2-train.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Train"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This notebook trains a bunch of classifiers on the like/dislike playlists downloaded in the **Get Spotify Data** notebook. It saves the best one (highest accuracy) in the `/classifiers` folder."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from pkg.spotifyclassifier import *\n",
    "from pkg.run import *\n",
    "from pkg.wrangling import *\n",
    "from pkg.constants import LABEL_LIST\n",
    "from pkg.serialize import save_classifier"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Prepare Training Data"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The metadata, feature space, and genres are read in and stored in the \"songs\" dictionary by `track_id`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "default_song_data = FrozenMap(read_data('data/data.csv', True))\n",
    "print('Song data reading complete.')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "default_feature_names = tuple(next(iter(default_song_data.values()))['features'].keys())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Sanity check\n",
    "sanity_check(default_song_data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_cluster_size(default_song_data, 10)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Compute clusters from training data for algorithms that use them."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "clustered_song_data, songs_by_cluster = {}, {}\n",
    "clustered_song_data, songs_by_cluster = get_kmeans_clusters(default_song_data, NUM_CLUSTERS)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# sanity check\n",
    "if set(next(iter(default_song_data.values()))['features'].keys()) == set(next(iter(clustered_song_data.values()))['features'].keys()):\n",
    "    raise ValueError('Default features messed up.')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "default_training_data, default_validation_data, clustered_training_data, clustered_validation_data = get_experiment_split(default_song_data, clustered_song_data)\n",
    "training_clusters = get_train_clusters(clustered_training_data, songs_by_cluster[NUM_CLUSTERS])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Train Classifiers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "print('----\\nUnclustered\\n----')\n",
    "active_unclustered_results = run_active_suite(default_song_data, default_training_data, default_validation_data, SUPPORTED_ALGS, AL_STRATS)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "print('----\\nClustered\\n----')\n",
    "active_clustered_results = run_active_suite(default_song_data, clustered_training_data, clustered_validation_data, SUPPORTED_ALGS, AL_STRATS)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "print('\\n----\\nClustered w/ Cluster Sampling\\n----')\n",
    "active_cluster_sampled_results = run_clusters_suite(default_song_data, clustered_training_data, clustered_validation_data, training_clusters)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "best_classifier = get_highest_benchmark(\n",
    "    default_song_data,\n",
    "    default_training_data,\n",
    "    default_validation_data,\n",
    "    clustered_training_data,\n",
    "    clustered_validation_data\n",
    ")\n",
    "save_classifier(best_classifier, None) # specify filename instead of None if desired\n",
    "\n",
    "print('Classifier saved successfully.')"
   ]
  }
 ],
 "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": "3.7.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}