b[1;34m(table, schema)\u001b[0m\n\u001b[0;32m   2239\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m table_column_names \u001b[38;5;241m<\u001b[39m\u001b[38;5;241m=\u001b[39m \u001b[38;5;28mset\u001b[39m(schema\u001b[38;5;241m.\u001b[39mnames):\n\u001b[1;32m-> 2240\u001b[0m     \u001b[38;5;28;01mraise\u001b[39;00m CastError(\n\u001b[0;32m   2241\u001b[0m         \u001b[38;5;124mf\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mCouldn\u001b[39m\u001b[38;5;124m'\u001b[39m\u001b[38;5;124mt cast\u001b[39m\u001b[38;5;130;01m\\n\u001b[39;00m\u001b[38;5;132;01m{\u001b[39;00m_short_str(table\u001b[38;5;241m.\u001b[39mschema)\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;130;01m\\n\u001b[39;00m\u001b[38;5;124mto\u001b[39m\u001b[38;5;130;01m\\n\u001b[39;00m\u001b[38;5;132;01m{\u001b[39;00m_short_str(features)\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;130;01m\\n\u001b[39;00m\u001b[38;5;124mbecause column names don\u001b[39m\u001b[38;5;124m'\u001b[39m\u001b[38;5;124mt match\u001b[39m\u001b[38;5;124m\"\u001b[39m,\n\u001b[0;32m   2242\u001b[0m         table_column_names\u001b[38;5;241m=\u001b[39mtable\u001b[38;5;241m.\u001b[39mcolumn_names,\n\u001b[0;32m   2243\u001b[0m         requested_column_names\u001b[38;5;241m=\u001b[39m\u001b[38;5;28mlist\u001b[39m(features),\n\u001b[0;32m   2244\u001b[0m     )\n\u001b[0;32m   2245\u001b[0m arrays \u001b[38;5;241m=\u001b[39m [\n\u001b[0;32m   2246\u001b[0m     cast_array_to_feature(\n\u001b[0;32m   2247\u001b[0m         table[name] \u001b[38;5;28;01mif\u001b[39;00m name \u001b[38;5;129;01min\u001b[39;00m table_column_names \u001b[38;5;28;01melse\u001b[39;00m pa\u001b[38;5;241m.\u001b[39marray([\u001b[38;5;28;01mNone\u001b[39;00m] \u001b[38;5;241m*\u001b[39m \u001b[38;5;28mlen\u001b[39m(table), \u001b[38;5;28mtype\u001b[39m\u001b[38;5;241m=\u001b[39mschema\u001b[38;5;241m.\u001b[39mfield(name)\u001b[38;5;241m.\u001b[39mtype),\n\u001b[1;32m   (...)\u001b[0m\n\u001b[0;32m   2250\u001b[0m     \u001b[38;5;28;01mfor\u001b[39;00m name, feature \u001b[38;5;129;01min\u001b[39;00m features\u001b[38;5;241m.\u001b[39mitems()\n\u001b[0;32m   2251\u001b[0m ]\n",
      "\u001b[1;31mCastError\u001b[0m: Couldn't cast\nid: string\nname: string\nauthor: string\ndescription: string\ntags: list<item: string>\n  child 0, item: string\nlikes: int64\nviewed: int64\npublished: string\ndate: string\ntime_retrieved: timestamp[s]\nimage_code: string\nimage_inputs: list<item: struct<id: string, previewfilepath: string, channel: int64, sampler: struct<filter: string, wrap: string, vflip: string, srgb: string, internal: string>, published: int64, src: string, ctype: string>>\n  child 0, item: struct<id: string, previewfilepath: string, channel: int64, sampler: struct<filter: string, wrap: string, vflip: string, srgb: string, internal: string>, published: int64, src: string, ctype: string>\n      child 0, id: string\n      child 1, previewfilepath: string\n      child 2, channel: int64\n      child 3, sampler: struct<filter: string, wrap: string, vflip: string, srgb: string, internal: string>\n          child 0, filter: string\n          child 1, wrap: string\n          child 2, vflip: string\n          child 3, srgb: string\n          child 4, internal: string\n      child 4, published: int64\n      child 5, src: string\n      child 6, ctype: string\ncommon_code: string\nsound_code: string\nsound_inputs: list<item: null>\n  child 0, item: null\nbuffer_a_code: string\nbuffer_a_inputs: list<item: null>\n  child 0, item: null\nbuffer_b_code: string\nbuffer_b_inputs: list<item: null>\n  child 0, item: null\nbuffer_c_code: string\nbuffer_c_inputs: list<item: null>\n  child 0, item: null\nbuffer_d_code: string\nbuffer_d_inputs: list<item: null>\n  child 0, item: null\ncube_a_code: string\ncube_a_inputs: list<item: null>\n  child 0, item: null\nthumbnail: string\naccess: string\nlicense: string\nfunctions: list<item: list<item: int64>>\n  child 0, item: list<item: int64>\n      child 0, item: int64\ntest: string\nto\n{'text': Value(dtype='string', id=None)}\nbecause column names don't match",
      "\nDuring handling of the above exception, another exception occurred:\n",
      "\u001b[1;31mDatasetGenerationCastError\u001b[0m                Traceback (most recent call last)",
      "Cell \u001b[1;32mIn[1], line 14\u001b[0m\n\u001b[0;32m     12\u001b[0m shader20k_files \u001b[38;5;241m=\u001b[39m [\u001b[38;5;124mf\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124m./data/annotated/\u001b[39m\u001b[38;5;132;01m{\u001b[39;00mf\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m\"\u001b[39m \u001b[38;5;28;01mfor\u001b[39;00m f \u001b[38;5;129;01min\u001b[39;00m shader20k_files \u001b[38;5;28;01mif\u001b[39;00m f\u001b[38;5;241m.\u001b[39mstartswith(\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124m20k\u001b[39m\u001b[38;5;124m\"\u001b[39m)]\n\u001b[0;32m     13\u001b[0m \u001b[38;5;66;03m# ds = load_dataset(\"json\", data_dir=\"./data/annotated\", download_mode=\"force_redownload\", cache_dir=None)\u001b[39;00m\n\u001b[1;32m---> 14\u001b[0m ds \u001b[38;5;241m=\u001b[39m \u001b[43mload_dataset\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43mjson\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mdata_files\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mshader20k_files\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mdownload_mode\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43mforce_redownload\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mcache_dir\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;28;43;01mNone\u001b[39;49;00m\u001b[43m)\u001b[49m\n\u001b[0;32m     15\u001b[0m \u001b[38;5;66;03m# ds = load_dataset(\"json\", data_dir=\"./data/annotated_test/\", download_mode=\"force_redownload\", cache_dir=None)\u001b[39;00m\n\u001b[0;32m     16\u001b[0m df \u001b[38;5;241m=\u001b[39m ds[\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mtrain\u001b[39m\u001b[38;5;124m\"\u001b[39m]\u001b[38;5;241m.\u001b[39mto_pandas()\n",
      "File \u001b[1;32mc:\\Users\\Jan\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\datasets\\load.py:2096\u001b[0m, in \u001b[0;36mload_dataset\u001b[1;34m(path, name, data_dir, data_files, split, cache_dir, features, download_config, download_mode, verification_mode, keep_in_memory, save_infos, revision, token, streaming, num_proc, storage_options, trust_remote_code, **config_kwargs)\u001b[0m\n\u001b[0;32m   2093\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m builder_instance\u001b[38;5;241m.\u001b[39mas_streaming_dataset(split\u001b[38;5;241m=\u001b[39msplit)\n\u001b[0;32m   2095\u001b[0m \u001b[38;5;66;03m# Download and prepare data\u001b[39;00m\n\u001b[1;32m-> 2096\u001b[0m \u001b[43mbuilder_instance\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mdownload_and_prepare\u001b[49m\u001b[43m(\u001b[49m\n\u001b[0;32m   2097\u001b[0m \u001b[43m    \u001b[49m\u001b[43mdownload_config\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mdownload_config\u001b[49m\u001b[43m,\u001b[49m\n\u001b[0;32m   2098\u001b[0m \u001b[43m    \u001b[49m\u001b[43mdownload_mode\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mdownload_mode\u001b[49m\u001b[43m,\u001b[49m\n\u001b[0;32m   2099\u001b[0m \u001b[43m    \u001b[49m\u001b[43mverification_mode\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mverification_mode\u001b[49m\u001b[43m,\u001b[49m\n\u001b[0;32m   2100\u001b[0m \u001b[43m    \u001b[49m\u001b[43mnum_proc\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mnum_proc\u001b[49m\u001b[43m,\u001b[49m\n\u001b[0;32m   2101\u001b[0m \u001b[43m    \u001b[49m\u001b[43mstorage_options\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mstorage_options\u001b[49m\u001b[43m,\u001b[49m\n\u001b[0;32m   2102\u001b[0m \u001b[43m\u001b[49m\u001b[43m)\u001b[49m\n\u001b[0;32m   2104\u001b[0m \u001b[38;5;66;03m# Build dataset for splits\u001b[39;00m\n\u001b[0;32m   2105\u001b[0m keep_in_memory \u001b[38;5;241m=\u001b[39m (\n\u001b[0;32m   2106\u001b[0m     keep_in_memory \u001b[38;5;28;01mif\u001b[39;00m keep_in_memory \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m \u001b[38;5;28;01melse\u001b[39;00m is_small_dataset(builder_instance\u001b[38;5;241m.\u001b[39minfo\u001b[38;5;241m.\u001b[39mdataset_size)\n\u001b[0;32m   2107\u001b[0m )\n",
      "File \u001b[1;32mc:\\Users\\Jan\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\datasets\\builder.py:924\u001b[0m, in \u001b[0;36mDatasetBuilder.download_and_prepare\u001b[1;34m(self, output_dir, download_config, download_mode, verification_mode, dl_manager, base_path, file_format, max_shard_size, num_proc, storage_options, **download_and_prepare_kwargs)\u001b[0m\n\u001b[0;32m    922\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m num_proc \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m:\n\u001b[0;32m    923\u001b[0m     prepare_split_kwargs[\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mnum_proc\u001b[39m\u001b[38;5;124m\"\u001b[39m] \u001b[38;5;241m=\u001b[39m num_proc\n\u001b[1;32m--> 924\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_download_and_prepare(\n\u001b[0;32m    925\u001b[0m     dl_manager\u001b[38;5;241m=\u001b[39mdl_manager,\n\u001b[0;32m    926\u001b[0m     verification_mode\u001b[38;5;241m=\u001b[39mverification_mode,\n\u001b[0;32m    927\u001b[0m     \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mprepare_split_kwargs,\n\u001b[0;32m    928\u001b[0m     \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mdownload_and_prepare_kwargs,\n\u001b[0;32m    929\u001b[0m )\n\u001b[0;32m    930\u001b[0m \u001b[38;5;66;03m# Sync info\u001b[39;00m\n\u001b[0;32m    931\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39minfo\u001b[38;5;241m.\u001b[39mdataset_size \u001b[38;5;241m=\u001b[39m \u001b[38;5;28msum\u001b[39m(split\u001b[38;5;241m.\u001b[39mnum_bytes \u001b[38;5;28;01mfor\u001b[39;00m split \u001b[38;5;129;01min\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39minfo\u001b[38;5;241m.\u001b[39msplits\u001b[38;5;241m.\u001b[39mvalues())\n",
      "File \u001b[1;32mc:\\Users\\Jan\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\datasets\\builder.py:999\u001b[0m, in \u001b[0;36mDatasetBuilder._download_and_prepare\u001b[1;34m(self, dl_manager, verification_mode, **prepare_split_kwargs)\u001b[0m\n\u001b[0;32m    995\u001b[0m split_dict\u001b[38;5;241m.\u001b[39madd(split_generator\u001b[38;5;241m.\u001b[39msplit_info)\n\u001b[0;32m    997\u001b[0m \u001b[38;5;28;01mtry\u001b[39;00m:\n\u001b[0;32m    998\u001b[0m     \u001b[38;5;66;03m# Prepare split will record examples associated to the split\u001b[39;00m\n\u001b[1;32m--> 999\u001b[0m     \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_prepare_split(split_generator, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mprepare_split_kwargs)\n\u001b[0;32m   1000\u001b[0m \u001b[38;5;28;01mexcept\u001b[39;00m \u001b[38;5;167;01mOSError\u001b[39;00m \u001b[38;5;28;01mas\u001b[39;00m e:\n\u001b[0;32m   1001\u001b[0m     \u001b[38;5;28;01mraise\u001b[39;00m \u001b[38;5;167;01mOSError\u001b[39;00m(\n\u001b[0;32m   1002\u001b[0m         \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mCannot find data file. \u001b[39m\u001b[38;5;124m\"\u001b[39m\n\u001b[0;32m   1003\u001b[0m         \u001b[38;5;241m+\u001b[39m (\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mmanual_download_instructions \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124m\"\u001b[39m)\n\u001b[0;32m   1004\u001b[0m         \u001b[38;5;241m+\u001b[39m \u001b[38;5;124m\"\u001b[39m\u001b[38;5;130;01m\\n\u001b[39;00m\u001b[38;5;124mOriginal error:\u001b[39m\u001b[38;5;130;01m\\n\u001b[39;00m\u001b[38;5;124m\"\u001b[39m\n\u001b[0;32m   1005\u001b[0m         \u001b[38;5;241m+\u001b[39m \u001b[38;5;28mstr\u001b[39m(e)\n\u001b[0;32m   1006\u001b[0m     ) \u001b[38;5;28;01mfrom\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m\n",
      "File \u001b[1;32mc:\\Users\\Jan\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\datasets\\builder.py:1740\u001b[0m, in \u001b[0;36mArrowBasedBuilder._prepare_split\u001b[1;34m(self, split_generator, file_format, num_proc, max_shard_size)\u001b[0m\n\u001b[0;32m   1738\u001b[0m job_id \u001b[38;5;241m=\u001b[39m \u001b[38;5;241m0\u001b[39m\n\u001b[0;32m   1739\u001b[0m \u001b[38;5;28;01mwith\u001b[39;00m pbar:\n\u001b[1;32m-> 1740\u001b[0m     \u001b[38;5;28;01mfor\u001b[39;00m job_id, done, content \u001b[38;5;129;01min\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_prepare_split_single(\n\u001b[0;32m   1741\u001b[0m         gen_kwargs\u001b[38;5;241m=\u001b[39mgen_kwargs, job_id\u001b[38;5;241m=\u001b[39mjob_id, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39m_prepare_split_args\n\u001b[0;32m   1742\u001b[0m     ):\n\u001b[0;32m   1743\u001b[0m         \u001b[38;5;28;01mif\u001b[39;00m done:\n\u001b[0;32m   1744\u001b[0m             result \u001b[38;5;241m=\u001b[39m content\n",
      "File \u001b[1;32mc:\\Users\\Jan\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\datasets\\builder.py:1871\u001b[0m, in \u001b[0;36mArrowBasedBuilder._prepare_split_single\u001b[1;34m(self, gen_kwargs, fpath, file_format, max_shard_size, job_id)\u001b[0m\n\u001b[0;32m   1869\u001b[0m     writer\u001b[38;5;241m.\u001b[39mwrite_table(table)\n\u001b[0;32m   1870\u001b[0m \u001b[38;5;28;01mexcept\u001b[39;00m CastError \u001b[38;5;28;01mas\u001b[39;00m cast_error:\n\u001b[1;32m-> 1871\u001b[0m     \u001b[38;5;28;01mraise\u001b[39;00m DatasetGenerationCastError\u001b[38;5;241m.\u001b[39mfrom_cast_error(\n\u001b[0;32m   1872\u001b[0m         cast_error\u001b[38;5;241m=\u001b[39mcast_error,\n\u001b[0;32m   1873\u001b[0m         builder_name\u001b[38;5;241m=\u001b[39m\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39minfo\u001b[38;5;241m.\u001b[39mbuilder_name,\n\u001b[0;32m   1874\u001b[0m         gen_kwargs\u001b[38;5;241m=\u001b[39mgen_kwargs,\n\u001b[0;32m   1875\u001b[0m         token\u001b[38;5;241m=\u001b[39m\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mtoken,\n\u001b[0;32m   1876\u001b[0m     )\n\u001b[0;32m   1877\u001b[0m num_examples_progress_update \u001b[38;5;241m+\u001b[39m\u001b[38;5;241m=\u001b[39m \u001b[38;5;28mlen\u001b[39m(table)\n\u001b[0;32m   1878\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m time\u001b[38;5;241m.\u001b[39mtime() \u001b[38;5;241m>\u001b[39m _time \u001b[38;5;241m+\u001b[39m config\u001b[38;5;241m.\u001b[39mPBAR_REFRESH_TIME_INTERVAL:\n",
      "\u001b[1;31mDatasetGenerationCastError\u001b[0m: An error occurred while generating the dataset\n\nAll the data files must have the same columns, but at some point there are 30 new columns ({'image_code', 'image_inputs', 'buffer_b_code', 'cube_a_inputs', 'author', 'license', 'date', 'buffer_d_code', 'buffer_d_inputs', 'sound_code', 'published', 'likes', 'buffer_a_code', 'buffer_a_inputs', 'buffer_b_inputs', 'tags', 'thumbnail', 'functions', 'sound_inputs', 'test', 'buffer_c_inputs', 'cube_a_code', 'description', 'common_code', 'buffer_c_code', 'id', 'name', 'viewed', 'time_retrieved', 'access'}) and 1 missing columns ({'text'}).\n\nThis happened while the json dataset builder was generating data using\n\nD:/Dokumente/Uni_OFFLINE/SoSe2024/shadertoys-dataset/./data/annotated/20k_2014-02.jsonl\n\nPlease either edit the data files to have matching columns, or separate them into different configurations (see docs at https://hf.co/docs/hub/datasets-manual-configuration#multiple-configurations)"
     ]
    }
   ],
   "source": [
    "from datasets import load_dataset, disable_caching\n",
    "import pandas as pd\n",
    "import os\n",
    "from tqdm.auto import tqdm\n",
    "\n",
    "# can infer the generic loader script this easily\n",
    "disable_caching()\n",
    "# cache location [click](%userprofile%/.cache/huggingface/datasets)\n",
    "# needs force_redownload to avoid loading incorrect cached stuff\n",
    "shader20k_files = os.listdir(\"./data/annotated\")\n",
    "# print(shader20k_files)\n",
    "shader20k_files = [f\"./data/annotated/{f}\" for f in shader20k_files if f.startswith(\"20k\")]\n",
    "# ds = load_dataset(\"json\", data_dir=\"./data/annotated\", download_mode=\"force_redownload\", cache_dir=None)\n",
    "ds = load_dataset(\"json\", data_files=shader20k_files, download_mode=\"force_redownload\", cache_dir=None)\n",
    "# ds = load_dataset(\"json\", data_dir=\"./data/annotated_test/\", download_mode=\"force_redownload\", cache_dir=None)\n",
    "df = ds[\"train\"].to_pandas()\n",
    "# read \"date\" column as unix timestamp\n",
    "df[\"date\"] = pd.to_datetime(df[\"date\"].astype(int), unit=\"s\")\n",
    "ds"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# fewer than 6k unique authors in shaders20k\n",
    "df[\"author\"].nunique()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# this is just from the 20k shaders dataset\n",
    "df[\"date\"].dt.year.value_counts().sort_index()#.plot(kind=\"bar\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "#redo the df\n",
    "\n",
    "df = ds[\"train\"].to_pandas()\n",
    "# read \"date\" column as unix timestamp\n",
    "df[\"date\"] = pd.to_datetime(df[\"date\"].astype(int), unit=\"s\")\n",
    "#"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "df[\"test\"].value_counts().head(10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# list of permissive licenses like: https://huggingface.co/datasets/bigcode/the-stack-v2/blob/main/license_stats.csv (gated?)\n",
