recommendation.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# A recommendation engine for books\n",
    "\n",
    "## Preparation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "from lenskit import batch, topn, util\n",
    "from lenskit import crossfold as xf\n",
    "from lenskit.algorithms import als, funksvd\n",
    "from lenskit.algorithms.item_knn import ItemItem\n",
    "from lenskit.algorithms.user_knn import UserUser\n",
    "from lenskit.metrics.predict import rmse, mae\n",
    "from lenskit.algorithms import Recommender"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We start by importing our datasets:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "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>ISBN</th>\n",
       "      <th>Book-Title</th>\n",
       "      <th>Book-Author</th>\n",
       "      <th>Year-Of-Publication</th>\n",
       "      <th>Publisher</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0195153448</td>\n",
       "      <td>Classical Mythology</td>\n",
       "      <td>Mark P. O. Morford</td>\n",
       "      <td>2002</td>\n",
       "      <td>Oxford University Press</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0002005018</td>\n",
       "      <td>Clara Callan</td>\n",
       "      <td>Richard Bruce Wright</td>\n",
       "      <td>2001</td>\n",
       "      <td>HarperFlamingo Canada</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0060973129</td>\n",
       "      <td>Decision in Normandy</td>\n",
       "      <td>Carlo D'Este</td>\n",
       "      <td>1991</td>\n",
       "      <td>HarperPerennial</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0374157065</td>\n",
       "      <td>Flu: The Story of the Great Influenza Pandemic...</td>\n",
       "      <td>Gina Bari Kolata</td>\n",
       "      <td>1999</td>\n",
       "      <td>Farrar Straus Giroux</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0393045218</td>\n",
       "      <td>The Mummies of Urumchi</td>\n",
       "      <td>E. J. W. Barber</td>\n",
       "      <td>1999</td>\n",
       "      <td>W. W. Norton &amp;amp; Company</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>271355</th>\n",
       "      <td>0440400988</td>\n",
       "      <td>There's a Bat in Bunk Five</td>\n",
       "      <td>Paula Danziger</td>\n",
       "      <td>1988</td>\n",
       "      <td>Random House Childrens Pub (Mm)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>271356</th>\n",
       "      <td>0525447644</td>\n",
       "      <td>From One to One Hundred</td>\n",
       "      <td>Teri Sloat</td>\n",
       "      <td>1991</td>\n",
       "      <td>Dutton Books</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>271357</th>\n",
       "      <td>006008667X</td>\n",
       "      <td>Lily Dale : The True Story of the Town that Ta...</td>\n",
       "      <td>Christine Wicker</td>\n",
       "      <td>2004</td>\n",
       "      <td>HarperSanFrancisco</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>271358</th>\n",
       "      <td>0192126040</td>\n",
       "      <td>Republic (World's Classics)</td>\n",
       "      <td>Plato</td>\n",
       "      <td>1996</td>\n",
       "      <td>Oxford University Press</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>271359</th>\n",
       "      <td>0767409752</td>\n",
       "      <td>A Guided Tour of Rene Descartes' Meditations o...</td>\n",
       "      <td>Christopher  Biffle</td>\n",
       "      <td>2000</td>\n",
       "      <td>McGraw-Hill Humanities/Social Sciences/Languages</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>271353 rows × 5 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "              ISBN                                         Book-Title  \\\n",
       "0       0195153448                                Classical Mythology   \n",
       "1       0002005018                                       Clara Callan   \n",
       "2       0060973129                               Decision in Normandy   \n",
       "3       0374157065  Flu: The Story of the Great Influenza Pandemic...   \n",
       "4       0393045218                             The Mummies of Urumchi   \n",
       "...            ...                                                ...   \n",
       "271355  0440400988                         There's a Bat in Bunk Five   \n",
       "271356  0525447644                            From One to One Hundred   \n",
       "271357  006008667X  Lily Dale : The True Story of the Town that Ta...   \n",
