LightFm.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "8d959c73",
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import scipy.sparse as sp\n",
    "from lightfm.evaluation import precision_at_k\n",
    "from lightfm import LightFM\n",
    "\n",
    "def matrix_mappings(df, user, item):\n",
    "    \"\"\"Map entities in interactions df to row and column indices\n",
    "    Parameters\n",
    "    ----------\n",
    "    df : DataFrame Interactions DataFrame.\n",
    "    user : str User Column.\n",
    "    item : str Item Column.\n",
    "    Returns\n",
    "    -------\n",
    "    uid_to_idx : dict maps user ID's to index\n",
    "    iid_to_idx : dict maps item ID's to index\n",
    "\n",
    "    \"\"\"\n",
    "    \n",
    "    uid_to_idx = {}\n",
    "    for (idx, rid) in enumerate(df[user].unique().tolist()):\n",
    "        uid_to_idx[rid] = idx\n",
    "        \n",
    "\n",
    "    iid_to_idx = {}\n",
    "    for (idx, cid) in enumerate(df[item].unique().tolist()):\n",
    "        iid_to_idx[cid] = idx\n",
    "\n",
    "    return uid_to_idx, iid_to_idx\n",
    "\n",
    "\n",
    "def sparse_matrix(df, user, item):\n",
    "    \"\"\"Take interactions dataframe and convert to a sparse matrix\n",
    "    Parameters\n",
    "    ----------\n",
    "    df : DataFrame\n",
    "    user : str\n",
    "    item : str\n",
    "    Returns\n",
    "    -------\n",
    "    interactions : sparse csr matrix\n",
    "    uid_to_idx : dict\n",
    "    iid_to_idx : dict\n",
    "    \"\"\"\n",
    "\n",
    "    uid_to_idx, iid_to_idx = matrix_mappings(df,user,item)\n",
    "\n",
    "    def map_ids(row, mapper):\n",
    "        return mapper[row]\n",
    "\n",
    "    I = df[user].apply(map_ids, args=[uid_to_idx]).to_numpy()\n",
    "    J = df[item].apply(map_ids, args=[iid_to_idx]).to_numpy()\n",
    "    \n",
    "    V = np.ones(I.shape[0])\n",
    "    interactions = sp.coo_matrix((V, (I, J)), dtype=np.float64)\n",
    "    interactions = interactions.tocsr()\n",
    "    \n",
    "    return interactions, uid_to_idx, iid_to_idx\n",
    "\n",
    "def split_matrix(interactions, uid_to_idx, iid_to_idx, ratings, childset):\n",
    "    \n",
    "    #Creating a dataframe of interactions not present in child set but in parent set\n",
    "    diff = ratings.merge(childset, how = 'outer' ,indicator=True).loc[lambda x : x['_merge']=='left_only']\n",
    "    \n",
    "    sub_mat = interactions.copy().tolil()\n",
    "    \n",
    "    #Indicating zero for these interactions\n",
    "    for user,movie in zip(diff['userId'].values,diff['movieId'].values):\n",
    "        uidx = uid_to_idx[user]\n",
    "        iidx = iid_to_idx[movie]\n",
    "\n",
    "        sub_mat[uidx, iidx] = 0.\n",
    "    \n",
    "    return sub_mat.tocsr()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "86adc031",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.066614166\n",
      "0.024320988\n",
      "0.02\n"
     ]
    }
   ],
   "source": [
    "#Reading data\n",
    "col_names=['userId', 'movieId', 'rating', 'timestamp']\n",
    "    \n",
    "ratings = pd.read_csv('ml-latest-small/ratings.csv', names=col_names)\n",
    "\n",
    "#Reading splits\n",
    "train_df = pd.read_csv('train-split-large.csv')#'combined.csv')\n",
    "val_df = pd.read_csv('valid-split-large.csv')#'combined_val.csv')\n",
    "test_df = pd.read_csv('test-split-large.csv')#'combined_test.csv')\n",
    "\n",
    "train_df.reset_index(inplace=True)\n",
    "\n",
    "interaction_matrix, uid_to_idx, iid_to_idx = sparse_matrix(ratings[['userId','movieId']], 'userId', 'movieId')\n",
    "\n",
    "#Splitting matrices\n",
    "train = split_matrix(interaction_matrix, uid_to_idx, iid_to_idx, ratings[['userId','movieId']], train_df[['userId', 'movieId']])\n",
    "val = split_matrix(interaction_matrix, uid_to_idx, iid_to_idx, ratings[['userId','movieId']], val_df[['userId', 'movieId']])\n",
    "test = split_matrix(interaction_matrix, uid_to_idx, iid_to_idx, ratings[['userId','movieId']], test_df[['userId', 'movieId']])\n",
    "\n",
    "#Creating light fm model\n",
    "model = LightFM(loss='warp', random_state=2016)\n",
    "    \n",
    "model.fit(train, epochs=10)\n",
    "\n",
    "#Model Evaluation\n",
    "train_precision = precision_at_k(model, train, k=100).mean()\n",
    "val_precision = precision_at_k(model, val, k=100).mean()\n",
    "test_precision = precision_at_k(model, test, k=100).mean()\n",
    "    \n",
    "print(train_precision)\n",
    "print(val_precision)\n",
    "print(test_precision)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "96f2b2dd",
   "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": "3.8.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}