    "# maybe this instead: https://huggingface.co/datasets/bigcode/the-stack/blob/main/licenses.json\n",
    "# seems to be the one they actually use: https://huggingface.co/datasets/bigcode-data/license_list/resolve/main/permissive_licenses.txt \n",
    "\n",
    "import requests\n",
    "permissive_list = requests.get(\"https://huggingface.co/datasets/bigcode-data/license_list/resolve/main/permissive_licenses.txt\").content.decode(\"utf-8\").split()\n",
    "permissive_list = [l.lower() for l in permissive_list]\n",
    "\n",
    "print(len(permissive_list))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# filtering down the dataset and keeping track of the numbers\n",
    "print(ds[\"train\"].unique(\"license\"))\n",
    "# filter permissive licenses\n",
    "filtered_ds = ds[\"train\"]\n",
    "num_all = len(filtered_ds)\n",
    "print(f\"{num_all=}\")\n",
    "filtered_ds = filtered_ds.filter(lambda x: x[\"published\"] == \"Public API\") # only take shaders published to the API\n",
    "num_public_api = len(filtered_ds)\n",
    "print(f\"{num_public_api=}\")\n",
    "filtered_ds = filtered_ds.filter(lambda x: x[\"license\"] in permissive_list + [\"CC-BY-NC-SA-3.0\"]) # permissive or base case(not actually permissive...) what about those AND and OR classifications?\n",
    "num_permissive = len(filtered_ds)\n",
    "print(f\"{num_permissive=}\")\n",
    "filtered_ds = filtered_ds.filter(lambda x: x[\"test\"] == \"ok\") # ok means it actually runs, valid was just that naga didn't error... but this also has issues\n",
    "num_working = len(filtered_ds)\n",
    "print(f\"{num_working=}\")\n",
    "filtered_ds"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# seems to never contain any common pass? or my preprocessing is wrong.\n",
    "print(df[\"common_code\"].unique()) \n",
    "\n",
    "# see how many shaders contain buffer passes (seemingly none?)\n",
    "ds[\"train\"].filter(lambda x: x[\"buffer_a_code\"] + x[\"buffer_b_code\"] + x[\"buffer_c_code\"] + x[\"buffer_d_code\"] != \"\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_per_month(column, stretched=False, data=None, **kwargs):\n",
    "    if data is None:\n",
    "        # lazy default case, to avoid name errror when not running cells above\n",
    "        data = df\n",
    "    feature_df = data[\"date\"].groupby([data[\"date\"].dt.to_period(\"M\"), data[column]]).count().unstack()\n",
    "    if stretched:\n",
    "        feature_df = feature_df.div(feature_df.sum(axis=1), axis=0)\n",
    "    return feature_df.plot(kind=\"bar\", stacked=True, figsize=(15,7), title=column, width=.9, **kwargs)\n",
    "\n",
    "def plot_per_year(column, stretched=False, data=None, **kwargs):\n",
    "    if data is None:\n",
    "        # lazy default case, to avoid name errror when not running cells above\n",
    "        data = df\n",
    "    feature_df = data[\"date\"].groupby([data[\"date\"].dt.to_period(\"Y\"), data[column]]).count().unstack()\n",
    "    if stretched:\n",
    "        feature_df = feature_df.div(feature_df.sum(axis=1), axis=0)\n",
    "    return feature_df.plot(kind=\"bar\", stacked=True, figsize=(15,7), title=column, width=.9, **kwargs)\n",
    "    \n",
    "# plot_per_month(\"license\", True, legend=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "plot_per_month(\"published\", False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# currently in development, another filtering step.\n",
    "plot_per_month(\"test\", stretched=True, legend=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "df[\"num_functions\"] = df[\"functions\"].apply(len)\n",
    "df[\"num_functions\"].describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "df[\"num_inputs\"] = df[\"image_inputs\"].apply(len)\n",
    "plot_per_month(\"num_inputs\", stretched=False, legend=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# these cases might be parsing failures? seem to always use macros inside functions or something.\n",
    "# some use no functions but Main(){...} which seems to work in glsl but not viable for our dataset\n",
    "from annotate import PARSER\n",
    "print(PARSER.parse(bytes(df[\"image_code\"][19616],encoding=\"utf-8\")).root_node.has_error)\n",
    "df.loc[df[\"num_functions\"] == 0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# TODO: refactor into a generator because I dislike the current shape. but figure out how batching and threadding works then?\n",
    "# part of the filter.py or prepare.py script that will be done soon.\n",
    "def split_functions(batch):\n",
    "    # per function fields\n",
    "    comments, headers, docstrings, bodies = [],[],[],[]\n",
    "    # per document fields\n",
    "    full_code, funcs, ids, dates = [],[],[],[]\n",
    "    for code, func_bytes, shader_id, shader_date in zip(batch[\"image_code\"], batch[\"functions\"], batch[\"id\"], batch[\"date\"]):\n",
    "        row_code = bytes(code,encoding=\"utf-8\")\n",
    "        for start_comment, start_header, end_header, end_docstring, end_function in func_bytes:\n",
    "            comments.append(row_code[start_comment:start_header].decode(encoding=\"utf-8\")),\n",
    "            headers.append(row_code[start_header:end_header].decode(encoding=\"utf-8\")),\n",
    "            docstrings.append(row_code[end_header:end_docstring].decode(encoding=\"utf-8\")),\n",
    "            bodies.append(row_code[end_docstring:end_function].decode(encoding=\"utf-8\")),\n",
    "            full_code.append(row_code.decode(encoding=\"utf-8\")),\n",
    "            funcs.append((start_comment, start_header, end_header, end_docstring, end_function))\n",
    "            ids.append(shader_id)\n",
    "            dates.append(shader_date)\n",
    "    return {\n",
    "            \"comment\": comments,\n",
    "            \"header\": headers,\n",
    "            \"docstring\": docstrings,\n",
    "            \"body\": bodies,\n",
    "            \"full_code\": full_code,\n",
    "            \"func_bytes\": funcs,\n",
    "            \"id\": ids,\n",
    "            \"date\": dates,\n",
    "            }\n",
    "\n",
    "\n",
    "# func_ds = ds.map(split_functions, batched=True, remove_columns=ds[\"train\"].column_names) # full dataset to functions\n",
    "func_ds = filtered_ds.map(split_functions, batched=True, remove_columns=filtered_ds.column_names) # only permissive shaders (the whole 20k set is single pass without common...)\n",
    "if \"train\" in func_ds:\n",
    "    func_ds = func_ds[\"train\"]\n",
    "func_ds"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "func_df = func_ds.to_pandas()\n",
    "func_df[\"date\"] = pd.to_datetime(func_ds[\"date\"], unit=\"s\") #didn't copy correctly?\n",
    "func_df.head() "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "func_df[\"has_comment\"] = func_df[\"comment\"] != \"\"\n",
    "func_df[\"has_docstring\"] = func_df[\"docstring\"] != \"\"\n",
    "# func_df[\"model_ctx\"] = func_df[\"comment\"] + func_df[\"header\"] + func_df[\"comment\"]\n",
    "# how many functions are there, and what is the share of comments, \n",
    "\n",
    "xtab = pd.crosstab(func_df[\"has_comment\"], func_df[\"has_docstring\"], margins=True)\n",
    "print(f\"functions with context = {xtab.iloc[-1,-1] - xtab.iloc[0,0]}\")\n",
    "xtab\n",
    "\n",
    "# for the permissive subset of shaders 21k the number goes down to like 253 for both... and 150 for valid shaders only."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# FINDING: only about half the comments are unique.\n",
    "# some frequent comment looks like a template maybe? should investigate\n",
    "print(func_df[\"comment\"].nunique())\n",
    "func_df[\"comment\"].value_counts(sort=True, ascending=False).head(10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# FINDING = only about a third of function-headers are unique?\n",
    "# it's expect to have one mainImage function per renderpass... but they are just whitesapce variable here - maybe we skip them in general\n",
    "print(func_df[\"header\"].nunique())\n",
    "func_df[\"header\"].value_counts(sort=True, ascending=False).head(10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# looks like some are not exact or might even use the common pass to bypass this\n",
    "func_df[\"header\"].str.startswith(\"void mainImage\").sum()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# half the docstrings aren't unique?\n",
    "# most common one is the shadertoy.com/new default for the main function\n",
    "print(func_df[\"docstring\"].nunique())\n",
    "func_df[\"docstring\"].value_counts(sort=True, ascending=False).head(10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# indicator where the docstring is just something commented out - not helpful instructions.\n",
    "func_df[\"docstring\"].str.endswith(\";\").sum()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# the comment is just a link to some website/reference? although sometimes it's a bunch of natural langauge too?\n",
    "print(func_df[\"comment\"].str.contains(\"https://\").sum())\n",
    "print(func_df[\"comment\"].str.contains(\"www.\").sum())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# figure out if a function is needed, by removing if from body and check if the code still runs (you should have filtered for only working shaders before)\n",
    "\n",
    "# step 1. subset the dataframe to only those rows that have comment and docstring\n",
    "both_df = func_df[func_df[\"has_comment\"] & func_df[\"has_docstring\"]]\n",
    "num_both= len(both_df)\n",
    "print(f\"{num_both=}\")\n",
    "# 253 functions form 168 shaders, only 179 of which are unique combinations...\n",
    "both_df[\"header\"].value_counts(sort=True, ascending=False).head(10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# this step can timeout/hang -.-\n",
    "# step 2. if code runs without the function.\n",
    "# takes ~1.4 minutes at home.\n",
    "from annotate import run_shader\n",
    "\n",
    "def check_function_needed(row):\n",
    "    code = row[\"full_code\"]\n",
    "    start_comment, start_header, end_header, end_docstring, end_function = row[\"func_bytes\"]\n",
    "    test_code = code[:start_comment] + code[end_function:] #TODO: this needs to be done in bytes -.-\n",
    "    status = run_shader(test_code)\n",
    "    return status == \"error\"\n",
    "\n",
    "\n",
    "both_df[\"needed\"] = both_df.apply(check_function_needed, axis=1)\n",
    "both_df[\"needed\"].value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# down select the only the needed and working examples.\n",
    "print(len(both_df))\n",
    "both_df = both_df.loc[both_df[\"needed\"]]\n",
    "num_needed = len(both_df)\n",
    "print(f\"{num_needed=}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from datasets import Dataset\n",
    "\n",
    "both_df.loc[:,\"input_both\"] = both_df[\"comment\"] + both_df[\"header\"] + both_df[\"docstring\"]\n",
    "both_df.loc[:,\"input_comment\"] = both_df[\"comment\"] + both_df[\"header\"]\n",
    "both_df.loc[:,\"input_docstring\"] = both_df[\"header\"] + both_df[\"docstring\"]\n",
    "both_df.loc[:,\"input_none\"] = both_df[\"header\"]\n",
    "print(both_df.shape)\n",
    "print(both_df[\"input_both\"].nunique())\n",
    "# remove the duplicates\n",
    "both_df.drop_duplicates(\"input_both\", inplace=True)\n",
    "experiment_ds = Dataset.from_pandas(both_df)\n",
    "experiment_ds"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# this pushes the dataset to the hub, uncomment with care!\n",
    "# experiment_ds.push_to_hub(\"Vipitis/Shadereval-experiments-dev\", private=False, commit_message=\"v0.2 filtered down to 257 examples, ensures working and Public+API\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# making a river plot (sankey diagram) of all out filtering steps\n",
    "# keeping track of number of functions and then in backets (number of shader programs)\n",
    "# filtering steps are in this order:\n",
    "# 1. on shader programs: published, license, test,\n",
    "# 2. on functions: comment+docstring, needed, deduplicate\n",
    "\n",
    "program_filter_steps = [num_all, num_public_api, num_permissive, num_working]\n",
    "import matplotlib.pyplot as plt\n",
    "from matplotlib.sankey import Sankey\n",
    "plt.figure(figsize=(20,10))\n",
    "plot = Sankey(head_angle=135, ax=plt.gca(), offset=-5000, unit=\" shaders\")\n",
    "# easily add a secondary data source?\n",
    "plot.add(flows=[-num_all], labels=[\"shaders20k\"], pathlengths=[9000])\n",
    "plot.add(flows=[num_all, -num_public_api, -(num_all-num_public_api)], patchlabel=\"published\", labels=[None,\"Published+API\", \"public/private\"], pathlengths=[5000, 1000, 0], orientations=[0,0,-1], prior=0, connect=(0,0), trunklength=5000)\n",
    "plot.add(flows=[num_public_api, -num_permissive, -(num_public_api-num_permissive)], labels=[None,\"permissive\", \"copyright, copyleft\"], pathlengths=[1000, 6000, (num_all-num_public_api)], orientations=[0,0,-1], prior=1, connect=(1,0), trunklength=5000)\n",
    "plot.add(flows=[num_permissive, -num_working, -(num_permissive-num_working)], labels=[None, \"working\", \"error\"], pathlengths=[2000, 8000, (num_all-num_permissive)], orientations=[0,0,-1], prior=2, connect=(1,0))\n",
    "\n",
    "# no we switch to function level\n",
    "plot.add(flows=[num_working, -len(func_df)],prior=3, connect=(1,0),orientations=[1], patchlabel=\"function parsing\", labels=[None, \"functions\"], pathlengths=[1000, 2000]) #need to change unit?\n",
    "# add function with comment and docstring = 329\n",
    "plot.add(flows=[len(func_df), -xtab.iloc[1,1], -xtab.iloc[0,0], -xtab.iloc[0,1], -xtab.iloc[1,0]],prior=4, connect=(1,0),orientations=[0,0,-1,-1,-1], labels=[None,\"both\", \"none\", \"docstring\",\"comment\"], pathlengths=[3000, 0, 0, 0, 0])\n",
    "# only needed functions = 318\n",
    "plot.add(flows=[xtab.iloc[1,1], -318, -11],prior=5, connect=(1,0),orientations=[0, 0, -1], labels=[None,\"needed\", \"unneeded\"], pathlengths=[5000, 5000, 10000])\n",
    "# deduplicated = 257\n",
    "plot.add(flows=[318, -len(both_df), -(318-len(both_df))],prior=6, connect=(1,0),orientations=[0, 0, -1], labels=[None, \"unique\", \"duplicates\"], pathlengths=[4000, 2000, 10000])\n",
    "\n",
    "# TODO: fancy this up - maybe figure out a scale\n",
    "# TODO: refactor this all into a function or generator to make the syntax less horrible.\n",
    "plot.finish()\n",
    "plt.savefig(\"../Thesis/report/texs/charts/data_filtering.png\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# plot unique headers per month\n",
    "# only looking at our \"permissive\" subset.\n",
    "# tag every function as unique until it shows up twice (we use the header, iterate by time)\n",
    "for i, row in tqdm(func_df.iterrows()):\n",
    "    if func_df.loc[:i,\"header\"].eq(row[\"header\"]).sum() > 1:\n",
    "        func_df.at[i, \"unique_header\"] = False\n",
    "    else:\n",
    "        func_df.at[i, \"unique_header\"] = True\n",
    "\n",
    "plot_per_month(\"unique_header\", stretched=False, data=func_df, legend=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Extension with API data\n",
    "\n",
    "### only downloaded years 2020 till 2023 for now. It takes around 1 hour per year....\n",
    "\n",
    "there is ~ 32k shaders on the API.\n",
    "we need to apply a few extra filters and then join them with the above..."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os \n",
    "import jsonlines\n",
    "import datasets\n",
    "import pandas as pd\n",
    "\n",
    "\n",
    "# awful code snippet pending https://github.com/huggingface/datasets/issues/7092 to use load_dataset instead (with both parts maybe).\n",
    "api_files = os.listdir(\"./data/annotated/api\")\n",
    "\n",
    "api_files = [f\"./data/annotated/api/{f}\" for f in api_files]\n",
    "\n",
    "api_file_contents = []\n",
    "for f in api_files:\n",
    "    with jsonlines.open(f) as reader:\n",
    "        for obj in reader:\n",
    "            api_file_contents.append(obj)\n",
    "\n",
    "ds_api = datasets.Dataset.from_list(api_file_contents)\n",
    "\n",
    "df_api = ds_api.to_pandas()\n",
    "\n",
    "df_api[\"date\"] = pd.to_datetime(df_api[\"date\"].astype(int), unit=\"s\")\n",
    "\n",
    "num_all = len(df_api)\n",
    "print(f\"{num_all=}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "# Filter 1: only shaders that are single pass (as in they don't have buffer passes etc)\n",
    "# TODO: sound could be allowed, since it doesn't impact the image.\n",
    "\n",
    "# TODO: maybe use common_code as that is trivially to combine?\n",
    "print(f'number of shaders with common code={df_api.loc[df_api[\"common_code\"] != \"\"].shape[0]}')\n",
    "other_passes = [col for col in df_api.columns if col.endswith(\"_code\") and col not in (\"image_code\")] #, \"common_code\", \"souncd_code\"\n",
    "\n",
    "df_api = df_api[(df_api.loc[:,other_passes] == \"\").all(axis=1)]\n",
    "num_single_pass = len(df_api)\n",
    "print(f\"{num_single_pass=}\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Filter 2: only shaders that don't require inputs (for the image_pass which is left now)\n",
    "# df_api[\"image_inputs\"].apply(len).value_counts() # -> there is quite something\n",