       "271358  0192126040                        Republic (World's Classics)   \n",
       "271359  0767409752  A Guided Tour of Rene Descartes' Meditations o...   \n",
       "\n",
       "                 Book-Author Year-Of-Publication  \\\n",
       "0         Mark P. O. Morford                2002   \n",
       "1       Richard Bruce Wright                2001   \n",
       "2               Carlo D'Este                1991   \n",
       "3           Gina Bari Kolata                1999   \n",
       "4            E. J. W. Barber                1999   \n",
       "...                      ...                 ...   \n",
       "271355        Paula Danziger                1988   \n",
       "271356            Teri Sloat                1991   \n",
       "271357      Christine Wicker                2004   \n",
       "271358                 Plato                1996   \n",
       "271359   Christopher  Biffle                2000   \n",
       "\n",
       "                                               Publisher  \n",
       "0                                Oxford University Press  \n",
       "1                                  HarperFlamingo Canada  \n",
       "2                                        HarperPerennial  \n",
       "3                                   Farrar Straus Giroux  \n",
       "4                             W. W. Norton &amp; Company  \n",
       "...                                                  ...  \n",
       "271355                   Random House Childrens Pub (Mm)  \n",
       "271356                                      Dutton Books  \n",
       "271357                                HarperSanFrancisco  \n",
       "271358                           Oxford University Press  \n",
       "271359  McGraw-Hill Humanities/Social Sciences/Languages  \n",
       "\n",
       "[271353 rows x 5 columns]"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "books = pd.read_csv(\"data/Books.csv\", dtype={3: str})  # string column 3 also contains float to signify NaN, thus specify with dtype\n",
    "books = books.dropna()\n",
    "books = books.drop(columns=[\"Image-URL-S\", \"Image-URL-M\", \"Image-URL-L\"])\n",
    "books"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Index(['Age', 'city', 'state', 'country'], dtype='object')"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "users = pd.read_csv(\"data/Users.csv\", index_col='User-ID')\n",
    "\n",
    "# split location column in different components\n",
    "location_split = users['Location'].str.split(',', expand=True)\n",
    "\n",
    "if location_split.shape[1] < 3:\n",
    "    location_split[2] = np.nan\n",
    "if location_split.shape[1] > 3:\n",
    "    location_split = location_split.iloc[:, :3]\n",
    "    \n",
    "location_split.columns = ['city', 'state', 'country']\n",
    "users = users.drop(columns=['Location']).join(location_split)\n",
    "\n",
    "users.columns"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The final data set contains explicit user feedback from $0$ (worst) to $10$ (best):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "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>user</th>\n",
       "      <th>item</th>\n",
       "      <th>rating</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>276725</td>\n",
       "      <td>034545104X</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>276726</td>\n",
       "      <td>0155061224</td>\n",
       "      <td>5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>276727</td>\n",
       "      <td>0446520802</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>276729</td>\n",
       "      <td>052165615X</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>276729</td>\n",
       "      <td>0521795028</td>\n",
       "      <td>6</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1149775</th>\n",
       "      <td>276704</td>\n",
       "      <td>1563526298</td>\n",
       "      <td>9</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1149776</th>\n",
       "      <td>276706</td>\n",
       "      <td>0679447156</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1149777</th>\n",
       "      <td>276709</td>\n",
       "      <td>0515107662</td>\n",
       "      <td>10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1149778</th>\n",
       "      <td>276721</td>\n",
       "      <td>0590442449</td>\n",
       "      <td>10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1149779</th>\n",
