CF PSO Echo Nest.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "8648138d",
   "metadata": {},
   "outputs": [],
   "source": [
    "# importing libraries\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import pyswarms as ps\n",
    "\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "from sklearn.metrics.pairwise import cosine_similarity\n",
    "from sklearn.cluster import KMeans\n",
    "from sklearn.metrics import silhouette_score\n",
    "\n",
    "from surprise import Dataset, Reader\n",
    "from surprise import SVD, NMF, KNNWithZScore, CoClustering\n",
    "from surprise.model_selection import ShuffleSplit\n",
    "from surprise.model_selection import train_test_split\n",
    "from sklearn.metrics.pairwise import cosine_similarity\n",
    "from surprise import accuracy\n",
    "\n",
    "from collections import defaultdict\n",
    "from statistics import mean\n",
    "import random"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "74359858",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Train Triplets Data\n",
    "train_triplets = 'K:/Datasets/EchoNest/Train Triplets/train_triplets.txt'\n",
    "train_ratings = pd.read_csv(train_triplets, sep=\"\\t\", header=None)\n",
    "train_ratings.columns = ['user_id', 'song_id', 'listen_count']\n",
    "\n",
    "# Spotify Metadata \n",
    "song_metadata = pd.read_csv('K:/Notebook Files/spotify_metadata.csv')\n",
    "\n",
    "# get ratings for songs that are in the metadata\n",
    "ratings = train_ratings[train_ratings['song_id'].isin(song_metadata['en_song_id'])]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "6c07c54a",
   "metadata": {},
   "outputs": [],
   "source": [
    "# how many songs each user has listened to\n",
    "user_counts = ratings.groupby('user_id')['song_id'].count()\n",
    "user_counts.describe().apply(lambda x: format(x, 'f'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "96eff1cb",
   "metadata": {},
   "outputs": [],
   "source": [
    "# how many users listen to the same song on average\n",
    "song_user = ratings.groupby('song_id')['user_id'].count()\n",
    "song_user.describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "dae75cd8",
   "metadata": {},
   "outputs": [],
   "source": [
    "# filter out users that have listened to less than 16 songs\n",
    "flt_users = user_counts[user_counts > 15].index.to_list() #25%\n",
    "\n",
    "# filter out songs that have less than 200 users\n",
    "flt_songs = song_user[song_user > 14].index.to_list() #50%\n",
    "\n",
    "# filter out dataset with user and song id's\n",
    "ratings_flt = ratings[(ratings['user_id'].isin(flt_users)) & (ratings['song_id'].isin(flt_songs))].reset_index(drop=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f461b1a5",
   "metadata": {},
   "outputs": [],
   "source": [
    "# top 10000 songs with most ratings\n",
    "top_songs = song_user.sort_values(ascending=False)[:10000].index.to_list()\n",
    "# filter for only top 10000 popular songs\n",
    "ratings_flt_pop = ratings_flt[(ratings_flt['song_id'].isin(top_songs))].reset_index(drop=True)\n",
    "\n",
    "# top 10000 users with most ratings\n",
    "most_ratings_user = user_counts.sort_values(ascending=False)[:10000].index.to_list()\n",
    "# filter for only top 10000 users\n",
    "ratings_flt_pop = ratings_flt_pop[(ratings_flt_pop['user_id'].isin(most_ratings_user))].reset_index(drop=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "cfd9cdf2",
   "metadata": {},
   "outputs": [],
   "source": [
    "# binning technique \n",
    "bins = [0,1,2,3,4,5,6,7,8,9,ratings_flt_pop['listen_count'].max()]\n",
    "ratings_flt_pop['rating'] = pd.cut(ratings_flt_pop['listen_count'], bins=bins, labels=[1,2,3,4,5,6,7,8,9,10])\n",
    "ratings_flt_pop['rating'] = ratings_flt_pop.rating.astype('int')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "3496585d",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Initialize Reader class with rating scale from 1 to 10\n",