    "# TODO: maybe only some channel_types are bad... perhaps keep the ones we support (.complete)\n",
    "\n",
    "df_api = df_api[df_api[\"image_inputs\"].apply(len) == 0]\n",
    "num_no_inputs = len(df_api)\n",
    "print(f\"{num_no_inputs=}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Filter 3: overlap with the shaders20k data.\n",
    "# we keep the \"older\" variant from the shaders20k dataset because that is static and won't change.\n",
    "\n",
    "ids_20k = set(df[\"id\"])\n",
    "\n",
    "df_api = df_api[~df_api[\"id\"].isin(ids_20k)]\n",
    "num_new = len(df_api)\n",
    "print(f\"{num_new=}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### combine and use similar filters as above..."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "all combined is 27857 shaders\n",
      "8235\n",
      "8235 shaderprograms remaining after filter_public_api\n",
      "304 shaderprograms remaining after filter_licenses\n",
      "225 shaderprograms remaining after filter_single_pass\n",
      "167 shaderprograms remaining after filter_no_inputs\n",
      "163 shaderprograms remaining after filter_words\n",
      "137 shaderprograms remaining after filter_working\n",
      "19622\n",
      "7263 shaderprograms remaining after filter_public_api\n",
      "408 shaderprograms remaining after filter_licenses\n",
      "408 shaderprograms remaining after filter_single_pass\n",
      "345 shaderprograms remaining after filter_no_inputs\n",
      "330 shaderprograms remaining after filter_words\n",
      "276 shaderprograms remaining after filter_working\n",
      "413\n"
     ]
    }
   ],
   "source": [
    "# puttint it all together\n",
    "import os\n",
    "import jsonlines\n",
    "import pandas as pd\n",
    "import requests\n",
    "\n",
    "\n",
    "def load_data(data_dir: os.PathLike) -> pd.DataFrame:\n",
    "    lines = []\n",
    "    for file in os.listdir(data_dir):\n",
    "        if file.endswith(\".jsonl\"):\n",
    "            with jsonlines.open(os.path.join(data_dir, file)) as reader:\n",
    "                for obj in reader:\n",
    "                    lines.append(obj)\n",
    "\n",
    "    return pd.DataFrame(lines)\n",
    "\n",
    "\n",
    "def filter_public_api(dataframe: pd.DataFrame, **kwargs) -> pd.DataFrame:\n",
    "    \"\"\"\n",
    "    only keep shaders that are published to the API.\n",
    "    \"\"\"\n",
    "    # TODO: Publish API shouldn't be in raw or annotated, this needs to be fixed in both datahalves.\n",
    "    return dataframe[dataframe[\"published\"].isin([\"Public API\", 3])]\n",
    "\n",
    "\n",
    "def filter_licenses(dataframe: pd.DataFrame, keep_base=False, **kwargs) -> pd.DataFrame:\n",
    "    \"\"\"\n",
    "    only keep permissive licenses.\n",
    "    \"\"\"\n",
    "    permissive_list = requests.get(\"https://huggingface.co/datasets/bigcode-data/license_list/resolve/main/permissive_licenses.txt\").content.decode(\"utf-8\").split()\n",
    "    permissive_list = [l.lower() for l in permissive_list]\n",
    "    #TODO: figure out cases with AND and OR in the detection.\n",
    "    if keep_base:\n",
    "        permissive_list.append(\"CC-BY-NC-SA-3.0\")\n",
    "    return dataframe[dataframe[\"license\"].isin(permissive_list)]\n",
    "\n",
    "\n",
    "def filter_single_pass(dataframe: pd.DataFrame, **kwargs) -> pd.DataFrame:\n",
    "    \"\"\"\n",
    "    only keep shaders that are single pass.\n",
    "    \"\"\"\n",
    "    other_passes = [col for col in dataframe.columns if col.endswith(\"_code\") and col not in (\"image_code\")]\n",
    "    #TODO: consider keeping sound_code and common_code (the later one needs to be prepended to the image_code)\n",
    "    return dataframe[(dataframe.loc[:,other_passes] == \"\").all(axis=1)]\n",
    "\n",
    "\n",
    "def filter_no_inputs(dataframe: pd.DataFrame, **kwargs) -> pd.DataFrame:\n",
    "    \"\"\"\n",
    "    only keep shaders that don't require inputs.\n",
    "    to be used after you filtered for single pass shaders.\n",
    "    \"\"\"\n",
    "    # TODO: consider keeping some channel_types\n",
    "    return dataframe[dataframe[\"image_inputs\"].apply(len) == 0]\n",
    "\n",
    "#TODO: inspect if this is the case\n",
    "def filter_words(dataframe: pd.DataFrame, words=[\"test\", \"bug\"], **kwargs) -> pd.DataFrame:\n",
    "    \"\"\"\n",
    "    Drop all shaders that contain works like \"test\", \"debug\", \"ai\", \"chatGPT\", in the title, description or tags.\n",
    "    \"\"\"\n",
    "    for word in words:\n",
    "        dataframe = dataframe[~dataframe[\"name\"].str.contains(word, case=False)]\n",
    "        dataframe = dataframe[~dataframe[\"description\"].str.contains(word, case=False)]\n",
    "        # dataframe = dataframe[~dataframe[\"tags\"]\" \".join().str.contains(word, case=False)] # can't do string opeerations on list of tags.\n",
    "\n",
    "    return dataframe\n",
    "\n",
    "\n",
    "def filter_working(dataframe: pd.DataFrame, untested=False, **kwargs) -> pd.DataFrame:\n",
    "    \"\"\"\n",
    "    only keep shaders that are working.\n",
    "    Note: the idea of using `untested=True` is to first run all filters and just get a list of IDs, to then test these.\n",
    "    Testing is slow and should therefore be only annotated where it's needed.\n",
    "    \"\"\"\n",
    "\n",
    "    # TODO: testing needs to really work before we can rely on this\n",
    "    drop_values = [\"error\", \"panic\", \"timeout\", \"timedout\", \"valid\", \"untested\"]\n",
    "    keep_values = [\"ok\"]\n",
    "    if untested:\n",
    "        drop_values.remove(\"untested\")\n",
    "        keep_values.append(\"untested\")\n",
    "    return dataframe[dataframe[\"test\"].isin(keep_values)]\n",
    "\n",
    "PROGRAM_FILTERS = [filter_public_api, filter_licenses, filter_single_pass, filter_no_inputs, filter_words, filter_working]\n",
    "\n",
    "def filter_programs(dataframe: pd.DataFrame, filters=PROGRAM_FILTERS, **kwargs) -> pd.DataFrame:\n",
    "    \"\"\"\n",
    "    apply a series of filters and print the resulting numbers\n",
    "    kwargs are passed to the filters\n",
    "    untested: keep untested shaders (default: False)\n",
    "    keep_base: keep the base license (default: False)\n",
    "    \"\"\"\n",
    "    print(len(dataframe))\n",
    "\n",
    "    # TODO: add plotting into sankey?  when plot=True\n",
    "    # plot = kwargs.pop(\"plot\", None)\n",
    "    # if plot is None:\n",
    "        # import matplotlib.pyplot as plt\n",
    "        # from matplotlib.sankey import Sankey\n",
    "        # plt.figure(figsize=(20,10))\n",
    "        # plot = Sankey(head_angle=135, ax=plt.gca(), offset=-5000, unit=\" shader programs\")\n",
    "        # # easily add a secondary data source?\n",
    "        # plot.add(flows=[-len(dataframe)], labels=[\"datasource\"], pathlengths=[9000])\n",
    "\n",
    "\n",
    "    for f in filters:\n",
    "        dataframe = f(dataframe, **kwargs)\n",
    "        print(f\"{len(dataframe)} shaderprograms remaining after {f.__name__}\")\n",
    "    return dataframe\n",
    "\n",
    "\n",
    "def combine_datasets(base_data, add_data):\n",
    "    \"\"\"\n",
    "    combine two datasets, keeping the base data.\n",
    "    \"\"\"\n",
    "    ids_base = set(base_data[\"id\"])\n",
    "    return pd.concat([base_data, add_data[~add_data[\"id\"].isin(ids_base)]])\n",
    "\n",
    "\n",
    "api_df = load_data(\"./data/annotated/api\")\n",
    "s20k_df = load_data(\"./data/annotated\")\n",
    "all_combined_df = combine_datasets(s20k_df, api_df) # make this early\n",
    "print(f\"all combined is {len(all_combined_df)} shaders\")\n",
    "api_df = filter_programs(api_df, untested=False, keep_base=False)\n",
    "s20k_df = filter_programs(s20k_df, untested=False, keep_base=False)\n",
    "\n",
    "combined_df = combine_datasets(s20k_df, api_df)\n",
    "combined_df[\"date\"] = pd.to_datetime(combined_df[\"date\"].astype(int), unit=\"s\")\n",
    "print(len(combined_df))\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 87,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\Jan\\AppData\\Local\\Temp\\ipykernel_5920\\1739740627.py:60: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
      "  dataframe[\"model_inp\"] = dataframe[\"comment\"] + dataframe[\"header\"]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "all functions from all uniuqe image passes: 202752\n",
      "2991\n",
      "693 functions remaining after filter_has_context\n",
      "693 functions remaining after construct_inp\n",
      "678 functions remaining after filter_length\n",
      "662 functions remaining after filter_alphabetic\n",
      "510 functions remaining after filter_duplicates\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 510/510 [23:48<00:00,  2.80s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "467 functions remaining after filter_needed\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "# expand to functions and then do function filters\n",
    " \n",
    "from annotate import run_shader\n",
    "from tqdm import tqdm\n",
    "\n",
    "tqdm.pandas()\n",
    "\n",
    "def expand_functions(dataframe: pd.DataFrame) -> pd.DataFrame:\n",
    "    \n",
    "    # TODO: likely redundant, since we keep everything... might need some cleanup instead\n",
    "    cols_to_keep = [\"id\", \"date\", \"image_code\", \"functions\", \"func_bytes\", \"author\", \"license\"]\n",
    "    #function byte indicies: start_comment, start_header, end_header, end_docstring, end_function\n",
    "    func_parts = [\"comment\", \"header\", \"docstring\", \"body\"]\n",
    "\n",
    "    # TODO: do we need the whole functions for anything downstream?\n",
    "    dataframe[\"functions_\"] = dataframe[\"functions\"] # make a copy to keep it past explode\n",
    "    func_df = dataframe.explode(\"functions\")\n",
    "    func_df.rename(columns={\"functions\": \"func_bytes\", \"functions_\": \"functions\"}, inplace=True)\n",
    "    func_df.dropna(subset=[\"func_bytes\"], inplace=True) # some shaders got not functions parsed ? -> TODO: check the tree-sitter.has_error() result?\n",
    "    func_df.reset_index(inplace=True)\n",
    "    for row_idx, row in func_df.iterrows():\n",
    "        # print(row_idx, row)\n",
    "        # break\n",
    "        code = row[\"image_code\"]\n",
    "        code_bytes = bytes(code, encoding=\"utf-8\")\n",
    "        # func_bytes = row[\"func_bytes\"]\n",
    "        for idx, part in enumerate(func_parts):\n",
    "            start = row[\"func_bytes\"][idx]\n",
    "            end = row[\"func_bytes\"][idx+1]\n",
    "            func_df.at[row_idx, part] = code_bytes[start:end].decode(encoding=\"utf-8\")\n",
    "        \n",
    "    func_df.drop(columns=func_df.columns.difference(cols_to_keep + func_parts), inplace=True)\n",
    "    # func_df[\"date\"] = pd.to_datetime(func_df[\"date\"].astype(int), unit=\"s\")\n",
    "\n",
    "    return func_df\n",
    "\n",
    "def filter_has_context(dataframe: pd.DataFrame, context=\"comment\", **kwargs) -> pd.DataFrame:\n",
    "    \"\"\"\n",
    "    only keep functions that have a specific context. (not exclusive)\n",
    "    context: one of \"comment\", \"docstring\", \"both\", \"none\"\n",
    "    \"\"\"\n",
    "    # TODO: not all are implemented\n",
    "    if context == \"comment\":\n",
    "        return dataframe[dataframe[\"comment\"] != \"\"]\n",
    "    elif context == \"docstring\":\n",
    "        return dataframe[dataframe[\"docstring\"] != \"\"]\n",
    "    elif context == \"both\":\n",
    "        return dataframe[(dataframe[\"comment\"] != \"\") & (dataframe[\"docstring\"] != \"\")]\n",
    "    elif context == \"none\":\n",
    "        raise NotImplementedError(f\"not implemented yet for context: {context}\")\n",
    "    else:\n",
    "        raise ValueError(f\"unknown context: {context}\")\n",
    "\n",
    "# TODO: combine construct_inp and filter_has_context into one function maybe?\n",
    "def construct_inp(dataframe: pd.DataFrame, context=\"comment\", **kwargs) -> pd.DataFrame:\n",
    "    \"\"\"\n",
    "    adds the model_inp column based on the context.\n",
    "    \"\"\"\n",
    "    if context == \"comment\":\n",
    "        dataframe[\"model_inp\"] = dataframe[\"comment\"] + dataframe[\"header\"]\n",
    "    elif context == \"docstring\":\n",
    "        dataframe[\"model_inp\"] = dataframe[\"header\"] + dataframe[\"docstring\"]\n",
    "    elif context == \"both\":\n",
    "        dataframe[\"model_inp\"] = dataframe[\"comment\"] + dataframe[\"header\"] + dataframe[\"docstring\"]\n",
    "    elif context == \"none\":\n",
    "        dataframe[\"model_inp\"] = dataframe[\"header\"]\n",
    "    else:\n",
    "        raise ValueError(f\"unknown context: {context}\")\n",
    "    return dataframe\n",
    "\n",
    "def filter_length(dataframe: pd.DataFrame, max_length=2500, **kwargs) -> pd.DataFrame:\n",
    "    \"\"\"\n",
    "    sort out function bodies that are really long. (likely machine generated)\n",
    "    \"\"\"\n",
    "    # TODO: why chose this number?\n",
    "    return dataframe[dataframe[\"body\"].apply(len) <= max_length]\n",
    "\n",
    "def filter_alphabetic(dataframe: pd.DataFrame, column=\"comment\", cutoff=0.25, **kwargs) -> pd.DataFrame:\n",
    "    \"\"\"\n",
    "    sort out functions that have an alphabetic ration above the cutoff.\n",
    "    \"\"\"\n",
    "    return dataframe[dataframe[column].apply(lambda x: sum(c.isalpha() for c in x) / len(x)) > cutoff]\n",
    "\n",
    "def filter_duplicates(dataframe: pd.DataFrame, sort_by=\"date\", **kwargs) -> pd.DataFrame:\n",
    "    \"\"\"\n",
    "    only keeps unique model_inp values. \n",
    "    sort_by specifies a column to sort by and keep the top value of. (default is \"date\")\n",
    "    the sorting is not kept\n",
    "    \"\"\"\n",
    "    if sort_by not in dataframe.columns:\n",
    "        raise ValueError(f\"unknown column to sort by: {sort_by}\")\n",
    "    # asceding=True and keep=\"first\" are the defualts, meaning we keep ealier entries! \n",
    "    # if sorting my views or likes it might might sense to either set ascending=False or keep=\"last\" (not both).\n",
    "    dataframe.sort_values(by=sort_by, inplace=True, ascending=True)\n",
    "    out_df = dataframe.drop_duplicates(\"model_inp\", keep=\"first\") #default is \"keep first\"\n",
    "    # undo the sort\n",
    "    return out_df.sort_index()\n",
    "\n",
    "def filter_needed(dataframe: pd.DataFrame, **kwargs) -> pd.DataFrame:\n",
    "    \"\"\"\n",
    "    only keep functions that are needed. By running the shader with the function removed and seeing if it errors...\n",
    "    \"\"\"\n",
    "    def row_needed(row):\n",
    "        # print(row[\"id\"])\n",
    "        code_bytes = bytes(row[\"image_code\"], encoding=\"utf-8\")\n",
    "        start_comment, start_header, end_header, end_docstring, end_function = row[\"func_bytes\"]\n",
    "        outer_bytes = code_bytes[:start_comment] + b\"\\n\" +  code_bytes[end_function:]\n",
    "        test_code = outer_bytes.decode(encoding=\"utf-8\")\n",
    "        status = run_shader(test_code)\n",
    "        return status != \"ok\"\n",
    "    dataframe[\"needed\"] = dataframe.progress_apply(row_needed, axis=1)\n",
    "    return dataframe[dataframe[\"needed\"]]\n",
    "\n",
    "FUNCTION_FILTERS = [filter_has_context, construct_inp, filter_length, filter_alphabetic, filter_duplicates, filter_needed]\n",
    "def filter_functions(dataframe: pd.DataFrame, filters=FUNCTION_FILTERS, **kwargs) -> pd.DataFrame:\n",
    "    \"\"\"\n",
    "    apply a series of filters and print the resulting numbers\n",
    "    kwargs are passed to the filters\n",
    "    \"\"\"\n",
    "    print(len(dataframe))\n",
    "\n",
    "    for f in filters:\n",
    "        dataframe = f(dataframe, **kwargs)\n",
    "        print(f\"{len(dataframe)} functions remaining after {f.__name__}\")\n",
    "    return dataframe\n",
    "\n",
    "func_df = expand_functions(combined_df)\n",
    "all_funcs = expand_functions(all_combined_df)\n",
    "print(f\"all functions from all uniuqe image passes: {len(all_funcs)}\")\n",
    "func_df = filter_functions(func_df, context=\"comment\")\n",
    "\n",
    "# func_df = expand_functions(combined_df)\n",
    "# print(len(func_df))\n",
    "# func_df = filter_has_context(func_df, context=\"comment\")\n",
    "# print(len(func_df))\n",
    "# func_df = construct_inp(func_df, context=\"comment\")\n",
    "# print(len(func_df)) # shouldn't change\n",
    "# func_df = filter_length(func_df)\n",
    "# print(len(func_df))\n",
    "# func_df = filter_duplicates(func_df)\n",
    "# print(len(func_df))\n",
    "# func_df = filter_needed(func_df) #this step takes nearly 3 minutes on my laptop! do we know if it ever hangs?\n",
    "# print(len(func_df))\n",
    "# func_df[\"body\"].apply(len).describe()\n",
    "# func_df[\"id\"].value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Axes: xlabel='date'>"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "func_df[\"date\"] = pd.to_datetime(func_df[\"date\"], unit=\"s\")\n",
    "func_df[\"date\"].dt.year.value_counts().sort_index().plot(kind=\"bar\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "2473    NstGDM\n",