       "      <td>276723</td>\n",
       "      <td>05162443314</td>\n",
       "      <td>8</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>1149780 rows × 3 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "           user         item  rating\n",
       "0        276725   034545104X       0\n",
       "1        276726   0155061224       5\n",
       "2        276727   0446520802       0\n",
       "3        276729   052165615X       3\n",
       "4        276729   0521795028       6\n",
       "...         ...          ...     ...\n",
       "1149775  276704   1563526298       9\n",
       "1149776  276706   0679447156       0\n",
       "1149777  276709   0515107662      10\n",
       "1149778  276721   0590442449      10\n",
       "1149779  276723  05162443314       8\n",
       "\n",
       "[1149780 rows x 3 columns]"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ratings = pd.read_csv(\"data/Ratings.csv\")\n",
    "ratings = ratings.rename(columns={'User-ID': 'user', 'ISBN': 'item', 'Book-Rating': 'rating'})  # expected format for LensKit library\n",
    "ratings\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1000x600 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "rating_counts = ratings['rating'].value_counts().sort_index()\n",
    "\n",
    "# Plot the bar graph\n",
    "plt.figure(figsize=(10, 6))\n",
    "rating_counts.plot(kind='bar', color='skyblue')\n",
    "plt.title('Number of Occurrences of Each Rating')\n",
    "plt.xlabel('Rating')\n",
    "plt.ylabel('Count')\n",
    "plt.xticks(rotation=0)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Recommendation Engines\n",
    "\n",
    "### Alternating Least Squares (ALS)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Cross-validation results:\n",
      "              RMSE       MAE\n",
      "AlgoName                    \n",
      "ALS       3.712356  2.897875\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<lenskit.algorithms.ranking.TopN at 0x75406b8633d0>"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Create the ALS algorithm\n",
    "als_algo = als.BiasedMF(features=20, iterations=20, reg=0.1, damping=5)\n",
    "\n",
    "# Create a full algorithm\n",
    "full_algo = Recommender.adapt(als_algo)\n",
    "\n",
    "# Define a function to evaluate the algorithm\n",
    "def eval_algo(aname, algo, train, test):\n",
    "    fittable = util.clone(algo)\n",
    "    fittable.fit(train)\n",
    "    \n",
    "    preds = batch.predict(fittable, test)\n",
    "    \n",
    "    merged = pd.merge(test, preds, on=['user', 'item'], how='left', suffixes=('_true', '_pred'))\n",
    "    merged = merged.dropna(subset=['prediction'])  # Ensure no missing predictions\n",
    "    \n",
    "    rmse_value = rmse(merged['rating_true'], merged['prediction'])\n",
    "    mae_value = mae(merged['rating_true'], merged['prediction'])\n",
    "\n",
    "    return pd.DataFrame({\n",
    "        'AlgoName': aname,\n",
    "        'RMSE': [rmse_value],\n",
    "        'MAE': [mae_value]\n",
    "    })\n",
    "\n",
    "# Perform cross-validation\n",
    "all_results = []\n",
    "for train, test in xf.partition_users(ratings, 5, xf.SampleFrac(0.2)):\n",
    "    results = eval_algo('ALS', full_algo, train, test)\n",
    "    all_results.append(results)\n",
    "\n",
    "# Combine all results\n",
    "all_results = pd.concat(all_results, ignore_index=True)\n",
    "\n",
    "# Print average results\n",
    "print(\"Cross-validation results:\")\n",
    "print(all_results.groupby('AlgoName').mean())\n",
    "\n",
    "# Train the algorithm on the full dataset\n",
    "full_algo.fit(ratings)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We observe that the mean absolute error in cross-validation is $< 3$, which is a nice result considering the algorithm only uses the rating dataframe. Let's use our model to make predictions: "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Top 10 recommendations for user 57725:\n",
      "ISBN: 1584855177, Title: Isabel: Taking Wing (Girls of Many Lands), Score: 14.30\n",
      "ISBN: 1591827191, Title: Eerie Queerie!, Score: 14.06\n",
      "ISBN: 0312956045, Title: Cannibal Killers, Score: 13.95\n",
      "ISBN: 081120118X, Title: Confessions of a Mask (New Directions Paperbook), Score: 13.58\n",
      "ISBN: 0340770724, Title: Title Not Found, Score: 13.03\n",
      "ISBN: 006097673X, Title: Autobiography of a Face, Score: 12.94\n",
      "ISBN: 1878685546, Title: Moon Shot : The Inside Story of America's Race to the Moon, Score: 12.93\n",
