swiss_roll.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "%load_ext autoreload\n",
    "%autoreload 2\n",
    "\n",
    "import itertools\n",
    "from mpl_toolkits.mplot3d import Axes3D\n",
    "import matplotlib.pyplot as plt\n",
    "from matplotlib import cm\n",
    "from matplotlib.ticker import LinearLocator, FormatStrFormatter\n",
    "import numpy as np\n",
    "\n",
    "import contextlib\n",
    "import os\n",
    "from importlib import reload\n",
    "\n",
    "import torchvision\n",
    "from torch import autograd\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "import torch\n",
    "\n",
    "plt.rcParams['axes.facecolor'] = 'white'\n",
    "plt.rcParams['font.family'] = 'serif'\n",
    "plt.rcParams['font.sans-serif'] = ['Lucida Grande']\n",
    "plt.rcParams['font.size'] = 12"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import gait\n",
    "import utils\n",
    "from gait import renyi_mixture_divergence, rbf_kernel, poly_kernel, generic_kernel\n",
    "from gait import test_mixture_divergence, renyi_mixture_divergence_stable\n",
    "from gait import renyi_mixture_divergence, renyi_sim_divergence, renyi_sim_divergence_stable\n",
    "from gait import breg_mixture_divergence_stable, breg_sim_divergence_stable\n",
    "from gait import breg_mixture_divergence, breg_sim_divergence"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# torch.set_default_dtype(torch.float64)\n",
    "\n",
    "class GeneratorFC(nn.Module):\n",
    "    def __init__(self, input_size, output_size, hidden_dims):\n",
    "        super(GeneratorFC, self).__init__()\n",
    "        self.layers = []\n",
    "\n",
    "        prev_dim = input_size\n",
    "        for hidden_dim in hidden_dims:\n",
    "            self.layers.append(nn.Linear(prev_dim, hidden_dim))\n",
    "#             self.layers.append(nn.BatchNorm1d(hidden_dim))\n",
    "            self.layers.append(nn.ReLU())\n",
    "            prev_dim = hidden_dim\n",
    "        self.layers.append(nn.Linear(prev_dim, output_size))\n",
    "\n",
    "        self.layer_module = nn.ModuleList(self.layers)\n",
    "\n",
    "    def forward(self, x):\n",
    "        out = x\n",
    "        for layer in self.layer_module:\n",
    "            out = layer(out)\n",
    "        return out"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.datasets.samples_generator import make_swiss_roll\n",
    "\n",
    "d, z_dim, gen_hiddens = 2, 2, [500, 500]\n",
    "batch_size_p = 200\n",
    "batch_size_q = 200"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "generator = GeneratorFC(z_dim, d, gen_hiddens).cuda()\n",
    "\n",
    "g_optimizer = torch.optim.Adam(generator.parameters(), lr=5e-4, betas=(0.9, 0.999))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "alpha = 1\n",
    "sigma_mult = 1\n",
    "\n",
    "# emp_probs = 2*torch.ones(1, batch_size//2).cuda()/batch_size\n",
    "emp_probs_p = torch.ones(1, batch_size_p).cuda()/batch_size_p\n",
    "emp_probs_q = torch.ones(1, batch_size_q).cuda()/batch_size_q\n",
    "\n",
    "kernel = lambda x, y: generic_kernel(x, y, lambda u, v: rbf_kernel(u, v, sigmas=[0.2 * sigma_mult], log=True))\n",