       "231     4d33z4\n",
       "232     4d33z4\n",
       "239     4d33z4\n",
       "240     4d33z4\n",
       "242     4d33z4\n",
       "243     4d33z4\n",
       "246     4d33z4\n",
       "247     4d33z4\n",
       "283     Xs3GRM\n",
       "Name: id, dtype: object"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# inspection for length\n",
    "# show the ids for the longest image_codes\n",
    "func_df.sort_values(\"image_code\", key=lambda x: x.str.len(), ascending=False).head(10)[\"id\"]\n",
    "# Stack v2 filters: 100k rows, https://github.com/go-enry/go-enry (is_generated), less than 25% of alphabetic characters, encoded, \n",
    "# https://www.shadertoy.com/view/NstGDM (contains a neural network in the code, don't want that)\n",
    "\n",
    "\n",
    "# func_df[\"comment\"].apply(len).describe(percentiles=[.5, .75, .9, .95, .99])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 86,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "date\n",
       "2014     2\n",
       "2015     6\n",
       "2016     4\n",
       "2017     6\n",
       "2018     4\n",
       "2019    16\n",
       "2020    16\n",
       "2021    16\n",
       "2022    21\n",
       "2023    25\n",
       "Name: count, dtype: int64"
      ]
     },
     "execution_count": 86,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "combined_df.sort_values(by=\"date\", ascending=True).drop_duplicates(\"author\")[\"date\"].dt.year.value_counts().sort_index()#.plot.bar(alpha=0.3, color=\"red\")\n",
    "# combined_df.sort_values(by=\"date\", ascending=False).drop_duplicates(\"author\")[\"date\"].dt.year.value_counts().sort_index()#.plot.bar(alpha=0.3, color=\"blue\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "ename": "AttributeError",
     "evalue": "Can only use .dt accessor with datetimelike values",
     "output_type": "error",
     "traceback": [
      "\u001b[1;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[1;31mAttributeError\u001b[0m                            Traceback (most recent call last)",
      "Cell \u001b[1;32mIn[10], line 3\u001b[0m\n\u001b[0;32m      1\u001b[0m \u001b[38;5;66;03m# Shaders per year, in the combined verison... will get published to API instead.\u001b[39;00m\n\u001b[0;32m      2\u001b[0m \u001b[38;5;66;03m# all_combined_df[\"date\"] = pd.to_datetime(all_combined_df[\"date\"].astype(int), unit=\"s\") # this should only be run once.\u001b[39;00m\n\u001b[1;32m----> 3\u001b[0m \u001b[43mall_combined_df\u001b[49m\u001b[43m[\u001b[49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43mdate\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[43m]\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mdt\u001b[49m\u001b[38;5;241m.\u001b[39myear\u001b[38;5;241m.\u001b[39mvalue_counts()\u001b[38;5;241m.\u001b[39msort_index()\u001b[38;5;66;03m#.plot(kind=\"bar\", title=\"shaders per year\")\u001b[39;00m\n",
      "File \u001b[1;32mc:\\Users\\Jan\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\pandas\\core\\generic.py:6299\u001b[0m, in \u001b[0;36mNDFrame.__getattr__\u001b[1;34m(self, name)\u001b[0m\n\u001b[0;32m   6292\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m (\n\u001b[0;32m   6293\u001b[0m     name \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;129;01min\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_internal_names_set\n\u001b[0;32m   6294\u001b[0m     \u001b[38;5;129;01mand\u001b[39;00m name \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;129;01min\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_metadata\n\u001b[0;32m   6295\u001b[0m     \u001b[38;5;129;01mand\u001b[39;00m name \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;129;01min\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_accessors\n\u001b[0;32m   6296\u001b[0m     \u001b[38;5;129;01mand\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_info_axis\u001b[38;5;241m.\u001b[39m_can_hold_identifiers_and_holds_name(name)\n\u001b[0;32m   6297\u001b[0m ):\n\u001b[0;32m   6298\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28mself\u001b[39m[name]\n\u001b[1;32m-> 6299\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28;43mobject\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[38;5;21;43m__getattribute__\u001b[39;49m\u001b[43m(\u001b[49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mname\u001b[49m\u001b[43m)\u001b[49m\n",
      "File \u001b[1;32mc:\\Users\\Jan\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\pandas\\core\\accessor.py:224\u001b[0m, in \u001b[0;36mCachedAccessor.__get__\u001b[1;34m(self, obj, cls)\u001b[0m\n\u001b[0;32m    221\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m obj \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m:\n\u001b[0;32m    222\u001b[0m     \u001b[38;5;66;03m# we're accessing the attribute of the class, i.e., Dataset.geo\u001b[39;00m\n\u001b[0;32m    223\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_accessor\n\u001b[1;32m--> 224\u001b[0m accessor_obj \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_accessor\u001b[49m\u001b[43m(\u001b[49m\u001b[43mobj\u001b[49m\u001b[43m)\u001b[49m\n\u001b[0;32m    225\u001b[0m \u001b[38;5;66;03m# Replace the property with the accessor object. Inspired by:\u001b[39;00m\n\u001b[0;32m    226\u001b[0m \u001b[38;5;66;03m# https://www.pydanny.com/cached-property.html\u001b[39;00m\n\u001b[0;32m    227\u001b[0m \u001b[38;5;66;03m# We need to use object.__setattr__ because we overwrite __setattr__ on\u001b[39;00m\n\u001b[0;32m    228\u001b[0m \u001b[38;5;66;03m# NDFrame\u001b[39;00m\n\u001b[0;32m    229\u001b[0m \u001b[38;5;28mobject\u001b[39m\u001b[38;5;241m.\u001b[39m\u001b[38;5;21m__setattr__\u001b[39m(obj, \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_name, accessor_obj)\n",
      "File \u001b[1;32mc:\\Users\\Jan\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\pandas\\core\\indexes\\accessors.py:643\u001b[0m, in \u001b[0;36mCombinedDatetimelikeProperties.__new__\u001b[1;34m(cls, data)\u001b[0m\n\u001b[0;32m    640\u001b[0m \u001b[38;5;28;01melif\u001b[39;00m \u001b[38;5;28misinstance\u001b[39m(data\u001b[38;5;241m.\u001b[39mdtype, PeriodDtype):\n\u001b[0;32m    641\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m PeriodProperties(data, orig)\n\u001b[1;32m--> 643\u001b[0m \u001b[38;5;28;01mraise\u001b[39;00m \u001b[38;5;167;01mAttributeError\u001b[39;00m(\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mCan only use .dt accessor with datetimelike values\u001b[39m\u001b[38;5;124m\"\u001b[39m)\n",
      "\u001b[1;31mAttributeError\u001b[0m: Can only use .dt accessor with datetimelike values"
     ]
    }
   ],
   "source": [
    "# Shaders per year, in the combined verison... will get published to API instead.\n",
    "# all_combined_df[\"date\"] = pd.to_datetime(all_combined_df[\"date\"].astype(int), unit=\"s\") # this should only be run once.\n",
    "all_combined_df[\"date\"].dt.year.value_counts().sort_index()#.plot(kind=\"bar\", title=\"shaders per year\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "all_combined_df[\"license\"].value_counts()/len(all_combined_df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>id</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>2229</th>\n",
       "      <td>NstGDM</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "          id\n",
       "2229  NstGDM"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# check for the percentage of alphabetic characters in the whole code.\n",
    "api_df[api_df[\"image_code\"].apply(lambda x: sum(c.isalpha() for c in x) / len(x)) < .25][[\"id\"]]#, \"comment\", \"header\"]]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# extract the shader_ids that are left in the end to run tests just on them.\n",
    "# with open(\"data/ids/v0-3.txt\", \"w\", encoding=\"utf-8\") as f:\n",
    "    # f.write(\"\\n\".join(func_df[\"id\"].unique().tolist()))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 89,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "204\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "header\n",
       "float              236\n",
       "vec3               109\n",
       "vec2                57\n",
       "vec4                37\n",
       "void                 7\n",
       "mat3                 5\n",
       "bool                 4\n",
       "bound3               3\n",
       "v33                  3\n",
       "tilingVal3D          2\n",
       "cast_ray_result      1\n",
       "hexParams            1\n",
       "ivec2                1\n",
       "ivec3                1\n",
       "Name: count, dtype: int64"
      ]
     },
     "execution_count": 89,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "print(func_df[\"id\"].nunique()) # our 719 functions are just from 290 programs?\n",
    "func_df[\"header\"].str.split(\" \").apply(lambda x: x[0]).value_counts()\n",
    "# func_df[\"header\"].str.startswith(\"void\").sum()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 90,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>id</th>\n",
       "      <th>author</th>\n",
       "      <th>date</th>\n",
       "      <th>image_code</th>\n",
       "      <th>license</th>\n",
       "      <th>func_bytes</th>\n",
       "      <th>functions</th>\n",
       "      <th>comment</th>\n",
       "      <th>header</th>\n",
       "      <th>docstring</th>\n",
       "      <th>body</th>\n",
       "      <th>model_inp</th>\n",
       "      <th>needed</th>\n",
       "      <th>function</th>\n",
       "      <th>function_frequency</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>15</th>\n",
       "      <td>lsfXWH</td>\n",
       "      <td>iq</td>\n",
       "      <td>2014-05-15 00:14:43</td>\n",
       "      <td>// The MIT License\\n// Copyright © 2014 Inigo ...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[3256, 3289, 3316, 3316, 3347]</td>\n",
       "      <td>[[1978, 2039, 2083, 2083, 3254], [3256, 3289, ...</td>\n",
       "      <td>// unrolled version of the above\\n</td>\n",
       "      <td>float SH_0_0( in vec3 s ) {</td>\n",
       "      <td></td>\n",
       "      <td>vec3 n = s.zxy; return  k01; }</td>\n",
       "      <td>// unrolled version of the above\\nfloat SH_0_0...</td>\n",
       "      <td>True</td>\n",
       "      <td>float SH_0_0( in vec3 s ) { vec3 n = s.zxy; re...</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>35</th>\n",
       "      <td>4ssSRl</td>\n",
       "      <td>iq</td>\n",
       "      <td>2014-07-18 08:21:44</td>\n",
       "      <td>// The MIT License\\n// Copyright © 2014 Inigo ...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[0, 1196, 1243, 1243, 1363]</td>\n",
       "      <td>[[0, 1196, 1243, 1243, 1363], [1365, 1365, 142...</td>\n",
       "      <td>// The MIT License\\n// Copyright © 2014 Inigo ...</td>\n",
       "      <td>float line( in vec2 a, in vec2 b, in vec2 p )\\n{</td>\n",
       "      <td></td>\n",
       "      <td>\\n\\tvec2 pa = p - a;\\n\\tvec2 ba = b - a;\\n\\tfl...</td>\n",
       "      <td>// The MIT License\\n// Copyright © 2014 Inigo ...</td>\n",
       "      <td>True</td>\n",
       "      <td>float line( in vec2 a, in vec2 b, in vec2 p )\\...</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>37</th>\n",
       "      <td>4djXzz</td>\n",
       "      <td>otaviogood</td>\n",
       "      <td>2014-08-21 06:53:07</td>\n",
       "      <td>/*--------------------------------------------...</td>\n",
       "      <td>cc0-1.0</td>\n",
       "      <td>[562, 589, 612, 612, 689]</td>\n",
       "      <td>[[562, 589, 612, 612, 689], [690, 690, 713, 71...</td>\n",
       "      <td>// various noise functions\\n</td>\n",
       "      <td>float Hash2d(vec2 uv)\\n{</td>\n",
       "      <td></td>\n",
       "      <td>\\n    float f = uv.x + uv.y * 47.0;\\n    retur...</td>\n",
       "      <td>// various noise functions\\nfloat Hash2d(vec2 ...</td>\n",
       "      <td>True</td>\n",
       "      <td>float Hash2d(vec2 uv)\\n{\\n    float f = uv.x +...</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>48</th>\n",
       "      <td>4dBXz3</td>\n",
       "      <td>iq</td>\n",
       "      <td>2014-10-24 08:55:07</td>\n",
       "      <td>// The MIT License\\n// Copyright © 2014 Inigo ...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[1272, 1348, 1388, 1388, 1443]</td>\n",
       "      <td>[[1272, 1348, 1388, 1388, 1443], [1445, 1524, ...</td>\n",
       "      <td>// Flip v if in the negative half plane define...</td>\n",
       "      <td>vec2 flipIfNeg( in vec2 v, in vec2 r )\\n{</td>\n",
       "      <td></td>\n",
       "      <td>\\n    float k = dot(v,r);\\n    return (k&gt;0.0) ...</td>\n",
       "      <td>// Flip v if in the negative half plane define...</td>\n",
       "      <td>True</td>\n",
       "      <td>vec2 flipIfNeg( in vec2 v, in vec2 r )\\n{\\n   ...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>49</th>\n",
       "      <td>4dBXz3</td>\n",
       "      <td>iq</td>\n",
       "      <td>2014-10-24 08:55:07</td>\n",
       "      <td>// The MIT License\\n// Copyright © 2014 Inigo ...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[1445, 1524, 1564, 1564, 1626]</td>\n",
       "      <td>[[1272, 1348, 1388, 1388, 1443], [1445, 1524, ...</td>\n",
       "      <td>// Reflect v if in the negative half plane def...</td>\n",
       "      <td>vec2 reflIfNeg( in vec2 v, in vec2 r )\\n{</td>\n",
       "      <td></td>\n",
       "      <td>\\n    float k = dot(v,r);\\n    return (k&gt;0.0) ...</td>\n",
       "      <td>// Reflect v if in the negative half plane def...</td>\n",
       "      <td>True</td>\n",
       "      <td>vec2 reflIfNeg( in vec2 v, in vec2 r )\\n{\\n   ...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2981</th>\n",
       "      <td>clKfWm</td>\n",
       "      <td>afl_ext</td>\n",
       "      <td>2023-12-09 21:45:07</td>\n",
       "      <td>// afl_ext 2017-2023\\n// MIT License\\n\\n// Use...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[4801, 4870, 4894, 4894, 5105]</td>\n",
       "      <td>[[993, 1070, 1130, 1130, 1523], [1525, 1592, 1...</td>\n",
       "      <td>// Pretty self explanatory FBM with some preci...</td>\n",
       "      <td>float cloudsFBM(vec3 p){</td>\n",
       "      <td></td>\n",
       "      <td>\\n    float a = 0.0;\\n    float w = 0.5;\\n    ...</td>\n",
       "      <td>// Pretty self explanatory FBM with some preci...</td>\n",
       "      <td>True</td>\n",
       "      <td>float cloudsFBM(vec3 p){\\n    float a = 0.0;\\n...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2983</th>\n",
       "      <td>clKfWm</td>\n",
       "      <td>afl_ext</td>\n",
       "      <td>2023-12-09 21:45:07</td>\n",
       "      <td>// afl_ext 2017-2023\\n// MIT License\\n\\n// Use...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[5221, 5393, 5424, 5424, 6194]</td>\n",
       "      <td>[[993, 1070, 1130, 1130, 1523], [1525, 1592, 1...</td>\n",
       "      <td>// this function probes the clouds densite at ...</td>\n",
       "      <td>vec2 cloudsDensity3D(vec3 pos){</td>\n",
       "      <td></td>\n",
       "      <td>\\n    float h = getHeightOverSurface(pos);\\n  ...</td>\n",
       "      <td>// this function probes the clouds densite at ...</td>\n",
       "      <td>True</td>\n",
       "      <td>vec2 cloudsDensity3D(vec3 pos){\\n    float h =...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2985</th>\n",
       "      <td>clKfWm</td>\n",
       "      <td>afl_ext</td>\n",
       "      <td>2023-12-09 21:45:07</td>\n",
       "      <td>// afl_ext 2017-2023\\n// MIT License\\n\\n// Use...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[7849, 8201, 8257, 8257, 8618]</td>\n",
       "      <td>[[993, 1070, 1130, 1130, 1523], [1525, 1592, 1...</td>\n",
       "      <td>// very native rendering for the ground, shado...</td>\n",
       "      <td>vec3 renderGround(vec3 point, float dist, floa...</td>\n",
       "      <td></td>\n",
       "      <td>\\n    float shadow = raymarchClouds(\\n        ...</td>\n",
       "      <td>// very native rendering for the ground, shado...</td>\n",
       "      <td>True</td>\n",
       "      <td>vec3 renderGround(vec3 point, float dist, floa...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2986</th>\n",
       "      <td>clKfWm</td>\n",
       "      <td>afl_ext</td>\n",
       "      <td>2023-12-09 21:45:07</td>\n",
       "      <td>// afl_ext 2017-2023\\n// MIT License\\n\\n// Use...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[8625, 8713, 8806, 8806, 9094]</td>\n",
       "      <td>[[993, 1070, 1130, 1130, 1523], [1525, 1592, 1...</td>\n",
       "      <td>// Straightforward, render raymarch, apply fog...</td>\n",
       "      <td>vec3 renderClouds(vec3 pointStart, vec3 pointE...</td>\n",
       "      <td></td>\n",
       "      <td>\\n    vec4 clouds = raymarchClouds(\\n         ...</td>\n",
       "      <td>// Straightforward, render raymarch, apply fog...</td>\n",
       "      <td>True</td>\n",
       "      <td>vec3 renderClouds(vec3 pointStart, vec3 pointE...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2987</th>\n",
       "      <td>clKfWm</td>\n",
       "      <td>afl_ext</td>\n",
       "      <td>2023-12-09 21:45:07</td>\n",
       "      <td>// afl_ext 2017-2023\\n// MIT License\\n\\n// Use...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[9097, 9187, 9218, 9218, 9649]</td>\n",