      "ISBN: 0151003130, Title: The Doubtful Guest, Score: 12.77\n",
      "ISBN: 0752826751, Title: Title Not Found, Score: 12.71\n",
      "ISBN: 0446611638, Title: Lost Light, Score: 12.61\n"
     ]
    }
   ],
   "source": [
    "# Function to get top-N recommendations for a user\n",
    "def get_top_n_recommendations(algo, user_id, n=10):\n",
    "    user_items = ratings[ratings['user'] == user_id]['item'].tolist()\n",
    "    recs = algo.recommend(user_id, n)\n",
    "    return [(isbn, score) for isbn, score in zip(recs['item'], recs['score'])]\n",
    "\n",
    "# Example: Get top 10 recommendations for user\n",
    "user_id = 57725\n",
    "recommendations = get_top_n_recommendations(full_algo, user_id, n=10)\n",
    "print(f\"Top 10 recommendations for user {user_id}:\")\n",
    "for isbn, score in recommendations:\n",
    "    # Check if the ISBN exists in the books DataFrame\n",
    "    if not books[books['ISBN'] == isbn].empty:\n",
    "        title = books[books['ISBN'] == isbn]['Book-Title'].values[0]\n",
    "    else:\n",
    "        title = \"Title Not Found\"\n",
    "    print(f\"ISBN: {isbn}, Title: {title}, Score: {score:.2f}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Top 10 similar books to ISBN 0971880107:\n",
      "ISBN: 0684873893, Title: The Blackwater Lightship : A Novel, Similarity: 5.17\n",
      "ISBN: 0553122509, Title: All Creatures Great and Small, Similarity: 5.10\n",
      "ISBN: 0307126854, Title: What a Bad Dream (A Golden Look-Look Book), Similarity: 4.87\n",
      "ISBN: 0679750789, Title: The Tightwad Gazette II: Promoting Thrift As a Viable Alternative Lifestyle, Similarity: 4.79\n",
      "ISBN: 0061097632, Title: The Eleventh Plague, Similarity: 4.76\n",
      "ISBN: 842266416X, Title: Title Not Found, Similarity: 4.73\n",
      "ISBN: 1583224890, Title: 9-11, Similarity: 4.69\n",
      "ISBN: 0373077351, Title: Bringing Benjy Home (Silhouette Intimate Moments, No 735), Similarity: 4.55\n",
      "ISBN: 0553569694, Title: The Killing of Monday Brown (A Phoebe Siegel Mystery), Similarity: 4.55\n",
      "ISBN: 8440699476, Title: El Catalejo Lacado (Pullman, Philip, His Dark Materials. 3.), Similarity: 4.48\n"
     ]
    }
   ],
   "source": [
    "# Function to get similar books\n",
    "def get_similar_books(als_algo, isbn, n=10):\n",
    "    # Check if the ISBN exists in the item index\n",
    "    if isbn not in als_algo.item_index_:\n",
    "        raise ValueError(f\"ISBN {isbn} not found in the trained model.\")\n",
    "    \n",
    "    # Get the index of the item (ISBN) in the item factors matrix\n",
    "    item_index = als_algo.item_index_.get_loc(isbn)\n",
    "    \n",
    "    # Get the factor vector for the specified book\n",
    "    item_factors = als_algo.item_features_\n",
    "    book_factor = item_factors[item_index]\n",
    "    \n",
    "    # Compute similarities with all other books\n",
    "    similarities = item_factors.dot(book_factor)\n",
    "    \n",
    "    # Get indices of the most similar books, excluding the book itself\n",
    "    similar_indices = np.argsort(similarities)[::-1][1:n+1]\n",
    "    \n",
    "    # Retrieve the ISBN and similarity score for each similar book\n",
    "    similar_books = [(als_algo.item_index_[i], similarities[i]) for i in similar_indices]\n",
    "    \n",
    "    return similar_books\n",
    "\n",
    "# Example: Get 10 similar books for a given ISBN\n",
    "isbn = \"0971880107\"\n",
    "similar_books = get_similar_books(als_algo, isbn, n=10)\n",
    "print(f\"\\nTop 10 similar books to ISBN {isbn}:\")\n",
    "\n",
    "for similar_isbn, similarity in similar_books:\n",
    "    # Check if the ISBN exists in the books DataFrame\n",
    "    if not books[books['ISBN'] == similar_isbn].empty:\n",
    "        title = books[books['ISBN'] == similar_isbn]['Book-Title'].values[0]\n",
    "    else:\n",
    "        title = \"Title Not Found\"\n",
    "    \n",
    "    print(f\"ISBN: {similar_isbn}, Title: {title}, Similarity: {similarity:.2f}\")\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### FunkSVD"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Cross-validation results:\n",
      "              RMSE      MAE\n",
      "AlgoName                   \n",
      "FunkSVD   3.620984  2.85036\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<lenskit.algorithms.ranking.TopN at 0x75403fa7ccd0>"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "svd_algo = funksvd.FunkSVD(features=20, iterations=100, lrate=0.001, reg=0.015, damping=5)\n",