    "# kernel = lambda x, y: generic_kernel(x, y, lambda u, v: multiquad_kernel(u, v, sigma=0.2))\n",
    "# kernel = lambda x, y: generic_kernel(x, y, lambda u, v: poly_kernel(u, v, degree=2))\n",
    "\n",
    "D = lambda x, y: breg_mixture_divergence_stable(emp_probs_p, x, emp_probs_q, y, kernel, symmetric=False)\n",
    "\n",
    "# D = lambda x, y: renyi_mixture_divergence_stable(emp_probs_p, x, emp_probs_q, y, kernel, alpha,\n",
    "#                                                  use_avg=False, use_full=False, symmetric=True)\n",
    "# D = lambda x, y: test_mixture_divergence(emp_probs_p, x, emp_probs_q, y, kernel,\n",
    "#                                          symmetric=True, use_avg=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 - 13.9981\n",
      "200 - 0.6382\n",
      "400 - 0.3631\n",
      "600 - 0.3934\n",
      "800 - 0.3738\n",
      "1000 - 0.3196\n",
      "1200 - 0.1488\n",
      "1400 - 0.2395\n",
      "1600 - 0.2924\n",
      "1800 - 0.2549\n",
      "2000 - 0.2633\n",
      "2200 - 0.2692\n",
      "2400 - 0.2586\n",
      "2600 - 0.2877\n",
      "2800 - 0.2050\n",
      "3000 - 0.1875\n",
      "3200 - 0.2751\n",
      "3400 - 0.2241\n",
      "3600 - 0.2868\n",
      "3800 - 0.2194\n",
      "4000 - 0.1518\n",
      "4200 - 0.2911\n",
      "4400 - 0.1658\n",
      "4600 - 0.2354\n",
      "4800 - 0.1664\n",
      "5000 - 0.1960\n",
      "5200 - 0.2201\n",
      "5400 - 0.1743\n",
      "5600 - 0.1775\n",
      "5800 - 0.1818\n",
      "6000 - 0.1848\n",
      "6200 - 0.1875\n",
      "6400 - 0.2487\n",
      "6600 - 0.1637\n",
      "6800 - 0.1560\n",
      "7000 - 0.2300\n",
      "7200 - 0.2163\n",
      "7400 - 0.2041\n",
      "7600 - 0.2136\n",
      "7800 - 0.2642\n",
      "8000 - 0.1647\n",
      "8200 - 0.2020\n",
      "8400 - 0.2004\n",
      "8600 - 0.1914\n",
      "8800 - 0.1626\n",
      "9000 - 0.2667\n",
      "9200 - 0.2073\n",
      "9400 - 0.2000\n",
      "9600 - 0.2161\n",
      "9800 - 0.1971\n",
      "10000 - 0.1574\n",
      "10200 - 0.2464\n",
      "10400 - 0.2022\n",
      "10600 - 0.1499\n",
      "10800 - 0.2182\n",
      "11000 - 0.3540\n",
      "11200 - 0.1996\n",
      "11400 - 0.1818\n",
      "11600 - 0.1049\n",
      "11800 - 0.2740\n",
      "12000 - 0.2245\n",
      "12200 - 0.2341\n",
      "12400 - 0.1377\n",
      "12600 - 0.1447\n",
      "12800 - 0.2296\n",
      "13000 - 0.1791\n",
      "13200 - 0.2110\n",
      "13400 - 0.1952\n",
      "13600 - 0.1500\n",
      "13800 - 0.1853\n",
      "14000 - 0.1668\n",
      "14200 - 0.1653\n",
      "14400 - 0.1680\n",
      "14600 - 0.1988\n",
      "14800 - 0.1987\n",
      "15000 - 0.1615\n",
      "15200 - 0.2825\n",
      "15400 - 0.1149\n",
      "15600 - 0.1889\n",
      "15800 - 0.1556\n",
      "16000 - 0.2652\n",
      "16200 - 0.1401\n",
      "16400 - 0.2231\n",
      "16600 - 0.1836\n",
      "16800 - 0.1629\n",
      "17000 - 0.2129\n",
      "17200 - 0.2621\n",
      "17400 - 0.2233\n",
      "17600 - 0.1802\n",
      "17800 - 0.2270\n",
      "18000 - 0.1792\n",
      "18200 - 0.2353\n",
      "18400 - 0.1171\n",
      "18600 - 0.2205\n",
      "18800 - 0.1565\n",