       "      <td>[[993, 1070, 1130, 1130, 1523], [1525, 1592, 1...</td>\n",
       "      <td>// Great tonemapping function from my other sh...</td>\n",
       "      <td>vec3 aces_tonemap(vec3 color) {</td>\n",
       "      <td></td>\n",
       "      <td>\\n  mat3 m1 = mat3(\\n    0.59719, 0.07600, 0...</td>\n",
       "      <td>// Great tonemapping function from my other sh...</td>\n",
       "      <td>True</td>\n",
       "      <td>vec3 aces_tonemap(vec3 color) {  \\n  mat3 m1 =...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>467 rows × 15 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "          id      author                date  \\\n",
       "15    lsfXWH          iq 2014-05-15 00:14:43   \n",
       "35    4ssSRl          iq 2014-07-18 08:21:44   \n",
       "37    4djXzz  otaviogood 2014-08-21 06:53:07   \n",
       "48    4dBXz3          iq 2014-10-24 08:55:07   \n",
       "49    4dBXz3          iq 2014-10-24 08:55:07   \n",
       "...      ...         ...                 ...   \n",
       "2981  clKfWm     afl_ext 2023-12-09 21:45:07   \n",
       "2983  clKfWm     afl_ext 2023-12-09 21:45:07   \n",
       "2985  clKfWm     afl_ext 2023-12-09 21:45:07   \n",
       "2986  clKfWm     afl_ext 2023-12-09 21:45:07   \n",
       "2987  clKfWm     afl_ext 2023-12-09 21:45:07   \n",
       "\n",
       "                                             image_code  license  \\\n",
       "15    // The MIT License\\n// Copyright © 2014 Inigo ...      mit   \n",
       "35    // The MIT License\\n// Copyright © 2014 Inigo ...      mit   \n",
       "37    /*--------------------------------------------...  cc0-1.0   \n",
       "48    // The MIT License\\n// Copyright © 2014 Inigo ...      mit   \n",
       "49    // The MIT License\\n// Copyright © 2014 Inigo ...      mit   \n",
       "...                                                 ...      ...   \n",
       "2981  // afl_ext 2017-2023\\n// MIT License\\n\\n// Use...      mit   \n",
       "2983  // afl_ext 2017-2023\\n// MIT License\\n\\n// Use...      mit   \n",
       "2985  // afl_ext 2017-2023\\n// MIT License\\n\\n// Use...      mit   \n",
       "2986  // afl_ext 2017-2023\\n// MIT License\\n\\n// Use...      mit   \n",
       "2987  // afl_ext 2017-2023\\n// MIT License\\n\\n// Use...      mit   \n",
       "\n",
       "                          func_bytes  \\\n",
       "15    [3256, 3289, 3316, 3316, 3347]   \n",
       "35       [0, 1196, 1243, 1243, 1363]   \n",
       "37         [562, 589, 612, 612, 689]   \n",
       "48    [1272, 1348, 1388, 1388, 1443]   \n",
       "49    [1445, 1524, 1564, 1564, 1626]   \n",
       "...                              ...   \n",
       "2981  [4801, 4870, 4894, 4894, 5105]   \n",
       "2983  [5221, 5393, 5424, 5424, 6194]   \n",
       "2985  [7849, 8201, 8257, 8257, 8618]   \n",
       "2986  [8625, 8713, 8806, 8806, 9094]   \n",
       "2987  [9097, 9187, 9218, 9218, 9649]   \n",
       "\n",
       "                                              functions  \\\n",
       "15    [[1978, 2039, 2083, 2083, 3254], [3256, 3289, ...   \n",
       "35    [[0, 1196, 1243, 1243, 1363], [1365, 1365, 142...   \n",
       "37    [[562, 589, 612, 612, 689], [690, 690, 713, 71...   \n",
       "48    [[1272, 1348, 1388, 1388, 1443], [1445, 1524, ...   \n",
       "49    [[1272, 1348, 1388, 1388, 1443], [1445, 1524, ...   \n",
       "...                                                 ...   \n",
       "2981  [[993, 1070, 1130, 1130, 1523], [1525, 1592, 1...   \n",
       "2983  [[993, 1070, 1130, 1130, 1523], [1525, 1592, 1...   \n",
       "2985  [[993, 1070, 1130, 1130, 1523], [1525, 1592, 1...   \n",
       "2986  [[993, 1070, 1130, 1130, 1523], [1525, 1592, 1...   \n",
       "2987  [[993, 1070, 1130, 1130, 1523], [1525, 1592, 1...   \n",
       "\n",
       "                                                comment  \\\n",
       "15                   // unrolled version of the above\\n   \n",
       "35    // The MIT License\\n// Copyright © 2014 Inigo ...   \n",
       "37                         // various noise functions\\n   \n",
       "48    // Flip v if in the negative half plane define...   \n",
       "49    // Reflect v if in the negative half plane def...   \n",
       "...                                                 ...   \n",
       "2981  // Pretty self explanatory FBM with some preci...   \n",
       "2983  // this function probes the clouds densite at ...   \n",
       "2985  // very native rendering for the ground, shado...   \n",
       "2986  // Straightforward, render raymarch, apply fog...   \n",
       "2987  // Great tonemapping function from my other sh...   \n",
       "\n",
       "                                                 header docstring  \\\n",
       "15                          float SH_0_0( in vec3 s ) {             \n",
       "35     float line( in vec2 a, in vec2 b, in vec2 p )\\n{             \n",
       "37                             float Hash2d(vec2 uv)\\n{             \n",
       "48            vec2 flipIfNeg( in vec2 v, in vec2 r )\\n{             \n",
       "49            vec2 reflIfNeg( in vec2 v, in vec2 r )\\n{             \n",
       "...                                                 ...       ...   \n",
       "2981                           float cloudsFBM(vec3 p){             \n",
       "2983                    vec2 cloudsDensity3D(vec3 pos){             \n",
       "2985  vec3 renderGround(vec3 point, float dist, floa...             \n",
       "2986  vec3 renderClouds(vec3 pointStart, vec3 pointE...             \n",
       "2987                    vec3 aces_tonemap(vec3 color) {             \n",
       "\n",
       "                                                   body  \\\n",
       "15                       vec3 n = s.zxy; return  k01; }   \n",
       "35    \\n\\tvec2 pa = p - a;\\n\\tvec2 ba = b - a;\\n\\tfl...   \n",
       "37    \\n    float f = uv.x + uv.y * 47.0;\\n    retur...   \n",
       "48    \\n    float k = dot(v,r);\\n    return (k>0.0) ...   \n",
       "49    \\n    float k = dot(v,r);\\n    return (k>0.0) ...   \n",
       "...                                                 ...   \n",
       "2981  \\n    float a = 0.0;\\n    float w = 0.5;\\n    ...   \n",
       "2983  \\n    float h = getHeightOverSurface(pos);\\n  ...   \n",
       "2985  \\n    float shadow = raymarchClouds(\\n        ...   \n",
       "2986  \\n    vec4 clouds = raymarchClouds(\\n         ...   \n",
       "2987    \\n  mat3 m1 = mat3(\\n    0.59719, 0.07600, 0...   \n",
       "\n",
       "                                              model_inp  needed  \\\n",
       "15    // unrolled version of the above\\nfloat SH_0_0...    True   \n",
       "35    // The MIT License\\n// Copyright © 2014 Inigo ...    True   \n",
       "37    // various noise functions\\nfloat Hash2d(vec2 ...    True   \n",
       "48    // Flip v if in the negative half plane define...    True   \n",
       "49    // Reflect v if in the negative half plane def...    True   \n",
       "...                                                 ...     ...   \n",
       "2981  // Pretty self explanatory FBM with some preci...    True   \n",
       "2983  // this function probes the clouds densite at ...    True   \n",
       "2985  // very native rendering for the ground, shado...    True   \n",
       "2986  // Straightforward, render raymarch, apply fog...    True   \n",
       "2987  // Great tonemapping function from my other sh...    True   \n",
       "\n",
       "                                               function  function_frequency  \n",
       "15    float SH_0_0( in vec3 s ) { vec3 n = s.zxy; re...                   3  \n",
       "35    float line( in vec2 a, in vec2 b, in vec2 p )\\...                   3  \n",
       "37    float Hash2d(vec2 uv)\\n{\\n    float f = uv.x +...                   4  \n",
       "48    vec2 flipIfNeg( in vec2 v, in vec2 r )\\n{\\n   ...                   1  \n",
       "49    vec2 reflIfNeg( in vec2 v, in vec2 r )\\n{\\n   ...                   1  \n",
       "...                                                 ...                 ...  \n",
       "2981  float cloudsFBM(vec3 p){\\n    float a = 0.0;\\n...                   1  \n",
       "2983  vec2 cloudsDensity3D(vec3 pos){\\n    float h =...                   1  \n",
       "2985  vec3 renderGround(vec3 point, float dist, floa...                   1  \n",
       "2986  vec3 renderClouds(vec3 pointStart, vec3 pointE...                   1  \n",
       "2987  vec3 aces_tonemap(vec3 color) {  \\n  mat3 m1 =...                   1  \n",
       "\n",
       "[467 rows x 15 columns]"
      ]
     },
     "execution_count": 90,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# some temporary columns\n",
    "all_funcs[\"function\"] = all_funcs[\"header\"] + all_funcs[\"body\"]\n",
    "func_df[\"function\"] = func_df[\"header\"] + func_df[\"body\"]\n",
    "func_df[\"function_frequency\"] = all_funcs[\"function\"].value_counts()[func_df[\"function\"]].values\n",
    "func_df[\"header_frequency\"] = all_funcs[\"header\"].value_counts()[func_df[\"header\"]].values\n",
    "func_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 92,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>id</th>\n",
       "      <th>author</th>\n",
       "      <th>date</th>\n",
       "      <th>image_code</th>\n",
       "      <th>license</th>\n",
       "      <th>func_bytes</th>\n",
       "      <th>functions</th>\n",
       "      <th>comment</th>\n",
       "      <th>header</th>\n",
       "      <th>body</th>\n",
       "      <th>model_inp</th>\n",
       "      <th>function_frequency</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>15</th>\n",
       "      <td>lsfXWH</td>\n",
       "      <td>iq</td>\n",
       "      <td>2014-05-15 00:14:43</td>\n",
       "      <td>// The MIT License\\n// Copyright © 2014 Inigo ...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[3256, 3289, 3316, 3316, 3347]</td>\n",
       "      <td>[[1978, 2039, 2083, 2083, 3254], [3256, 3289, ...</td>\n",
       "      <td>// unrolled version of the above\\n</td>\n",
       "      <td>float SH_0_0( in vec3 s ) {</td>\n",
       "      <td>vec3 n = s.zxy; return  k01; }</td>\n",
       "      <td>// unrolled version of the above\\nfloat SH_0_0...</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>35</th>\n",
       "      <td>4ssSRl</td>\n",
       "      <td>iq</td>\n",
       "      <td>2014-07-18 08:21:44</td>\n",
       "      <td>// The MIT License\\n// Copyright © 2014 Inigo ...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[0, 1196, 1243, 1243, 1363]</td>\n",
       "      <td>[[0, 1196, 1243, 1243, 1363], [1365, 1365, 142...</td>\n",
       "      <td>// The MIT License\\n// Copyright © 2014 Inigo ...</td>\n",
       "      <td>float line( in vec2 a, in vec2 b, in vec2 p )\\n{</td>\n",
       "      <td>\\n\\tvec2 pa = p - a;\\n\\tvec2 ba = b - a;\\n\\tfl...</td>\n",
       "      <td>// The MIT License\\n// Copyright © 2014 Inigo ...</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>37</th>\n",
       "      <td>4djXzz</td>\n",
       "      <td>otaviogood</td>\n",
       "      <td>2014-08-21 06:53:07</td>\n",
       "      <td>/*--------------------------------------------...</td>\n",
       "      <td>cc0-1.0</td>\n",
       "      <td>[562, 589, 612, 612, 689]</td>\n",
       "      <td>[[562, 589, 612, 612, 689], [690, 690, 713, 71...</td>\n",
       "      <td>// various noise functions\\n</td>\n",
       "      <td>float Hash2d(vec2 uv)\\n{</td>\n",
       "      <td>\\n    float f = uv.x + uv.y * 47.0;\\n    retur...</td>\n",
       "      <td>// various noise functions\\nfloat Hash2d(vec2 ...</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>48</th>\n",
       "      <td>4dBXz3</td>\n",
       "      <td>iq</td>\n",
       "      <td>2014-10-24 08:55:07</td>\n",
       "      <td>// The MIT License\\n// Copyright © 2014 Inigo ...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[1272, 1348, 1388, 1388, 1443]</td>\n",
       "      <td>[[1272, 1348, 1388, 1388, 1443], [1445, 1524, ...</td>\n",
       "      <td>// Flip v if in the negative half plane define...</td>\n",
       "      <td>vec2 flipIfNeg( in vec2 v, in vec2 r )\\n{</td>\n",
       "      <td>\\n    float k = dot(v,r);\\n    return (k&gt;0.0) ...</td>\n",
       "      <td>// Flip v if in the negative half plane define...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>49</th>\n",
       "      <td>4dBXz3</td>\n",
       "      <td>iq</td>\n",
       "      <td>2014-10-24 08:55:07</td>\n",
       "      <td>// The MIT License\\n// Copyright © 2014 Inigo ...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[1445, 1524, 1564, 1564, 1626]</td>\n",
       "      <td>[[1272, 1348, 1388, 1388, 1443], [1445, 1524, ...</td>\n",
       "      <td>// Reflect v if in the negative half plane def...</td>\n",
       "      <td>vec2 reflIfNeg( in vec2 v, in vec2 r )\\n{</td>\n",
       "      <td>\\n    float k = dot(v,r);\\n    return (k&gt;0.0) ...</td>\n",
       "      <td>// Reflect v if in the negative half plane def...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2981</th>\n",
       "      <td>clKfWm</td>\n",
       "      <td>afl_ext</td>\n",
       "      <td>2023-12-09 21:45:07</td>\n",
       "      <td>// afl_ext 2017-2023\\n// MIT License\\n\\n// Use...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[4801, 4870, 4894, 4894, 5105]</td>\n",
       "      <td>[[993, 1070, 1130, 1130, 1523], [1525, 1592, 1...</td>\n",
       "      <td>// Pretty self explanatory FBM with some preci...</td>\n",
       "      <td>float cloudsFBM(vec3 p){</td>\n",
       "      <td>\\n    float a = 0.0;\\n    float w = 0.5;\\n    ...</td>\n",
       "      <td>// Pretty self explanatory FBM with some preci...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2983</th>\n",
       "      <td>clKfWm</td>\n",
       "      <td>afl_ext</td>\n",
       "      <td>2023-12-09 21:45:07</td>\n",
       "      <td>// afl_ext 2017-2023\\n// MIT License\\n\\n// Use...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[5221, 5393, 5424, 5424, 6194]</td>\n",
       "      <td>[[993, 1070, 1130, 1130, 1523], [1525, 1592, 1...</td>\n",
       "      <td>// this function probes the clouds densite at ...</td>\n",
       "      <td>vec2 cloudsDensity3D(vec3 pos){</td>\n",
       "      <td>\\n    float h = getHeightOverSurface(pos);\\n  ...</td>\n",
       "      <td>// this function probes the clouds densite at ...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2985</th>\n",
       "      <td>clKfWm</td>\n",
       "      <td>afl_ext</td>\n",
       "      <td>2023-12-09 21:45:07</td>\n",
       "      <td>// afl_ext 2017-2023\\n// MIT License\\n\\n// Use...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[7849, 8201, 8257, 8257, 8618]</td>\n",
       "      <td>[[993, 1070, 1130, 1130, 1523], [1525, 1592, 1...</td>\n",
       "      <td>// very native rendering for the ground, shado...</td>\n",
       "      <td>vec3 renderGround(vec3 point, float dist, floa...</td>\n",
       "      <td>\\n    float shadow = raymarchClouds(\\n        ...</td>\n",
       "      <td>// very native rendering for the ground, shado...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2986</th>\n",
       "      <td>clKfWm</td>\n",
       "      <td>afl_ext</td>\n",
       "      <td>2023-12-09 21:45:07</td>\n",
       "      <td>// afl_ext 2017-2023\\n// MIT License\\n\\n// Use...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[8625, 8713, 8806, 8806, 9094]</td>\n",
       "      <td>[[993, 1070, 1130, 1130, 1523], [1525, 1592, 1...</td>\n",
       "      <td>// Straightforward, render raymarch, apply fog...</td>\n",
       "      <td>vec3 renderClouds(vec3 pointStart, vec3 pointE...</td>\n",
       "      <td>\\n    vec4 clouds = raymarchClouds(\\n         ...</td>\n",
       "      <td>// Straightforward, render raymarch, apply fog...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2987</th>\n",
       "      <td>clKfWm</td>\n",
       "      <td>afl_ext</td>\n",
       "      <td>2023-12-09 21:45:07</td>\n",
       "      <td>// afl_ext 2017-2023\\n// MIT License\\n\\n// Use...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[9097, 9187, 9218, 9218, 9649]</td>\n",
       "      <td>[[993, 1070, 1130, 1130, 1523], [1525, 1592, 1...</td>\n",
       "      <td>// Great tonemapping function from my other sh...</td>\n",
       "      <td>vec3 aces_tonemap(vec3 color) {</td>\n",
       "      <td>\\n  mat3 m1 = mat3(\\n    0.59719, 0.07600, 0...</td>\n",
       "      <td>// Great tonemapping function from my other sh...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>467 rows × 12 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "          id      author                date  \\\n",
       "15    lsfXWH          iq 2014-05-15 00:14:43   \n",
       "35    4ssSRl          iq 2014-07-18 08:21:44   \n",
       "37    4djXzz  otaviogood 2014-08-21 06:53:07   \n",
       "48    4dBXz3          iq 2014-10-24 08:55:07   \n",
       "49    4dBXz3          iq 2014-10-24 08:55:07   \n",
       "...      ...         ...                 ...   \n",
       "2981  clKfWm     afl_ext 2023-12-09 21:45:07   \n",