    "reader = Reader(rating_scale=(1, 10))\n",
    "\n",
    "# load dataset class with ratings \n",
    "data = Dataset.load_from_df(ratings_flt_pop[['user_id', 'song_id', 'rating']], reader)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "1e12207f",
   "metadata": {},
   "outputs": [],
   "source": [
    "# code from suprise library website\n",
    "# source: https://surprise.readthedocs.io/en/stable/FAQ.html\n",
    "\n",
    "# Return precision and recall at k metrics for each user\n",
    "def precision_recall_at_k(predictions, k=20, threshold=3):\n",
    "    \n",
    "    # First map the predictions to each user.\n",
    "    user_est_true = defaultdict(list)\n",
    "    for uid, _, true_r, est, _ in predictions:\n",
    "        user_est_true[uid].append((est, true_r))\n",
    "    \n",
    "    precisions = dict()\n",
    "    recalls = dict()\n",
    "    for uid, user_ratings in user_est_true.items():\n",
    "        \n",
    "        # Sort user ratings by estimated value\n",
    "        user_ratings.sort(key=lambda x: x[0], reverse=True)\n",
    "        # Number of relevant items\n",
    "        n_rel = sum((true_r >= threshold) for (_, true_r) in user_ratings)\n",
    "        # Number of recommended items in top k\n",
    "        n_rec_k = sum((est >= threshold) for (est, _) in user_ratings[:k])\n",
    "        # Number of relevant and recommended items in top k\n",
    "        n_rel_and_rec_k = sum(((true_r >= threshold) and (est >= threshold)) for (est, true_r) in user_ratings[:k])\n",
    "        # Precision@K: Proportion of recommended items that are relevant\n",
    "        precisions[uid] = n_rel_and_rec_k / n_rec_k if n_rec_k != 0 else 1\n",
    "        # Recall@K: Proportion of relevant items that are recommended\n",
    "        recalls[uid] = n_rel_and_rec_k / n_rel if n_rel != 0 else 1\n",
    "    \n",
    "    return precisions, recalls"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "b2e30957",
   "metadata": {},
   "outputs": [],
   "source": [
    "# open-source code by https://github.com/mikema2019/tune-hyperparameters-with-PSO\n",
    "\n",
    "class optimisation():\n",
    "    def __init__(self, n_particles=10):\n",
    "        self.n_particles = n_particles\n",
    "        \n",
    "    def PSO(self, train_set, test_set, algo, parameters):\n",
    "        \n",
    "        self.train_set = train_set\n",
    "        self.test_set = test_set\n",
    "        self.algo = algo\n",
    "        self.parameters = parameters\n",
    "        self.dimensions = len(parameters)\n",
    "        \n",
    "        # The Minimum and Maximum Dimension Boundary\n",
    "        maximum = np.zeros(self.dimensions)\n",
    "        minimum = np.zeros(self.dimensions)\n",
    "        \n",
    "        for count, (key, value) in enumerate(self.parameters.items()):\n",
    "            minimum[count] = value[1]\n",
    "            maximum[count] = value[2]\n",
    "        boundaries = (minimum, maximum)\n",
    "        \n",
    "        # parameters\n",
    "        options = {'c1':0.5, \n",
    "                   'c2':0.6, \n",
    "                   'w':0.8}\n",
    "        \n",
    "        # the GlobalBestPSO method finds the best solution from a set of condidate solutions\n",
    "        optimizer = ps.single.GlobalBestPSO(n_particles=self.n_particles, dimensions=self.dimensions, \n",
    "                                            options=options, bounds=boundaries)\n",
    "        \n",
    "        # omptimize method return tuple with optimal cost and position after optimizing \n",
    "        best_cost, best_pos = optimizer.optimize(self.searchParameters, iters=10)\n",
    "        self.BestParams = {}\n",
    "\n",
    "        # using the best found position for particle, train and test model with best hyper-parameters\n",