    "full_algo_alt = Recommender.adapt(svd_algo)\n",
    "\n",
    "# Perform cross-validation\n",
    "all_results = []\n",
    "for train, test in xf.partition_users(ratings, 5, xf.SampleFrac(0.2)):\n",
    "    results = eval_algo('FunkSVD', full_algo_alt, train, test)\n",
    "    all_results.append(results)\n",
    "\n",
    "# Combine all results\n",
    "all_results = pd.concat(all_results, ignore_index=True)\n",
    "\n",
    "# Print average results\n",
    "print(\"Cross-validation results:\")\n",
    "print(all_results.groupby('AlgoName').mean())\n",
    "\n",
    "# Train the algorithm on the full dataset\n",
    "full_algo_alt.fit(ratings)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The mean absolute error in cross-validation is around the same as for ALS."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Top 10 recommendations for user 57725:\n",
      "ISBN: 0142001430, Title: Year of Wonders, Score: 11.27\n",
      "ISBN: 0316769509, Title: Nine Stories, Score: 10.87\n",
      "ISBN: 0060248025, Title: Falling Up, Score: 10.09\n",
      "ISBN: 0898159954, Title: FAT!SO? : Because You Don't Have to Apologize for Your Size, Score: 9.61\n",
      "ISBN: 0553577123, Title: The Diary of a Young Girl: The Definitive Edition, Score: 9.57\n",
      "ISBN: 0140430725, Title: Pride and Prejudice (The Penguin English Library, El 72), Score: 9.57\n",
      "ISBN: 0385199570, Title: The Stand (The Complete and Uncut Edition), Score: 9.53\n",
      "ISBN: 0804106304, Title: The Joy Luck Club, Score: 9.32\n",
      "ISBN: 0786808551, Title: The Arctic Incident (Artemis Fowl, Book 2), Score: 9.25\n",
      "ISBN: 0380432811, Title: Bridge to Terabithia, Score: 9.23\n"
     ]
    }
   ],
   "source": [
    "# Example: Get top 10 recommendations for user\n",
    "user_id = 57725\n",
    "recommendations = get_top_n_recommendations(full_algo_alt, user_id, n=10)\n",
    "print(f\"Top 10 recommendations for user {user_id}:\")\n",
    "for isbn, score in recommendations:\n",
    "    # Check if the ISBN exists in the books DataFrame\n",
    "    if not books[books['ISBN'] == isbn].empty:\n",
    "        title = books[books['ISBN'] == isbn]['Book-Title'].values[0]\n",
    "    else:\n",
    "        title = \"Title Not Found\"\n",
    "    print(f\"ISBN: {isbn}, Title: {title}, Score: {score:.2f}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Top 10 similar books to ISBN 0971880107:\n",
      "ISBN: 0971880107, Title: Wild Animus, Similarity: 6.99\n",
      "ISBN: 0385504209, Title: The Da Vinci Code, Similarity: 6.75\n",
      "ISBN: 0786868716, Title: The Five People You Meet in Heaven, Similarity: 5.86\n",
      "ISBN: 0345423402, Title: A Kiss of Shadows (Meredith Gentry Novels (Paperback)), Similarity: 5.51\n",
      "ISBN: 0142001740, Title: The Secret Life of Bees, Similarity: 4.97\n",
      "ISBN: 0439404371, Title: The Thief Lord, Similarity: 4.96\n",
      "ISBN: 0767902521, Title: A Walk in the Woods: Rediscovering America on the Appalachian Trail (Official Guides to the Appalachian Trail), Similarity: 4.79\n",
      "ISBN: 067102423X, Title: Bag of Bones, Similarity: 4.54\n",
      "ISBN: 0446532231, Title: Dude, Where's My Country?, Similarity: 4.46\n",
      "ISBN: 0802139256, Title: Peace Like a River, Similarity: 4.21\n"
     ]
    }
   ],
   "source": [
    "# Example: Get 10 similar books for a given ISBN\n",
    "isbn = \"0971880107\"\n",
    "similar_books = get_similar_books(svd_algo, isbn, n=10)\n",
    "print(f\"\\nTop 10 similar books to ISBN {isbn}:\")\n",
    "\n",
    "for similar_isbn, similarity in similar_books:\n",
    "    # Check if the ISBN exists in the books DataFrame\n",
    "    if not books[books['ISBN'] == similar_isbn].empty:\n",
    "        title = books[books['ISBN'] == similar_isbn]['Book-Title'].values[0]\n",
    "    else:\n",
    "        title = \"Title Not Found\"\n",
    "    \n",
    "    print(f\"ISBN: {similar_isbn}, Title: {title}, Similarity: {similarity:.2f}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### k-Nearest Neighbour"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/mia/.local/lib/python3.10/site-packages/lenskit/algorithms/item_knn.py:119: NumbaTypeSafetyWarning: \u001b[1m\u001b[1m\u001b[1munsafe cast from uint64 to int64. Precision may be lost.\u001b[0m\u001b[0m\u001b[0m\n",
      "  b = blocks[bi]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Cross-validation results:\n",