       "2983  clKfWm     afl_ext 2023-12-09 21:45:07   \n",
       "2985  clKfWm     afl_ext 2023-12-09 21:45:07   \n",
       "2986  clKfWm     afl_ext 2023-12-09 21:45:07   \n",
       "2987  clKfWm     afl_ext 2023-12-09 21:45:07   \n",
       "\n",
       "                                             image_code  license  \\\n",
       "15    // The MIT License\\n// Copyright © 2014 Inigo ...      mit   \n",
       "35    // The MIT License\\n// Copyright © 2014 Inigo ...      mit   \n",
       "37    /*--------------------------------------------...  cc0-1.0   \n",
       "48    // The MIT License\\n// Copyright © 2014 Inigo ...      mit   \n",
       "49    // The MIT License\\n// Copyright © 2014 Inigo ...      mit   \n",
       "...                                                 ...      ...   \n",
       "2981  // afl_ext 2017-2023\\n// MIT License\\n\\n// Use...      mit   \n",
       "2983  // afl_ext 2017-2023\\n// MIT License\\n\\n// Use...      mit   \n",
       "2985  // afl_ext 2017-2023\\n// MIT License\\n\\n// Use...      mit   \n",
       "2986  // afl_ext 2017-2023\\n// MIT License\\n\\n// Use...      mit   \n",
       "2987  // afl_ext 2017-2023\\n// MIT License\\n\\n// Use...      mit   \n",
       "\n",
       "                          func_bytes  \\\n",
       "15    [3256, 3289, 3316, 3316, 3347]   \n",
       "35       [0, 1196, 1243, 1243, 1363]   \n",
       "37         [562, 589, 612, 612, 689]   \n",
       "48    [1272, 1348, 1388, 1388, 1443]   \n",
       "49    [1445, 1524, 1564, 1564, 1626]   \n",
       "...                              ...   \n",
       "2981  [4801, 4870, 4894, 4894, 5105]   \n",
       "2983  [5221, 5393, 5424, 5424, 6194]   \n",
       "2985  [7849, 8201, 8257, 8257, 8618]   \n",
       "2986  [8625, 8713, 8806, 8806, 9094]   \n",
       "2987  [9097, 9187, 9218, 9218, 9649]   \n",
       "\n",
       "                                              functions  \\\n",
       "15    [[1978, 2039, 2083, 2083, 3254], [3256, 3289, ...   \n",
       "35    [[0, 1196, 1243, 1243, 1363], [1365, 1365, 142...   \n",
       "37    [[562, 589, 612, 612, 689], [690, 690, 713, 71...   \n",
       "48    [[1272, 1348, 1388, 1388, 1443], [1445, 1524, ...   \n",
       "49    [[1272, 1348, 1388, 1388, 1443], [1445, 1524, ...   \n",
       "...                                                 ...   \n",
       "2981  [[993, 1070, 1130, 1130, 1523], [1525, 1592, 1...   \n",
       "2983  [[993, 1070, 1130, 1130, 1523], [1525, 1592, 1...   \n",
       "2985  [[993, 1070, 1130, 1130, 1523], [1525, 1592, 1...   \n",
       "2986  [[993, 1070, 1130, 1130, 1523], [1525, 1592, 1...   \n",
       "2987  [[993, 1070, 1130, 1130, 1523], [1525, 1592, 1...   \n",
       "\n",
       "                                                comment  \\\n",
       "15                   // unrolled version of the above\\n   \n",
       "35    // The MIT License\\n// Copyright © 2014 Inigo ...   \n",
       "37                         // various noise functions\\n   \n",
       "48    // Flip v if in the negative half plane define...   \n",
       "49    // Reflect v if in the negative half plane def...   \n",
       "...                                                 ...   \n",
       "2981  // Pretty self explanatory FBM with some preci...   \n",
       "2983  // this function probes the clouds densite at ...   \n",
       "2985  // very native rendering for the ground, shado...   \n",
       "2986  // Straightforward, render raymarch, apply fog...   \n",
       "2987  // Great tonemapping function from my other sh...   \n",
       "\n",
       "                                                 header  \\\n",
       "15                          float SH_0_0( in vec3 s ) {   \n",
       "35     float line( in vec2 a, in vec2 b, in vec2 p )\\n{   \n",
       "37                             float Hash2d(vec2 uv)\\n{   \n",
       "48            vec2 flipIfNeg( in vec2 v, in vec2 r )\\n{   \n",
       "49            vec2 reflIfNeg( in vec2 v, in vec2 r )\\n{   \n",
       "...                                                 ...   \n",
       "2981                           float cloudsFBM(vec3 p){   \n",
       "2983                    vec2 cloudsDensity3D(vec3 pos){   \n",
       "2985  vec3 renderGround(vec3 point, float dist, floa...   \n",
       "2986  vec3 renderClouds(vec3 pointStart, vec3 pointE...   \n",
       "2987                    vec3 aces_tonemap(vec3 color) {   \n",
       "\n",
       "                                                   body  \\\n",
       "15                       vec3 n = s.zxy; return  k01; }   \n",
       "35    \\n\\tvec2 pa = p - a;\\n\\tvec2 ba = b - a;\\n\\tfl...   \n",
       "37    \\n    float f = uv.x + uv.y * 47.0;\\n    retur...   \n",
       "48    \\n    float k = dot(v,r);\\n    return (k>0.0) ...   \n",
       "49    \\n    float k = dot(v,r);\\n    return (k>0.0) ...   \n",
       "...                                                 ...   \n",
       "2981  \\n    float a = 0.0;\\n    float w = 0.5;\\n    ...   \n",
       "2983  \\n    float h = getHeightOverSurface(pos);\\n  ...   \n",
       "2985  \\n    float shadow = raymarchClouds(\\n        ...   \n",
       "2986  \\n    vec4 clouds = raymarchClouds(\\n         ...   \n",
       "2987    \\n  mat3 m1 = mat3(\\n    0.59719, 0.07600, 0...   \n",
       "\n",
       "                                              model_inp  function_frequency  \n",
       "15    // unrolled version of the above\\nfloat SH_0_0...                   3  \n",
       "35    // The MIT License\\n// Copyright © 2014 Inigo ...                   3  \n",
       "37    // various noise functions\\nfloat Hash2d(vec2 ...                   4  \n",
       "48    // Flip v if in the negative half plane define...                   1  \n",
       "49    // Reflect v if in the negative half plane def...                   1  \n",
       "...                                                 ...                 ...  \n",
       "2981  // Pretty self explanatory FBM with some preci...                   1  \n",
       "2983  // this function probes the clouds densite at ...                   1  \n",
       "2985  // very native rendering for the ground, shado...                   1  \n",
       "2986  // Straightforward, render raymarch, apply fog...                   1  \n",
       "2987  // Great tonemapping function from my other sh...                   1  \n",
       "\n",
       "[467 rows x 12 columns]"
      ]
     },
     "execution_count": 92,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# drop some columns we don't need in the dataset\n",
    "clean_func_df = func_df.drop(columns=[\"function\", \"docstring\", \"needed\"])\n",
    "clean_func_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# maybe something to show which kind of shaders are easy/simple?\n",
    "# the common tags in our dataset are 2d (73 out of 394), sdf (61), 3d (47), distance (39) etc?\n",
    "# according to the full dataset, our filters have blindspots in \"fractal\", \"raymarching\", \"noise\", \"procedural\"\n",
    "from collections import Counter\n",
    "print(\"all:\", end=\"\")\n",
    "all_tags = []\n",
    "for t in all_combined_df[\"tags\"]:\n",
    "    all_tags.extend(t)\n",
    "print(len(all_tags))\n",
    "tag_counts = Counter(all_tags)\n",
    "print(tag_counts.most_common(10))\n",
    "print(\"eval:\", end=\"\")\n",
    "eval_ids = func_df[\"id\"].tolist()\n",
    "all_tags = []\n",
    "for t in all_combined_df[all_combined_df[\"id\"].isin(eval_ids)][\"tags\"]:\n",
    "    all_tags.extend(t)\n",
    "print(len(all_tags))\n",
    "tag_counts = Counter(all_tags)\n",
    "print(tag_counts.most_common(10))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "4bcb95f42e044930aae674766eb4557d",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Uploading the dataset shards:   0%|          | 0/1 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "34ee8b1325dd4065b3bd8678cd73ce80",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Creating parquet from Arrow format:   0%|          | 0/1 [00:00<?, ?ba/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "11d6658098e04ffba95b9210d8b2c152",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "README.md:   0%|          | 0.00/1.58k [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "CommitInfo(commit_url='https://huggingface.co/datasets/Vipitis/Shadereval-experiments-dev/commit/6bb952fd56153f6c57c19ea0e2c4acd46421b370', commit_message='initial v0.4 with 467', commit_description='', oid='6bb952fd56153f6c57c19ea0e2c4acd46421b370', pr_url=None, pr_revision=None, pr_num=None)"
      ]
     },
     "execution_count": 95,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# publishing the v0.3 version with context=\"comment\" and all the filters.\n",
    "from datasets import Dataset\n",
    "func_ds = Dataset.from_pandas(clean_func_df)\n",
    "func_ds = func_ds.remove_columns(['__index_level_0__'])\n",
    "func_ds\n",
    "# func_ds.push_to_hub(\"Vipitis/Shadereval-experiments-dev\", commit_message=\"initial v0.4 with 467\", private=False, split=\"test\") # don't easily uncomment again..."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 94,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "55\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "license\n",
       "mit           299\n",
       "cc0-1.0       145\n",
       "cc-by-4.0      15\n",
       "libpng          5\n",
       "cc-by-3.0       1\n",
       "isc             1\n",
       "apache-2.0      1\n",
       "Name: count, dtype: int64"
      ]
     },
     "execution_count": 94,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "print(func_df[\"author\"].nunique())\n",
    "func_df[\"license\"].value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "plot_per_month(\"license\", stretched=False, data=combined_df, legend=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 88,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Axes: title={'center': 'id'}, xlabel='date'>"
      ]
     },
     "execution_count": 88,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1500x700 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# this is skewed due to duplicates -.-\n",
    "plot_per_year(\"id\", stretched=False, data=func_df, legend=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Axes: title={'center': 'id'}, xlabel='date'>"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1500x700 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# this is recent perfer\n",
    "plot_per_year(\"id\", stretched=False, data=func_df, legend=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Dataset({\n",
       "    features: ['id', 'author', 'date', 'image_code', 'license', 'func_bytes', 'functions', 'comment', 'header', 'body', 'model_inp', 'function_frequency'],\n",
       "    num_rows: 467\n",
       "})"
      ]
     },
     "execution_count": 1,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from datasets import load_dataset, disable_caching\n",
    "# disable_caching()\n",
    "ref_ds = load_dataset(\"Vipitis/Shadereval-experiments-dev\", revision=\"main\", split=\"test\")# , download_mode=\"force_redownload\") #might be revision=\"0.3\" eventually.\n",
    "ref_ds"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# some functions appear multiple times but at most thirce, and a few twice...\n",
    "# comments appear up to 15 times which is odd, but all this SDF references I think.\n",
    "import pandas as pd\n",
    "ref_df = ref_ds.to_pandas()\n",
    "ref_df[\"header\"].value_counts().head(20)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "for row in ref_ds:\n",
    "    print(row)\n",
    "    break"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# all_funcs[\"id\"].nunique() # 27428 shaders/programs\n",
    "# all_funcs[\"header\"].nunique() # 78256 unique headers\n",
    "# all_funcs[\"body\"].nunique() # 130948 unique bodies\n",
    "all_funcs[\"function\"] = all_funcs[\"header\"] + all_funcs[\"body\"]\n",
    "all_funcs[\"function\"].nunique() # 137308 unique functions\n",
    "\n",
    "all_funcs[\"function\"].value_counts().head(20)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# function/header frequency in the eval set.\n",
    "all_funcs[\"header\"].value_counts()[func_df[\"header\"]].sort_values(ascending=False).plot(ylim=(0,200), xticks=[]) \n",
    "all_funcs[\"body\"].value_counts()[func_df[\"body\"]].sort_values(ascending=False).plot(xticks=[])\n",
    "all_funcs[\"function\"].value_counts()[func_df[\"function\"]].sort_values(ascending=False).plot(xticks=[])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# unique timestamps per year\n",
    "\n",
    "df_dates = pd.DataFrame(func_df[\"date\"].unique(), columns=[\"date\"])\n",
    "df_dates[\"date\"].dt.year.value_counts().sort_index()#.plot(kind=\"bar\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "# collect all panics\n",
    "import tempfile\n",
    "import subprocess\n",
    "import os\n",
    "from tqdm.auto import tqdm\n",
    "\n",
    "_file_template = \"\"\"\n",
    "from wgpu_shadertoy import Shadertoy\n",
    "\n",
    "shader_code = '''{}'''\n",
    "\n",
    "shader = Shadertoy(shader_code, shader_type=\"glsl\", offscreen=True)\n",
    "\n",
    "if __name__ == \"__main__\":\n",
    "    shader.show()\n",
    "    shader.snapshot(123.45)\n",
    "\"\"\"\n",
    "\n",
    "def _run_shader_in_subprocess(shader_code:str, timeout:float=10) -> str:\n",
    "    \"\"\"\n",
    "    writes the shadercode into a temporary file, and tries to run the shader with a snapshot. This will catch any kind of errors or panics. Even the really bad ones.\n",
    "    a timeout can be specified. But the enumerate adapter is rather slow, so it can take over 5 seconds just to do that on a slow computer. Therefore a timeout in the range of 10 seconds is needed to avoid false detections.\n",
    "    returns either \"ok\", \"timeout\" or \"error\"\n",
    "    #TODO: not tested on unix systems, might required a change in the python command to call.\n",
    "    \"\"\"\n",
    "    status = \"ok\" # default case\n",
    "    with tempfile.NamedTemporaryFile(mode=\"w\", suffix=\".py\", delete=False, encoding=\"utf-8\") as f:\n",
    "        f.write(_file_template.format(shader_code))\n",
    "        f.flush()\n",
    "        try:\n",
    "            p = subprocess.run([\"python\", f.name], capture_output=True, timeout=timeout) # this might not work as expect on Linux ...\n",
    "            \n",
    "        except Exception as e:\n",
    "            if isinstance(e, subprocess.TimeoutExpired):\n",
    "                status = \"timeout\"\n",
    "            else:\n",
    "                status = \"other validation error\"\n",
    "    \n",
    "    # cleanup temp file, delete_on_close was only added in Python 3.12?\n",
    "    os.remove(f.name)\n",
    "        \n",
    "    if status == \"ok\":\n",
    "        if p.returncode != 0:\n",
    "            msg = p.stderr.decode(\"utf-8\")\n",
    "            if \"panic\" in msg:\n",
    "                status = \"panic\"\n",
    "                print(msg)\n",
    "            else:\n",
    "                status = \"error\"\n",
    "    return status\n",
    "\n",
    "# load the full thing?\n",
    "api_df = load_data(\"./data/annotated/api\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 7 in the first 1578\n",
    "for row in tqdm(api_df.itertuples()):\n",
    "    res = _run_shader_in_subprocess(row.image_code)\n",
    "    if res == \"panic\":\n",
    "        print(\"\\n from:\", row.id)\n",
    "        print(res)\n",
    "        # break"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "467\n"
     ]
    }
   ],
   "source": [
    "new_code = \"\"\"\n",
    "void mainImage( out vec4 fragColor, in vec2 fragCoord )\n",
    "{\n",
    "    // Normalized pixel coordinates (from 0 to 1)\n",
    "    vec2 uv = fragCoord/iResolution.xy;\n",
    "\n",
    "    // Time varying pixel color\n",
    "    vec3 col = 0.5 + 0.5*cos(iTime+uv.xyx+vec3(0,2,4));\n",
    "\n",
    "    // Output to screen\n",
    "    fragColor = vec4(col,1.0);\n",
    "}\n",
    "\"\"\"\n",
    "\n",
    "new_codes = [new_code] * len(ref_ds)\n",
    "\n",
    "print(len(new_codes))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "# testing against the new code to make sure it works. expected 1.0 variation and also might take 30 minutes...\n",
    "import evaluate\n",
    "\n",
    "shadermatch = evaluate.load(\"Vipitis/shadermatch\") # public\n",
    "\n",
    "# res = shadermatch.compute(references=ref_ds[\"image_code\"], predictions=new_codes)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import datasets\n",
    "\n",
    "loaded_ds = datasets.load_from_disk(\"./data/prepared/\")\n",
    "loaded_df = loaded_ds.to_pandas()\n",
    "# loaded_df.drop(columns=[\"function\", \"docstring\", \"needed\"], inplace=True)\n",
    "\n",
    "ref_ds = datasets.load_dataset(\"Vipitis/Shadereval-experiments-dev\", revision=\"main\", split=\"test\")\n",
    "ref_df = ref_ds.to_pandas()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# order is still the same\n",
    "loaded_ds[\"model_inp\"] == ref_ds[\"model_inp\"]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>id</th>\n",
       "      <th>author</th>\n",
       "      <th>date</th>\n",