      "19000 - 0.1672\n",
      "19200 - 0.2069\n",
      "19400 - 0.2644\n",
      "19600 - 0.1681\n",
      "19800 - 0.2325\n",
      "20000 - 0.1530\n",
      "20200 - 0.2407\n",
      "20400 - 0.2741\n",
      "20600 - 0.2480\n",
      "20800 - 0.2142\n",
      "21000 - 0.1295\n",
      "21200 - 0.1934\n",
      "21400 - 0.1731\n",
      "21600 - 0.2164\n",
      "21800 - 0.1843\n",
      "22000 - 0.1531\n",
      "22200 - 0.1855\n",
      "22400 - 0.1584\n",
      "22600 - 0.1766\n",
      "22800 - 0.1554\n",
      "23000 - 0.2193\n",
      "23200 - 0.2339\n",
      "23400 - 0.1772\n",
      "23600 - 0.1131\n",
      "23800 - 0.1560\n",
      "24000 - 0.2287\n",
      "24200 - 0.2195\n",
      "24400 - 0.2032\n",
      "24600 - 0.1512\n",
      "24800 - 0.2349\n",
      "25000 - 0.1718\n",
      "25200 - 0.1081\n",
      "25400 - 0.1921\n",
      "25600 - 0.1807\n",
      "25800 - 0.2479\n",
      "26000 - 0.1629\n",
      "26200 - 0.1763\n",
      "26400 - 0.2311\n",
      "26600 - 0.1913\n",
      "26800 - 0.1590\n",
      "27000 - 0.1715\n",
      "27200 - 0.1783\n",
      "27400 - 0.1834\n",
      "27600 - 0.1292\n",
      "27800 - 0.2065\n",
      "28000 - 0.1650\n",
      "28200 - 0.1908\n",
      "28400 - 0.1704\n",
      "28600 - 0.1593\n",
      "28800 - 0.1667\n",
      "29000 - 0.2251\n",
      "29200 - 0.1477\n",
      "29400 - 0.1747\n",
      "29600 - 0.1405\n",
      "29800 - 0.1501\n",
      "30000 - 0.1632\n",
      "30200 - 0.1599\n",
      "30400 - 0.1851\n",
      "30600 - 0.1703\n",
      "30800 - 0.1813\n",
      "31000 - 0.1358\n",
      "31200 - 0.2055\n",
      "31400 - 0.1610\n",
      "31600 - 0.1820\n",
      "31800 - 0.1406\n",
      "32000 - 0.1875\n",
      "32200 - 0.1248\n",
      "32400 - 0.1438\n",
      "32600 - 0.1818\n",
      "32800 - 0.1519\n",
      "33000 - 0.2100\n",
      "33200 - 0.1495\n",
      "33400 - 0.1469\n",
      "33600 - 0.1104\n",
      "33800 - 0.1854\n",
      "34000 - 0.2031\n",
      "34200 - 0.1709\n",
      "34400 - 0.1973\n",
      "34600 - 0.2336\n",
      "34800 - 0.1801\n",
      "35000 - 0.1718\n",
      "35200 - 0.1202\n",
      "35400 - 0.1861\n",
      "35600 - 0.2324\n",
      "35800 - 0.1803\n",
      "36000 - 0.1146\n",
      "36200 - 0.2196\n",
      "36400 - 0.1268\n",
      "36600 - 0.1348\n",
      "36800 - 0.2197\n",
      "37000 - 0.1387\n",
      "37200 - 0.2407\n",
      "37400 - 0.1169\n",
      "37600 - 0.2161\n",
      "37800 - 0.2018\n",
      "38000 - 0.2483\n",
      "38200 - 0.2391\n",
      "38400 - 0.1715\n",
      "38600 - 0.2346\n",
      "38800 - 0.2036\n",
      "39000 - 0.1308\n",
      "39200 - 0.2080\n",
      "39400 - 0.1999\n",
      "39600 - 0.1290\n",
      "39800 - 0.1660\n",
      "40000 - 0.1868\n",
      "40200 - 0.1588\n",
      "40400 - 0.2928\n",
      "40600 - 0.1871\n",
      "40800 - 0.1758\n",
      "41000 - 0.1435\n",
      "41200 - 0.1988\n",
      "41400 - 0.2365\n",
      "41600 - 0.1948\n",
      "41800 - 0.1946\n",
      "42000 - 0.1604\n",
      "42200 - 0.1483\n",
      "42400 - 0.1648\n",
      "42600 - 0.2116\n",
      "42800 - 0.1524\n",
      "43000 - 0.0973\n",
      "43200 - 0.1838\n",