    "        for count, (key, value) in enumerate(self.parameters.items()):\n",
    "            if value[0].__name__ == 'choice':\n",
    "                index = value[0](best_pos[count])\n",
    "                self.BestParams[key] = value[3][index]\n",
    "            else:\n",
    "                self.BestParams[key] = value[0](best_pos[count])\n",
    "        print(\"\\n\")\n",
    "        self.final_model = self.algo(**self.BestParams)\n",
    "        self.final_model.fit(train_set)\n",
    "        print(f\"PSO selected hyper-parameters: {self.BestParams}\")\n",
    "        \n",
    "        # testing model and returning predictions \n",
    "        predictions = self.final_model.test(test_set)\n",
    "        result = accuracy.rmse(predictions)\n",
    "        print(result)\n",
    "        \n",
    "        return predictions\n",
    "\n",
    "    # objective function that finds the best combinations of hyper-parmaters\n",
    "    def searchParameters(self, parameters):\n",
    "        score_array = np.zeros((self.n_particles))\n",
    "        SearchParams = {}\n",
    "        \n",
    "        for i in range(self.n_particles):\n",
    "            for count, (key, value) in enumerate(self.parameters.items()):\n",
    "                if value[0].__name__ == 'choice':\n",
    "                    index = value[0](parameters[i,count])\n",
    "                    SearchParams[key] = value[3][index]\n",
    "                else:\n",
    "                    SearchParams[key] = value[0](parameters[i,count])\n",
    "                \n",
    "                algorithm = self.algo(**SearchParams)\n",
    "                algorithm.fit(self.train_set) \n",
    "\n",
    "                predictions = algorithm.test(self.test_set)\n",
    "                score_array[i] = accuracy.rmse(predictions)\n",
    "            \n",
    "            # get average of RMSE score and return\n",
    "            print(f'AVG: {-np.mean(score_array)}')\n",
    "            return -np.mean(score_array)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "798fd3d6",
   "metadata": {},
   "outputs": [],
   "source": [
    "algos = [SVD, NMF, KNNWithZScore, CoClustering]\n",
    "test_size = ['0.25', '0.30', '0.40']\n",
    "n_splits = 5\n",
    "\n",
    "results = {}\n",
    "\n",
    "parameters = defaultdict(dict)\n",
    "parameters['SVD'] = {\n",
    "    'n_factors':[choice, 0, 6, [20, 30, 50, 75, 100, 125]],\n",
    "    'lr_all':[uniform, 0.001, 0.002, 0.003, 0.05],\n",
    "    'reg_all':[uniform, 0.002, 0.004, 0.006, 0.08]\n",
    "}\n",
    "\n",
    "parameters['NMF'] = {\n",
    "    'n_factors':[choice, 0, 6, [25, 30, 50, 75, 100, 125]],\n",
    "    'reg_pu': [uniform, 0.001, 0.003, 0.005, 0.01],\n",
    "    'reg_qi':[uniform, 0.001, 0.003, 0.005, 0.01]\n",
    "}\n",
    "\n",
    "parameters['KNNWithZScore'] = {\n",
    "    'k':[choice, 0, 5, [5, 10, 15, 20, 30]],\n",
    "    'min_k':[choice, 0, 3, [2, 4, 6]],\n",
    "}\n",
    "\n",
    "parameters['CoClustering'] = {\n",
    "    'n_cltr_u':[choice, 0, 4, [3, 5, 10, 15, 20]],\n",
    "    'n_cltr_i': [choice, 0, 4, [3, 5, 10, 15, 20]]\n",
    "}\n",
    "\n",
    "for algorithm in algos:\n",
    "    print(f'Current algorithm: {algorithm}')\n",
    "    \n",
    "    algorithm_name = str(algorithm).split(\".\")[3].split(\" \")[0].split(\"'\")[0]\n",
    "    \n",
    "    for size in test_size:\n",
    "        # split the dataset into a training set and testing set\n",
    "        trainset, testset = train_test_split(data, test_size=float(size))\n",
    "        \n",
    "        # load the ParticleSwarmOptimisation class and call the PSO method\n",
    "        pso = optimisation()\n",
    "        pso.PSO(trainset, testset, algorithm, parameters[algorithm_name])"
   ]
  }
 ],
 "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": "3.8.11"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}