       "      <th>image_code</th>\n",
       "      <th>license</th>\n",
       "      <th>func_bytes</th>\n",
       "      <th>functions</th>\n",
       "      <th>comment</th>\n",
       "      <th>header</th>\n",
       "      <th>body</th>\n",
       "      <th>model_inp</th>\n",
       "      <th>function_frequency</th>\n",
       "      <th>header_frequency</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>68</th>\n",
       "      <td>MsyXDV</td>\n",
       "      <td>SparkX120</td>\n",
       "      <td>2016-06-23 04:30:06</td>\n",
       "      <td>/**\\n * Adapted from my Mandelbrot-JS Project\\...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[0, 1289, 1346, 1346, 2956]</td>\n",
       "      <td>[[0, 1289, 1346, 1346, 2956]]</td>\n",
       "      <td>/**\\n * Adapted from my Mandelbrot-JS Project\\...</td>\n",
       "      <td>void mainImage( out vec4 fragColor, in vec2 fr...</td>\n",
       "      <td>\\n    const float iterations = 512.;\\n    c...</td>\n",
       "      <td>/**\\n * Adapted from my Mandelbrot-JS Project\\...</td>\n",
       "      <td>1</td>\n",
       "      <td>16281</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>315</th>\n",
       "      <td>NdfBDl</td>\n",
       "      <td>jackakers13</td>\n",
       "      <td>2022-02-22 10:11:58</td>\n",
       "      <td>// Created by Jack Akers on February 22, 2022....</td>\n",
       "      <td>cc0-1.0</td>\n",
       "      <td>[0, 140, 195, 214, 746]</td>\n",
       "      <td>[[0, 140, 195, 214, 746]]</td>\n",
       "      <td>// Created by Jack Akers on February 22, 2022....</td>\n",
       "      <td>void mainImage(out vec4 fragColor, in vec2 fra...</td>\n",
       "      <td>\\n    vec2 center = vec2(iResolution.x/2.0, iR...</td>\n",
       "      <td>// Created by Jack Akers on February 22, 2022....</td>\n",
       "      <td>1</td>\n",
       "      <td>901</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>263</th>\n",
       "      <td>NlXXzs</td>\n",
       "      <td>iq</td>\n",
       "      <td>2021-07-14 17:22:14</td>\n",
       "      <td>// The MIT License\\n// Copyright © 2021 Inigo ...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[2362, 2433, 2464, 2464, 2555]</td>\n",
       "      <td>[[1209, 1475, 1517, 1517, 1710], [1712, 1742, ...</td>\n",
       "      <td>// https://iquilezles.org/www/articles/distfun...</td>\n",
       "      <td>float sdBox( vec3 p, vec3 b )\\n{</td>\n",
       "      <td>\\n    vec3 d = abs(p) - b;\\n    return min(max...</td>\n",
       "      <td>// https://iquilezles.org/www/articles/distfun...</td>\n",
       "      <td>61</td>\n",
       "      <td>579</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>181</th>\n",
       "      <td>Ws3XWl</td>\n",
       "      <td>iq</td>\n",
       "      <td>2019-11-13 23:45:13</td>\n",
       "      <td>// The MIT License\\n// Copyright © 2019 Inigo ...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[1856, 1926, 1957, 1957, 2048]</td>\n",
       "      <td>[[1655, 1729, 1751, 1751, 1854], [1856, 1926, ...</td>\n",
       "      <td>// http://iquilezles.org/www/articles/distfunc...</td>\n",
       "      <td>float sdBox( vec3 p, vec3 b )\\n{</td>\n",
       "      <td>\\n    vec3 d = abs(p) - b;\\n    return min(max...</td>\n",
       "      <td>// http://iquilezles.org/www/articles/distfunc...</td>\n",
       "      <td>61</td>\n",
       "      <td>579</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16</th>\n",
       "      <td>XljGDz</td>\n",
       "      <td>otaviogood</td>\n",
       "      <td>2015-04-07 03:58:40</td>\n",
       "      <td>/*--------------------------------------------...</td>\n",
       "      <td>cc0-1.0</td>\n",
       "      <td>[2998, 3034, 3075, 3075, 3168]</td>\n",
       "      <td>[[856, 856, 879, 879, 908], [909, 909, 932, 93...</td>\n",
       "      <td>// polynomial smooth min (k = 0.1);\\n</td>\n",
       "      <td>float smin( float a, float b, float k )\\n{</td>\n",
       "      <td>\\n    float h = clamp( 0.5+0.5*(b-a)/k, 0.0, 1...</td>\n",
       "      <td>// polynomial smooth min (k = 0.1);\\nfloat smi...</td>\n",
       "      <td>176</td>\n",
       "      <td>382</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>116</th>\n",
       "      <td>lt3BW2</td>\n",
       "      <td>iq</td>\n",
       "      <td>2018-11-01 04:49:23</td>\n",
       "      <td>// The MIT License\\n// Copyright © 2018 Inigo ...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[4550, 4614, 4646, 4646, 4880]</td>\n",
       "      <td>[[1702, 1702, 1739, 1739, 1764], [1766, 1766, ...</td>\n",
       "      <td>// http://iquilezles.org/www/articles/normalsS...</td>\n",
       "      <td>vec3 calcNormal( in vec3 pos )\\n{</td>\n",
       "      <td>\\n    const float ep = 0.0001;\\n    vec2 e = v...</td>\n",
       "      <td>// http://iquilezles.org/www/articles/normalsS...</td>\n",
       "      <td>6</td>\n",
       "      <td>360</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>310</th>\n",
       "      <td>stKSzc</td>\n",
       "      <td>iq</td>\n",
       "      <td>2022-01-03 03:36:24</td>\n",
       "      <td>// The MIT License\\n// Copyright © 2021 Inigo ...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[2294, 2340, 2372, 2372, 2611]</td>\n",
       "      <td>[[1588, 1643, 1699, 1699, 1990], [1992, 1992, ...</td>\n",
       "      <td>// https://iquilezles.org/articles/normalsSDF\\n</td>\n",
       "      <td>vec3 calcNormal( in vec3 pos )\\n{</td>\n",
       "      <td>\\n    vec2 e = vec2(1.0,-1.0)*0.5773;\\n    con...</td>\n",
       "      <td>// https://iquilezles.org/articles/normalsSDF\\...</td>\n",
       "      <td>23</td>\n",
       "      <td>360</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>455</th>\n",
       "      <td>dtGBRG</td>\n",
       "      <td>Dain</td>\n",
       "      <td>2023-12-11 01:18:25</td>\n",
       "      <td>// https://www.shadertoy.com/view/dtGBDz ortho...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[6779, 6843, 6885, 6885, 7175]</td>\n",
       "      <td>[[2307, 2307, 2430, 2430, 3560], [3562, 3562, ...</td>\n",
       "      <td>// https://www.shadertoy.com/view/Xds3zN rayma...</td>\n",
       "      <td>float calcAO( in vec3 pos, in vec3 nor )\\n{</td>\n",
       "      <td>\\n    float occ = 0.0;\\n    float sca = 1.0;\\n...</td>\n",
       "      <td>// https://www.shadertoy.com/view/Xds3zN rayma...</td>\n",
       "      <td>1</td>\n",
       "      <td>259</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>256</th>\n",
       "      <td>7tj3DG</td>\n",
       "      <td>mrange</td>\n",
       "      <td>2021-06-25 18:20:12</td>\n",
       "      <td>// License CC0: Random friday fractal\\n// Resu...</td>\n",
       "      <td>cc0-1.0</td>\n",
       "      <td>[584, 637, 659, 659, 828]</td>\n",
       "      <td>[[458, 458, 486, 506, 582], [584, 637, 659, 65...</td>\n",
       "      <td>// From: https://stackoverflow.com/a/17897228/...</td>\n",
       "      <td>vec3 hsv2rgb(vec3 c) {</td>\n",
       "      <td>\\n  const vec4 K = vec4(1.0, 2.0 / 3.0, 1.0 / ...</td>\n",
       "      <td>// From: https://stackoverflow.com/a/17897228/...</td>\n",
       "      <td>22</td>\n",
       "      <td>237</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>239</th>\n",
       "      <td>NdS3WW</td>\n",
       "      <td>mrange</td>\n",
       "      <td>2021-04-02 13:52:21</td>\n",
       "      <td>// License CC0: The monolith, 1x4x9\\n\\n#define...</td>\n",
       "      <td>cc0-1.0</td>\n",
       "      <td>[994, 1073, 1095, 1095, 1264]</td>\n",
       "      <td>[[715, 715, 743, 763, 839], [841, 865, 904, 90...</td>\n",
       "      <td>// https://stackoverflow.com/questions/1509590...</td>\n",
       "      <td>vec3 hsv2rgb(vec3 c) {</td>\n",
       "      <td>\\n  const vec4 K = vec4(1.0, 2.0 / 3.0, 1.0 / ...</td>\n",
       "      <td>// https://stackoverflow.com/questions/1509590...</td>\n",
       "      <td>22</td>\n",
       "      <td>237</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         id       author                date  \\\n",
       "68   MsyXDV    SparkX120 2016-06-23 04:30:06   \n",
       "315  NdfBDl  jackakers13 2022-02-22 10:11:58   \n",
       "263  NlXXzs           iq 2021-07-14 17:22:14   \n",
       "181  Ws3XWl           iq 2019-11-13 23:45:13   \n",
       "16   XljGDz   otaviogood 2015-04-07 03:58:40   \n",
       "116  lt3BW2           iq 2018-11-01 04:49:23   \n",
       "310  stKSzc           iq 2022-01-03 03:36:24   \n",
       "455  dtGBRG         Dain 2023-12-11 01:18:25   \n",
       "256  7tj3DG       mrange 2021-06-25 18:20:12   \n",
       "239  NdS3WW       mrange 2021-04-02 13:52:21   \n",
       "\n",
       "                                            image_code  license  \\\n",
       "68   /**\\n * Adapted from my Mandelbrot-JS Project\\...      mit   \n",
       "315  // Created by Jack Akers on February 22, 2022....  cc0-1.0   \n",
       "263  // The MIT License\\n// Copyright © 2021 Inigo ...      mit   \n",
       "181  // The MIT License\\n// Copyright © 2019 Inigo ...      mit   \n",
       "16   /*--------------------------------------------...  cc0-1.0   \n",
       "116  // The MIT License\\n// Copyright © 2018 Inigo ...      mit   \n",
       "310  // The MIT License\\n// Copyright © 2021 Inigo ...      mit   \n",
       "455  // https://www.shadertoy.com/view/dtGBDz ortho...      mit   \n",
       "256  // License CC0: Random friday fractal\\n// Resu...  cc0-1.0   \n",
       "239  // License CC0: The monolith, 1x4x9\\n\\n#define...  cc0-1.0   \n",
       "\n",
       "                         func_bytes  \\\n",
       "68      [0, 1289, 1346, 1346, 2956]   \n",
       "315         [0, 140, 195, 214, 746]   \n",
       "263  [2362, 2433, 2464, 2464, 2555]   \n",
       "181  [1856, 1926, 1957, 1957, 2048]   \n",
       "16   [2998, 3034, 3075, 3075, 3168]   \n",
       "116  [4550, 4614, 4646, 4646, 4880]   \n",
       "310  [2294, 2340, 2372, 2372, 2611]   \n",
       "455  [6779, 6843, 6885, 6885, 7175]   \n",
       "256       [584, 637, 659, 659, 828]   \n",
       "239   [994, 1073, 1095, 1095, 1264]   \n",
       "\n",
       "                                             functions  \\\n",
       "68                       [[0, 1289, 1346, 1346, 2956]]   \n",
       "315                          [[0, 140, 195, 214, 746]]   \n",
       "263  [[1209, 1475, 1517, 1517, 1710], [1712, 1742, ...   \n",
       "181  [[1655, 1729, 1751, 1751, 1854], [1856, 1926, ...   \n",
       "16   [[856, 856, 879, 879, 908], [909, 909, 932, 93...   \n",
       "116  [[1702, 1702, 1739, 1739, 1764], [1766, 1766, ...   \n",
       "310  [[1588, 1643, 1699, 1699, 1990], [1992, 1992, ...   \n",
       "455  [[2307, 2307, 2430, 2430, 3560], [3562, 3562, ...   \n",
       "256  [[458, 458, 486, 506, 582], [584, 637, 659, 65...   \n",
       "239  [[715, 715, 743, 763, 839], [841, 865, 904, 90...   \n",
       "\n",
       "                                               comment  \\\n",
       "68   /**\\n * Adapted from my Mandelbrot-JS Project\\...   \n",
       "315  // Created by Jack Akers on February 22, 2022....   \n",
       "263  // https://iquilezles.org/www/articles/distfun...   \n",
       "181  // http://iquilezles.org/www/articles/distfunc...   \n",
       "16               // polynomial smooth min (k = 0.1);\\n   \n",
       "116  // http://iquilezles.org/www/articles/normalsS...   \n",
       "310    // https://iquilezles.org/articles/normalsSDF\\n   \n",
       "455  // https://www.shadertoy.com/view/Xds3zN rayma...   \n",
       "256  // From: https://stackoverflow.com/a/17897228/...   \n",
       "239  // https://stackoverflow.com/questions/1509590...   \n",
       "\n",
       "                                                header  \\\n",
       "68   void mainImage( out vec4 fragColor, in vec2 fr...   \n",
       "315  void mainImage(out vec4 fragColor, in vec2 fra...   \n",
       "263                   float sdBox( vec3 p, vec3 b )\\n{   \n",
       "181                   float sdBox( vec3 p, vec3 b )\\n{   \n",
       "16          float smin( float a, float b, float k )\\n{   \n",
       "116                  vec3 calcNormal( in vec3 pos )\\n{   \n",
       "310                  vec3 calcNormal( in vec3 pos )\\n{   \n",
       "455        float calcAO( in vec3 pos, in vec3 nor )\\n{   \n",
       "256                             vec3 hsv2rgb(vec3 c) {   \n",
       "239                             vec3 hsv2rgb(vec3 c) {   \n",
       "\n",
       "                                                  body  \\\n",
       "68      \\n    const float iterations = 512.;\\n    c...   \n",
       "315  \\n    vec2 center = vec2(iResolution.x/2.0, iR...   \n",
       "263  \\n    vec3 d = abs(p) - b;\\n    return min(max...   \n",
       "181  \\n    vec3 d = abs(p) - b;\\n    return min(max...   \n",
       "16   \\n    float h = clamp( 0.5+0.5*(b-a)/k, 0.0, 1...   \n",
       "116  \\n    const float ep = 0.0001;\\n    vec2 e = v...   \n",
       "310  \\n    vec2 e = vec2(1.0,-1.0)*0.5773;\\n    con...   \n",
       "455  \\n    float occ = 0.0;\\n    float sca = 1.0;\\n...   \n",
       "256  \\n  const vec4 K = vec4(1.0, 2.0 / 3.0, 1.0 / ...   \n",
       "239  \\n  const vec4 K = vec4(1.0, 2.0 / 3.0, 1.0 / ...   \n",
       "\n",
       "                                             model_inp  function_frequency  \\\n",
       "68   /**\\n * Adapted from my Mandelbrot-JS Project\\...                   1   \n",
       "315  // Created by Jack Akers on February 22, 2022....                   1   \n",
       "263  // https://iquilezles.org/www/articles/distfun...                  61   \n",
       "181  // http://iquilezles.org/www/articles/distfunc...                  61   \n",
       "16   // polynomial smooth min (k = 0.1);\\nfloat smi...                 176   \n",
       "116  // http://iquilezles.org/www/articles/normalsS...                   6   \n",
       "310  // https://iquilezles.org/articles/normalsSDF\\...                  23   \n",
       "455  // https://www.shadertoy.com/view/Xds3zN rayma...                   1   \n",
       "256  // From: https://stackoverflow.com/a/17897228/...                  22   \n",
       "239  // https://stackoverflow.com/questions/1509590...                  22   \n",
       "\n",
       "     header_frequency  \n",
       "68              16281  \n",
       "315               901  \n",
       "263               579  \n",
       "181               579  \n",
       "16                382  \n",
       "116               360  \n",
       "310               360  \n",
       "455               259  \n",
       "256               237  \n",
       "239               237  "
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "loaded_df.sort_values(\"header_frequency\", ascending=False).head(10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Axes: >"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "loaded_df.sort_values(by=\"header_frequency\").reset_index(drop=True)[[\"header_frequency\", \"function_frequency\"]].plot(ylim=(0,200), xticks=[])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Axes: >"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "loaded_df[[\"header_frequency\", \"function_frequency\"]].plot.box(ylim=(0,50))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>id</th>\n",
       "      <th>author</th>\n",
       "      <th>date</th>\n",
       "      <th>image_code</th>\n",
       "      <th>license</th>\n",
       "      <th>func_bytes</th>\n",
       "      <th>functions</th>\n",
       "      <th>comment</th>\n",
       "      <th>header</th>\n",
       "      <th>body</th>\n",
       "      <th>model_inp</th>\n",
       "      <th>function_frequency</th>\n",
       "      <th>header_frequency</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>308</th>\n",
       "      <td>7tVXRt</td>\n",
       "      <td>iq</td>\n",
       "      <td>2022-01-03 19:29:53</td>\n",
       "      <td>// The MIT License\\n// Copyright © 2022 Inigo ...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[1914, 1959, 2046, 2046, 2304]</td>\n",
       "      <td>[[1396, 1470, 1532, 1532, 1716], [1718, 1718, ...</td>\n",
       "      <td>// https://iquilezles.org/articles/rmshadows\\n</td>\n",
       "      <td>float calcSoftshadow( in vec3 ro, in vec3 rd, ...</td>\n",
       "      <td>\\n\\tfloat res = 1.0;\\n    float t = tmin;\\n   ...</td>\n",
       "      <td>// https://iquilezles.org/articles/rmshadows\\n...</td>\n",
       "      <td>1</td>\n",
       "      <td>10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>171</th>\n",
       "      <td>WsV3zz</td>\n",
       "      <td>iq</td>\n",
       "      <td>2019-10-01 00:21:04</td>\n",
       "      <td>// The MIT License\\n// Copyright © 2019 Inigo ...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[2155, 2173, 2201, 2201, 2359]</td>\n",
       "      <td>[[1950, 1989, 2011, 2011, 2153], [2155, 2173, ...</td>\n",
       "      <td>// gradient noise\\n</td>\n",
       "      <td>float gnoise( in float p )\\n{</td>\n",
       "      <td>\\n    int   i = int(floor(p));\\n    float f = ...</td>\n",
       "      <td>// gradient noise\\nfloat gnoise( in float p )\\n{</td>\n",
       "      <td>1</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>378</th>\n",