      "43400 - 0.1665\n",
      "43600 - 0.1684\n",
      "43800 - 0.1081\n",
      "44000 - 0.1806\n",
      "44200 - 0.1236\n",
      "44400 - 0.1032\n",
      "44600 - 0.1868\n",
      "44800 - 0.1720\n",
      "45000 - 0.1373\n",
      "45200 - 0.2180\n",
      "45400 - 0.2049\n",
      "45600 - 0.1376\n",
      "45800 - 0.1504\n",
      "46000 - 0.1831\n",
      "46200 - 0.1703\n",
      "46400 - 0.1630\n",
      "46600 - 0.1886\n",
      "46800 - 0.1438\n",
      "47000 - 0.1682\n",
      "47200 - 0.2384\n",
      "47400 - 0.2318\n",
      "47600 - 0.1423\n",
      "47800 - 0.1593\n",
      "48000 - 0.1806\n",
      "48200 - 0.1627\n",
      "48400 - 0.1510\n",
      "48600 - 0.1828\n",
      "48800 - 0.1550\n",
      "49000 - 0.1510\n",
      "49200 - 0.2110\n",
      "49400 - 0.1935\n",
      "49600 - 0.2275\n",
      "49800 - 0.2801\n"
     ]
    }
   ],
   "source": [
    "# for g in g_optimizer.param_groups:\n",
    "#     g['lr'] = 1e-4\n",
    "\n",
    "for _ in range(50000):\n",
    "    \n",
    "    g_optimizer.zero_grad()\n",
    "    \n",
    "    x_real = torch.Tensor(make_swiss_roll(batch_size_p, noise=0.1)[0][:, [0, 2]]/10).cuda()\n",
    "    x_real += 0.01 * torch.randn_like(x_real)\n",
    "    \n",
    "    z = torch.randn(batch_size_q, z_dim).cuda()\n",
    "    with contextlib.nullcontext():  # autograd.detect_anomaly():\n",
    "        x_fake = generator(z)\n",
    "\n",
    "#         x = x_real[:batch_size//2]\n",
    "#         x_prime = x_real[batch_size//2:]\n",
    "#         y = x_fake[:batch_size//2]\n",
    "#         y_prime = x_fake[batch_size//2:]\n",
    "\n",
    "#         loss = D(x, y) + D(x_prime, y) + D(x, y_prime) + D(x_prime, y_prime) - 2*D(y, y_prime)\n",
    "\n",
    "        loss = D(x_real, x_fake)\n",
    "#         loss = D(x_fake, x_real)\n",
    "\n",
    "        if _ % 200 == 0:\n",
    "            print(\"%d - %.4f\" % (_, loss.item()))\n",
    "\n",
    "        loss.backward()\n",
    "\n",
    "    g_optimizer.step()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 360x360 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "x_real = make_swiss_roll(500, noise=0.1)[0][:, [0, 2]]/10\n",
    "z = torch.randn(500, z_dim).cuda()\n",
    "x_fake = generator(z).data.cpu().numpy()\n",
    "\n",
    "plt.figure(figsize=(5, 5))\n",
    "plt.plot(x_real[:, 0], x_real[:, 1], 'g*', alpha=0.5, label='True', markersize=3)\n",
    "plt.plot(x_fake[:, 0], x_fake[:, 1], 'ro', alpha=0.5, label='Fake', markersize=3)\n",
    "plt.legend(loc='upper center', bbox_to_anchor=(0.5, -0.00), fancybox=True, shadow=True, ncol=2);\n",
    "\n",
    "cur_axes = plt.gca()\n",
    "cur_axes.axes.get_xaxis().set_visible(False)\n",
    "cur_axes.axes.get_yaxis().set_visible(False)\n",
    "\n",
    "cur_axes.spines['top'].set_visible(False)\n",
    "cur_axes.spines['right'].set_visible(False)\n",
    "cur_axes.spines['bottom'].set_visible(False)\n",
    "cur_axes.spines['left'].set_visible(False)\n",
    "\n",
    "# plt.show()\n",
    "plt.savefig('swiss_roll.pdf', dpi=100)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.7.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}