energy_prediction.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Predicting podcast audio energy using ALPR.\n",
    "\n",
    "This notebook shows how pretrained **A**dversarial **L**earned **P**odcast **R**epresentation (ALPR) can be used to predict the energy of audio snippets. The seriousness and popularity predictions can be conducted in a similar manner.\n",
    "\n",
    "### **Step 1:** load labels for training and testing classifiers."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import json\n",
    "\n",
    "# Load training labels\n",
    "train_pos_samples = []\n",
    "train_neg_samples = []\n",
    "with open('attributes_labels/energy_train.txt', 'r') as fin:\n",
    "    for line in fin:\n",
    "        snippet = json.loads(line)\n",
    "        if snippet['label'] == 1:\n",
    "            train_pos_samples.append([snippet['id'], snippet['offset']])\n",
    "        else:\n",
    "            train_neg_samples.append([snippet['id'], snippet['offset']])\n",
    "\n",
    "# Load testing labels\n",
    "test_pos_samples = []\n",
    "test_neg_samples = []\n",
    "with open('attributes_labels/energy_test.txt', 'r') as fin:\n",
    "    for line in fin:\n",
    "        snippet = json.loads(line)\n",
    "        if snippet['label'] == 1:\n",
    "            test_pos_samples.append([snippet['id'], snippet['offset']])\n",
    "        else:\n",
    "            test_neg_samples.append([snippet['id'], snippet['offset']])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### **Step 2:** load the pretrained ALPR model using the ALPRExtractor."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Restoring parameters from pretrained_model/alpr\n"
     ]
    }
   ],
   "source": [
    "from alpr_extractor import ALPRExtractor\n",
    "\n",
    "# Load ALPR extractor\n",
    "extractor = ALPRExtractor()\n",
    "extractor.load_model(path='pretrained_model/alpr')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### **Step 3:** extract ALPR for podcast audio spectrograms."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 270/270 [00:00<00:00, 2525.82it/s]\n",
      "100%|██████████| 270/270 [00:00<00:00, 3163.32it/s]\n",
      "100%|██████████| 270/270 [00:00<00:00, 3252.11it/s]\n",
      "100%|██████████| 270/270 [00:00<00:00, 3191.90it/s]\n",
      "100%|██████████| 17/17 [00:00<00:00, 17.33it/s]\n",
      "100%|██████████| 17/17 [00:00<00:00, 68.77it/s]\n",
      "100%|██████████| 17/17 [00:00<00:00, 69.10it/s]\n",
      "100%|██████████| 17/17 [00:00<00:00, 69.17it/s]\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "from tqdm import tqdm\n",
    "\n",
    "def get_spectrograms(samples):\n",
    "    samples_spectrograms = []\n",
    "    for s in tqdm(samples):\n",
    "        samples_spectrograms.append(np.load('data/attributes_prediction_spectrograms/e_%s_%d.npy' % (s[0], s[1]))\\\n",
    "                              .reshape((1, 128, 512, 1)))\n",
    "    samples_spectrograms = np.concatenate(samples_spectrograms, axis=0)\n",
    "    return (samples_spectrograms + 2.0) / 2\n",
    "\n",
    "def extract_features(samples_spectrograms, extractor=extractor, batch_size=16):\n",
    "    num_batches = int(1 + (len(samples_spectrograms) - 1) / batch_size)\n",
    "    features_output = []\n",
    "    for step in tqdm(range(num_batches)):\n",
    "        in_spectrograms = samples_spectrograms[step*batch_size:(step+1)*batch_size]\n",
    "        features_output.append(extractor.forward(in_spectrograms))\n",
    "    return np.concatenate(features_output, axis=0)\n",
    "\n",
    "train_pos_samples_spectrograms = get_spectrograms(train_pos_samples)\n",
    "train_neg_samples_spectrograms = get_spectrograms(train_neg_samples)\n",
    "test_pos_samples_spectrograms = get_spectrograms(test_pos_samples)\n",
    "test_neg_samples_spectrograms = get_spectrograms(test_neg_samples)\n",