       "      <td>fddBD2</td>\n",
       "      <td>iq</td>\n",
       "      <td>2022-07-07 22:16:10</td>\n",
       "      <td>// The MIT License\\n// Copyright © 2022 Inigo ...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[4334, 4392, 4457, 4475, 4820]</td>\n",
       "      <td>[[2473, 2473, 2893, 2893, 3577], [3579, 3579, ...</td>\n",
       "      <td>// https://iquilezles.org/articles/filterablep...</td>\n",
       "      <td>float gridTextureGradBox( in vec2 p, in vec2 d...</td>\n",
       "      <td>\\n    vec2 w = max(abs(ddx), abs(ddy)) + 0.01;...</td>\n",
       "      <td>// https://iquilezles.org/articles/filterablep...</td>\n",
       "      <td>2</td>\n",
       "      <td>6</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>390</th>\n",
       "      <td>NtKBWh</td>\n",
       "      <td>mrange</td>\n",
       "      <td>2022-09-29 05:45:01</td>\n",
       "      <td>// License CC0: Dark chocolate FBM\\n//  Workin...</td>\n",
       "      <td>cc0-1.0</td>\n",
       "      <td>[1768, 1868, 1907, 1907, 1992]</td>\n",
       "      <td>[[592, 592, 614, 614, 760], [1055, 1155, 1174,...</td>\n",
       "      <td>// License: MIT, author: Inigo Quilez, found: ...</td>\n",
       "      <td>float pmin(float a, float b, float k) {</td>\n",
       "      <td>\\n  float h = clamp(0.5+0.5*(b-a)/k, 0.0, 1.0)...</td>\n",
       "      <td>// License: MIT, author: Inigo Quilez, found: ...</td>\n",
       "      <td>63</td>\n",
       "      <td>107</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>319</th>\n",
       "      <td>7sXfDH</td>\n",
       "      <td>mrange</td>\n",
       "      <td>2022-02-13 22:28:54</td>\n",
       "      <td>// License CC0: Follow the light\\n// Result af...</td>\n",
       "      <td>cc0-1.0</td>\n",
       "      <td>[1522, 1605, 1644, 1644, 1733]</td>\n",
       "      <td>[[968, 1068, 1089, 1089, 1162], [1163, 1263, 1...</td>\n",
       "      <td>// License: MIT, author: Inigo Quilez, found: ...</td>\n",
       "      <td>float pmin(float a, float b, float k) {</td>\n",
       "      <td>\\n  float h = clamp( 0.5+0.5*(b-a)/k, 0.0, 1.0...</td>\n",
       "      <td>// License: MIT, author: Inigo Quilez, found: ...</td>\n",
       "      <td>22</td>\n",
       "      <td>107</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>259</th>\n",
       "      <td>slj3Dd</td>\n",
       "      <td>iq</td>\n",
       "      <td>2021-06-30 05:08:46</td>\n",
       "      <td>// The MIT License\\n// Copyright © 2021 Inigo ...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[1261, 1351, 1415, 1499, 2515]</td>\n",
       "      <td>[[1261, 1351, 1415, 1499, 2515], [2517, 2517, ...</td>\n",
       "      <td>// The arrow goes from a to b. It's thickness ...</td>\n",
       "      <td>float sdArrow( in vec2 p, vec2 a, vec2 b, floa...</td>\n",
       "      <td>\\n    const float k = 3.0;   // arrow head rat...</td>\n",
       "      <td>// The arrow goes from a to b. It's thickness ...</td>\n",
       "      <td>1</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>263</th>\n",
       "      <td>NlXXzs</td>\n",
       "      <td>iq</td>\n",
       "      <td>2021-07-14 17:22:14</td>\n",
       "      <td>// The MIT License\\n// Copyright © 2021 Inigo ...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[2362, 2433, 2464, 2464, 2555]</td>\n",
       "      <td>[[1209, 1475, 1517, 1517, 1710], [1712, 1742, ...</td>\n",
       "      <td>// https://iquilezles.org/www/articles/distfun...</td>\n",
       "      <td>float sdBox( vec3 p, vec3 b )\\n{</td>\n",
       "      <td>\\n    vec3 d = abs(p) - b;\\n    return min(max...</td>\n",
       "      <td>// https://iquilezles.org/www/articles/distfun...</td>\n",
       "      <td>61</td>\n",
       "      <td>579</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>69</th>\n",
       "      <td>MsVXDt</td>\n",
       "      <td>otaviogood</td>\n",
       "      <td>2016-07-20 06:51:21</td>\n",
       "      <td>/*--------------------------------------------...</td>\n",
       "      <td>cc0-1.0</td>\n",
       "      <td>[2400, 2602, 2636, 2636, 2747]</td>\n",
       "      <td>[[1174, 1203, 1222, 1222, 1267], [1268, 1268, ...</td>\n",
       "      <td>// ---- shapes defined by distance fields ----...</td>\n",
       "      <td>float sdBox(vec3 p, vec3 radius)\\n{</td>\n",
       "      <td>\\n  vec3 dist = abs(p) - radius;\\n  return min...</td>\n",
       "      <td>// ---- shapes defined by distance fields ----...</td>\n",
       "      <td>10</td>\n",
       "      <td>10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>408</th>\n",
       "      <td>ddd3zs</td>\n",
       "      <td>iq</td>\n",
       "      <td>2023-03-02 04:52:06</td>\n",
       "      <td>// The MIT License\\n// Copyright © 2023 Inigo ...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[2980, 3029, 3081, 3081, 3195]</td>\n",
       "      <td>[[1213, 1264, 1328, 1451, 2914], [2980, 3029, ...</td>\n",
       "      <td>// https://iquilezles.org/articles/distfunctio...</td>\n",
       "      <td>float sdCapsule( vec3 p, vec3 a, vec3 b, float...</td>\n",
       "      <td>\\n\\tvec3 pa = p-a, ba = b-a;\\n\\tfloat h = clam...</td>\n",
       "      <td>// https://iquilezles.org/articles/distfunctio...</td>\n",
       "      <td>56</td>\n",
       "      <td>160</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>300</th>\n",
       "      <td>sl3XRn</td>\n",
       "      <td>iq</td>\n",
       "      <td>2021-12-04 00:18:43</td>\n",
       "      <td>// The MIT License\\n// Copyright © 2021 Inigo ...</td>\n",
       "      <td>mit</td>\n",
       "      <td>[2083, 2158, 2208, 2208, 2236]</td>\n",
       "      <td>[[1114, 1114, 1149, 1149, 1177], [1179, 1179, ...</td>\n",
       "      <td>// https://iquilezles.org/www/articles/distfun...</td>\n",
       "      <td>float sdDisk( in vec2 p, in vec2 c, in float r...</td>\n",
       "      <td>\\n    return length(p-c)-r;\\n}</td>\n",
       "      <td>// https://iquilezles.org/www/articles/distfun...</td>\n",
       "      <td>2</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         id      author                date  \\\n",
       "308  7tVXRt          iq 2022-01-03 19:29:53   \n",
       "171  WsV3zz          iq 2019-10-01 00:21:04   \n",
       "378  fddBD2          iq 2022-07-07 22:16:10   \n",
       "390  NtKBWh      mrange 2022-09-29 05:45:01   \n",
       "319  7sXfDH      mrange 2022-02-13 22:28:54   \n",
       "259  slj3Dd          iq 2021-06-30 05:08:46   \n",
       "263  NlXXzs          iq 2021-07-14 17:22:14   \n",
       "69   MsVXDt  otaviogood 2016-07-20 06:51:21   \n",
       "408  ddd3zs          iq 2023-03-02 04:52:06   \n",
       "300  sl3XRn          iq 2021-12-04 00:18:43   \n",
       "\n",
       "                                            image_code  license  \\\n",
       "308  // The MIT License\\n// Copyright © 2022 Inigo ...      mit   \n",
       "171  // The MIT License\\n// Copyright © 2019 Inigo ...      mit   \n",
       "378  // The MIT License\\n// Copyright © 2022 Inigo ...      mit   \n",
       "390  // License CC0: Dark chocolate FBM\\n//  Workin...  cc0-1.0   \n",
       "319  // License CC0: Follow the light\\n// Result af...  cc0-1.0   \n",
       "259  // The MIT License\\n// Copyright © 2021 Inigo ...      mit   \n",
       "263  // The MIT License\\n// Copyright © 2021 Inigo ...      mit   \n",
       "69   /*--------------------------------------------...  cc0-1.0   \n",
       "408  // The MIT License\\n// Copyright © 2023 Inigo ...      mit   \n",
       "300  // The MIT License\\n// Copyright © 2021 Inigo ...      mit   \n",
       "\n",
       "                         func_bytes  \\\n",
       "308  [1914, 1959, 2046, 2046, 2304]   \n",
       "171  [2155, 2173, 2201, 2201, 2359]   \n",
       "378  [4334, 4392, 4457, 4475, 4820]   \n",
       "390  [1768, 1868, 1907, 1907, 1992]   \n",
       "319  [1522, 1605, 1644, 1644, 1733]   \n",
       "259  [1261, 1351, 1415, 1499, 2515]   \n",
       "263  [2362, 2433, 2464, 2464, 2555]   \n",
       "69   [2400, 2602, 2636, 2636, 2747]   \n",
       "408  [2980, 3029, 3081, 3081, 3195]   \n",
       "300  [2083, 2158, 2208, 2208, 2236]   \n",
       "\n",
       "                                             functions  \\\n",
       "308  [[1396, 1470, 1532, 1532, 1716], [1718, 1718, ...   \n",
       "171  [[1950, 1989, 2011, 2011, 2153], [2155, 2173, ...   \n",
       "378  [[2473, 2473, 2893, 2893, 3577], [3579, 3579, ...   \n",
       "390  [[592, 592, 614, 614, 760], [1055, 1155, 1174,...   \n",
       "319  [[968, 1068, 1089, 1089, 1162], [1163, 1263, 1...   \n",
       "259  [[1261, 1351, 1415, 1499, 2515], [2517, 2517, ...   \n",
       "263  [[1209, 1475, 1517, 1517, 1710], [1712, 1742, ...   \n",
       "69   [[1174, 1203, 1222, 1222, 1267], [1268, 1268, ...   \n",
       "408  [[1213, 1264, 1328, 1451, 2914], [2980, 3029, ...   \n",
       "300  [[1114, 1114, 1149, 1149, 1177], [1179, 1179, ...   \n",
       "\n",
       "                                               comment  \\\n",
       "308     // https://iquilezles.org/articles/rmshadows\\n   \n",
       "171                                // gradient noise\\n   \n",
       "378  // https://iquilezles.org/articles/filterablep...   \n",
       "390  // License: MIT, author: Inigo Quilez, found: ...   \n",
       "319  // License: MIT, author: Inigo Quilez, found: ...   \n",
       "259  // The arrow goes from a to b. It's thickness ...   \n",
       "263  // https://iquilezles.org/www/articles/distfun...   \n",
       "69   // ---- shapes defined by distance fields ----...   \n",
       "408  // https://iquilezles.org/articles/distfunctio...   \n",
       "300  // https://iquilezles.org/www/articles/distfun...   \n",
       "\n",
       "                                                header  \\\n",
       "308  float calcSoftshadow( in vec3 ro, in vec3 rd, ...   \n",
       "171                      float gnoise( in float p )\\n{   \n",
       "378  float gridTextureGradBox( in vec2 p, in vec2 d...   \n",
       "390            float pmin(float a, float b, float k) {   \n",
       "319            float pmin(float a, float b, float k) {   \n",
       "259  float sdArrow( in vec2 p, vec2 a, vec2 b, floa...   \n",
       "263                   float sdBox( vec3 p, vec3 b )\\n{   \n",
       "69                 float sdBox(vec3 p, vec3 radius)\\n{   \n",
       "408  float sdCapsule( vec3 p, vec3 a, vec3 b, float...   \n",
       "300  float sdDisk( in vec2 p, in vec2 c, in float r...   \n",
       "\n",
       "                                                  body  \\\n",
       "308  \\n\\tfloat res = 1.0;\\n    float t = tmin;\\n   ...   \n",
       "171  \\n    int   i = int(floor(p));\\n    float f = ...   \n",
       "378  \\n    vec2 w = max(abs(ddx), abs(ddy)) + 0.01;...   \n",
       "390  \\n  float h = clamp(0.5+0.5*(b-a)/k, 0.0, 1.0)...   \n",
       "319  \\n  float h = clamp( 0.5+0.5*(b-a)/k, 0.0, 1.0...   \n",
       "259  \\n    const float k = 3.0;   // arrow head rat...   \n",
       "263  \\n    vec3 d = abs(p) - b;\\n    return min(max...   \n",
       "69   \\n  vec3 dist = abs(p) - radius;\\n  return min...   \n",
       "408  \\n\\tvec3 pa = p-a, ba = b-a;\\n\\tfloat h = clam...   \n",
       "300                     \\n    return length(p-c)-r;\\n}   \n",
       "\n",
       "                                             model_inp  function_frequency  \\\n",
       "308  // https://iquilezles.org/articles/rmshadows\\n...                   1   \n",
       "171   // gradient noise\\nfloat gnoise( in float p )\\n{                   1   \n",
       "378  // https://iquilezles.org/articles/filterablep...                   2   \n",
       "390  // License: MIT, author: Inigo Quilez, found: ...                  63   \n",
       "319  // License: MIT, author: Inigo Quilez, found: ...                  22   \n",
       "259  // The arrow goes from a to b. It's thickness ...                   1   \n",
       "263  // https://iquilezles.org/www/articles/distfun...                  61   \n",
       "69   // ---- shapes defined by distance fields ----...                  10   \n",
       "408  // https://iquilezles.org/articles/distfunctio...                  56   \n",
       "300  // https://iquilezles.org/www/articles/distfun...                   2   \n",
       "\n",
       "     header_frequency  \n",
       "308                10  \n",
       "171                 2  \n",
       "378                 6  \n",
       "390               107  \n",
       "319               107  \n",
       "259                 2  \n",
       "263               579  \n",
       "69                 10  \n",
       "408               160  \n",
       "300                 2  "
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "loaded_df[loaded_df[\"header\"].duplicated()].sort_values(\"header\").head(10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "// License: MIT, author: Inigo Quilez, found: https://iquilezles.org/articles/spherefunctions\n",
      "float sphered(vec3 ro, vec3 rd, vec4 sph, float dbuffer) {\n",
      "// License: MIT, author: Inigo Quilez, found: https://www.iquilezles.org/www/articles/spherefunctions/spherefunctions.htm\n",
      "float sphered(vec3 ro, vec3 rd, vec4 sph, float dbuffer) {\n"
     ]
    }
   ],
   "source": [
    "print(loaded_df[\"model_inp\"][320])\n",
    "print(loaded_df[\"model_inp\"][386])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "ceea255050e34812b90c2224f491a08b",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Uploading the dataset shards:   0%|          | 0/1 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "879d07e913d345e980393414e889dc2a",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Creating parquet from Arrow format:   0%|          | 0/1 [00:00<?, ?ba/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "CommitInfo(commit_url='https://huggingface.co/datasets/Vipitis/Shadereval-experiments-dev/commit/274eb4d3017d59da2a1f48bc7194be1545de919f', commit_message='adding header_frequency', commit_description='', oid='274eb4d3017d59da2a1f48bc7194be1545de919f', pr_url=None, pr_revision=None, pr_num=None)"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import datasets\n",
    "loaded_df\n",
    "func_ds = datasets.Dataset.from_pandas(loaded_df)\n",
    "# func_ds = func_ds.remove_columns(['__index_level_0__'])\n",
    "func_ds\n",
    "# func_ds.push_to_hub(\"Vipitis/Shadereval-experiments-dev\", commit_message=\"adding header_frequency\", private=False, split=\"test\") # don't easily uncomment again..."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "19622\n",
      "3858\n"
     ]
    }
   ],
   "source": [
    "# make a plot of shaders per year. compare s20k, api and plot the overlap too!\n",
    "import os\n",
    "import jsonlines\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "def load_data(data_dir: os.PathLike) -> pd.DataFrame:\n",
    "    lines = []\n",
    "    for file in os.listdir(data_dir):\n",
    "        if file.endswith(\".jsonl\"):\n",
    "            with jsonlines.open(os.path.join(data_dir, file)) as reader:\n",
    "                for obj in reader:\n",
    "                    lines.append(obj)\n",
    "\n",
    "    out_df = pd.DataFrame(lines)\n",
    "    out_df[\"date\"] = pd.to_datetime(out_df[\"date\"].astype(int), unit=\"s\")\n",
    "    return out_df\n",
    "\n",
    "all_shaders20k = load_data(\"./data/annotated\")\n",
    "# all_api = load_data(\"./data/annotated/api\")\n",
    "print(len(all_shaders20k))\n",
    "# print(len(all_api))\n",
    "\n",
    "s20k_ids = {k:v for k,v in all_shaders20k.groupby(all_shaders20k[\"date\"].dt.year)[\"id\"].unique().items()}\n",
    "print(len(s20k_ids[2021]))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "4725"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ids_path = \"./data/ids/\"\n",
    "api_ids = {}\n",
    "for file in os.listdir(ids_path):\n",
    "    if file.endswith(\".txt\") and file.startswith(\"api_\"):\n",
    "        year = int(file.split(\"_\")[1].removesuffix(\".txt\"))\n",
    "        with open(os.path.join(ids_path, file), \"r\", encoding=\"utf-8\") as f:\n",
    "            api_ids[year] = list(f.read().splitlines())\n",
    "\n",
    "len(api_ids[2021])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{2014: 156,\n",
       " 2015: 570,\n",
       " 2016: 695,\n",
       " 2017: 851,\n",
       " 2018: 818,\n",
       " 2019: 1179,\n",
       " 2020: 1456,\n",
       " 2021: 1569}"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "overlap = {k: len(set(v).intersection(api_ids[k])) for k,v in s20k_ids.items()}\n",
    "overlap"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 800x500 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(8,5))\n",
    "plt.bar(api_ids.keys(), [len(api_ids[k]) for k in api_ids.keys()], label=\"API\", alpha=0.7)\n",
    "plt.bar(s20k_ids.keys(), [len(s20k_ids[k]) for k in s20k_ids.keys()], label=\"shaders20k\", alpha=0.7, bottom=[-len(s20k_ids[k])+overlap[k] for k in s20k_ids.keys()])\n",
    "plt.legend()\n",
    "plt.xticks(list(api_ids.keys()))\n",
    "# plt.yticks(list(range(-3000, 5000, 1000)))\n",
    "plt.savefig(\"../Thesis/report/texs/charts/shaders_per_year.png\")\n",
    "plt.show()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}