      "              RMSE       MAE\n",
      "AlgoName                    \n",
      "ItemItem  3.689338  2.845022\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<lenskit.algorithms.ranking.TopN at 0x7540241f6c20>"
      ]
     },
     "execution_count": 43,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Create the ItemItem KNN algorithm\n",
    "knn_algo = ItemItem(nnbrs=20, min_nbrs=1, min_sim=1e-06, feedback='explicit')\n",
    "\n",
    "# Create a full algorithm\n",
    "full_knn_algo = Recommender.adapt(knn_algo)\n",
    "\n",
    "# Perform cross-validation\n",
    "all_results = []\n",
    "for train, test in xf.partition_users(ratings, 5, xf.SampleFrac(0.2)):\n",
    "    results = eval_algo('ItemItem', full_knn_algo, train, test)\n",
    "    all_results.append(results)\n",
    "\n",
    "# Combine all results\n",
    "all_results = pd.concat(all_results, ignore_index=True)\n",
    "\n",
    "# Print average results\n",
    "print(\"Cross-validation results:\")\n",
    "print(all_results.groupby('AlgoName').mean())\n",
    "\n",
    "# Train the algorithm on the full dataset\n",
    "full_knn_algo.fit(ratings)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Cross-validation results:\n",
      "              RMSE       MAE\n",
      "AlgoName                    \n",
      "ItemItem  4.001583  2.999032\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<lenskit.algorithms.ranking.TopN at 0x754055a5d570>"
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Create the ItemItem KNN algorithm\n",
    "knn_algo = UserUser(nnbrs=20, min_nbrs=1, min_sim=1e-06, feedback='explicit')\n",
    "\n",
    "# Create a full algorithm\n",
    "full_knn_algo = Recommender.adapt(knn_algo)\n",
    "\n",
    "# Perform cross-validation\n",
    "all_results = []\n",
    "for train, test in xf.partition_users(ratings, 5, xf.SampleFrac(0.2)):\n",
    "    results = eval_algo('ItemItem', full_knn_algo, train, test)\n",
    "    all_results.append(results)\n",
    "\n",
    "# Combine all results\n",
    "all_results = pd.concat(all_results, ignore_index=True)\n",
    "\n",
    "# Print average results\n",
    "print(\"Cross-validation results:\")\n",
    "print(all_results.groupby('AlgoName').mean())\n",
    "\n",
    "# Train the algorithm on the full dataset\n",
    "full_knn_algo.fit(ratings)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "So $k$-Nearest Neighbours leads to a similar mean average, with User-User performing worse."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Top 10 recommendations for user 57725:\n",
      "ISBN: 0385319290, Title: Grave Misgivings: A Mother Lavinia Grey Mystery (Mother Lavinia Grey Mysteries), Score: 12.57\n",
      "ISBN: 0375505008, Title: Franklin and Winston: An Intimate Portrait of an Epic Friendship, Score: 12.46\n",
      "ISBN: 0312117787, Title: Dog Talk: Training Your Dog Through a Canine Point of View, Score: 12.43\n",
      "ISBN: 0340666269, Title: Dog Who Loved Too Much Tales Treatments, Score: 12.43\n",
      "ISBN: 0486241785, Title: Elizabeth Zimmermann's Knitters' Almanac, Score: 12.42\n",
      "ISBN: 0517884518, Title: Woof! : The Funny and Fabulous Trials and Tribulations of 25 Years as a Dog Trainer, Score: 12.42\n",
      "ISBN: 0786410116, Title: Graves' Disease: A Practical Guide, Score: 12.42\n",
      "ISBN: 0882400851, Title: Alaska Sourdough, Score: 12.42\n",
      "ISBN: 1580626408, Title: The Everything Vegetarian Cookbook: 300 Healthy Recipes Everyone Will Enjoy (Everything Series), Score: 12.42\n",
      "ISBN: 0316108537, Title: Goodnight Opus, Score: 12.37\n"
     ]
    }
   ],
   "source": [
    "# Example: Get top 10 recommendations for user\n",
    "user_id = 57725\n",
    "recommendations = full_knn_algo.recommend(user_id, n=10)\n",
    "\n",
    "print(f\"Top 10 recommendations for user {user_id}:\")\n",
    "for _, rec in recommendations.iterrows():\n",
    "    isbn = rec['item']\n",
    "    score = rec['score']\n",
    "    # Check if the ISBN exists in the books DataFrame\n",
    "    if not books[books['ISBN'] == isbn].empty:\n",
    "        title = books[books['ISBN'] == isbn]['Book-Title'].values[0]\n",
    "    else:\n",
    "        title = \"Title Not Found\"\n",
    "    print(f\"ISBN: {isbn}, Title: {title}, Score: {score:.2f}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "(c) Mia Müßig"
   ]
  }
 ],
 "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": "3.10.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}