    "\n",
    "train_pos_samples_features = extract_features(train_pos_samples_spectrograms)\n",
    "train_neg_samples_features = extract_features(train_neg_samples_spectrograms)\n",
    "test_pos_samples_features = extract_features(test_pos_samples_spectrograms)\n",
    "test_neg_samples_features = extract_features(test_neg_samples_spectrograms)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### **Step 4:** train and test classifiers, and visualize PR and AUC curves."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.8981481481481481\n",
      "AUC: 0.9500274348422497\n"
     ]
    }
   ],
   "source": [
    "from sklearn.linear_model import LogisticRegression as LR\n",
    "from sklearn.model_selection import cross_val_score\n",
    "from sklearn.metrics import roc_auc_score\n",
    "from sklearn.metrics import roc_curve\n",
    "from sklearn.metrics import precision_recall_curve\n",
    "\n",
    "train_features = np.concatenate((train_pos_samples_features, \n",
    "                           train_neg_samples_features), axis=0)\n",
    "train_labels = np.array([1]*len(train_pos_samples) + \\\n",
    "                        [0]*len(train_neg_samples))\n",
    "\n",
    "best_perform = 0\n",
    "best_C = 0.0001\n",
    "for C in [0.0001, 0.001, 0.01, 0.1, 1, 10, 100, 1000, 10000]:\n",
    "    lr_model = LR(C=C)\n",
    "    perform = np.mean(cross_val_score(lr_model, train_features, train_labels, cv=5))\n",
    "    if perform > best_perform:\n",
    "        best_perform = perform\n",
    "        best_C = C\n",
    "\n",
    "model = LR(C=best_C)\n",
    "model.fit(train_features, train_labels)\n",
    "\n",
    "test_features = np.concatenate((test_pos_samples_features, \n",
    "                                test_neg_samples_features), axis=0)\n",
    "test_labels = np.array([1]*len(test_pos_samples) + \\\n",
    "                       [0]*len(test_neg_samples))\n",
    "test_scores = model.decision_function(test_features)\n",
    "\n",
    "# Accuracy and AUC\n",
    "print('Accuracy:', model.score(test_features, test_labels))\n",
    "print('AUC:', roc_auc_score(test_labels, test_scores))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAWIAAAD1CAYAAABqSP81AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nWDzkuQAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 360x216 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAW0AAADnCAYAAADVXQcoAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nOeT/AJjV4abbd05JAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 360x216 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "\n",
    "# AUC curve and Precision-Recall curve\n",
    "test_precision, test_recall, _ = precision_recall_curve(y_true=test_labels, probas_pred=test_scores)\n",
    "test_fpr, test_tpr, _ = roc_curve(y_true=test_labels, y_score=test_scores)\n",
    "\n",
    "plt.figure(figsize=(5, 3))\n",
    "plt.plot(test_fpr, test_tpr, color='#c0392b', linewidth=2.5)\n",
    "plt.xlim([0, 1])\n",
    "plt.ylim([0, 1])\n",
    "plt.xticks([0, 0.5, 1], [0, 0.5, 1], fontsize=17)\n",
    "plt.yticks([0, 0.5, 1], [0, 0.5, 1], fontsize=17)\n",
    "plt.xlabel(\"False Postive Rate (FPR)\", fontsize=17)\n",
    "plt.ylabel(\"True Positive Rate (TPR)\", fontsize=17)\n",
    "plt.show()\n",
    "\n",
    "plt.figure(figsize=(5, 3))\n",
    "plt.plot(test_recall, test_precision, color='#c0392b', linewidth=2.5)\n",
    "plt.xlim([0, 1])\n",
    "plt.ylim([0.5, 1])\n",
    "plt.xticks([0, 0.5, 1], [0, 0.5, 1], fontsize=17)\n",
    "plt.yticks([0.5, 0.75, 1.0], [0.5, 0.75, 1], fontsize=17)\n",
    "plt.xlabel(\"Recall\", fontsize=17)\n",
    "plt.ylabel(\"Precision\", fontsize=17)\n",
    "plt.show()"
   ]
  }
 ],
 "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": "3.5.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}