Quantile Loss.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/jacob/anaconda3/envs/kiromic/lib/python3.6/site-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.\n",
      "  from ._conv import register_converters as _register_converters\n"
     ]
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "%matplotlib inline\n",
    "\n",
    "# Initialize session\n",
    "sess = tf.Session()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load data\n",
    "mcycle = pd.read_csv('mcycle',delimiter='\\t')\n",
    "mcycle.times = (mcycle.times - mcycle.times.mean())/mcycle.times.std()\n",
    "mcycle.accel = (mcycle.accel - mcycle.accel.mean())/mcycle.accel.std()\n",
    "\n",
    "# Reshape to input format for network\n",
    "times = np.expand_dims(mcycle.times.values, 1)\n",
    "accel = np.expand_dims(mcycle.accel.values, 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Create network\n",
    "class q_model:\n",
    "    def __init__(self, \n",
    "                 sess, \n",
    "                 quantiles, \n",
    "                 in_shape=1, \n",
    "                 out_shape=1, \n",
    "                 batch_size=32):\n",
    "        \n",
    "        self.sess = sess\n",
    "        \n",
    "        self.quantiles = quantiles\n",
    "        self.num_quantiles = len(quantiles)\n",
    "        \n",
    "        self.in_shape = in_shape\n",
    "        self.out_shape = out_shape\n",
    "        self.batch_size = batch_size\n",
    "        \n",
    "        self.outputs = []\n",
    "        self.losses = []\n",
    "        self.loss_history = []\n",
    "        \n",
    "        self.build_model()\n",
    "        \n",
    "    def build_model(self, scope='q_model', reuse=tf.AUTO_REUSE): \n",
    "        with tf.variable_scope(scope, reuse=reuse) as scope:\n",
    "            self.x = tf.placeholder(tf.float32, shape=(None, self.in_shape))\n",
    "            self.y = tf.placeholder(tf.float32, shape=(None, self.out_shape))\n",
    "\n",
    "            self.layer0 = tf.layers.dense(self.x, \n",
    "                                    units=32, \n",
    "                                    activation=tf.nn.relu)\n",
    "            self.layer1 = tf.layers.dense(self.layer0, \n",
    "                                    units=32, \n",
    "                                    activation=tf.nn.relu)\n",
    "\n",
    "            # Create outputs and losses for all quantiles\n",
    "            for i in range(self.num_quantiles):\n",
    "                q = self.quantiles[i]\n",
    "                \n",
    "                # Get output layers \n",
    "                output = tf.layers.dense(self.layer1, 1, name=\"{}_q{}\".format(i, int(q*100)))\n",
    "                self.outputs.append(output)\n",
    "                \n",
    "                # Create losses\n",
    "                \n",
    "                error = tf.subtract(self.y, output)\n",
    "                loss = tf.reduce_mean(tf.maximum(q*error, (q-1)*error), axis=-1)\n",
    "\n",
    "                self.losses.append(loss)\n",
    "\n",
    "            # Create combined loss\n",
    "            self.combined_loss = tf.reduce_mean(tf.add_n(self.losses))\n",
    "            self.train_step = tf.train.AdamOptimizer().minimize(self.combined_loss)\n",
    "\n",
    "    def fit(self, x, y, epochs=100):  \n",
    "        for epoch in range(epochs):\n",
    "            epoch_losses = []\n",
    "            for idx in range(0, x.shape[0], self.batch_size):\n",
    "                batch_x = x[idx : min(idx + self.batch_size, x.shape[0]),:]\n",
    "                batch_y = y[idx : min(idx + self.batch_size, y.shape[0]),:]\n",
    "\n",
    "                feed_dict = {self.x: batch_x,\n",
    "                             self.y: batch_y}\n",
    "\n",
    "                _, c_loss = self.sess.run([self.train_step, self.combined_loss], feed_dict)\n",
    "                epoch_losses.append(c_loss)\n",
    "                \n",
    "            epoch_loss =  np.mean(epoch_losses)\n",
    "            self.loss_history.append(epoch_loss)\n",
    "            if epoch % 100 == 0:\n",
    "                print(\"Epoch {}: {}\".format(epoch, epoch_loss))\n",
    "            \n",
    "    def predict(self, x):   \n",
    "        # Run model to get outputs\n",
    "        feed_dict = {self.x: x}\n",
    "        predictions = sess.run(self.outputs, feed_dict)\n",
    "        \n",
    "        return predictions\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Instantiate model\n",
    "quantiles = [.1, .5, .9]\n",
    "model = q_model(sess, quantiles, batch_size=32)\n",
    "\n",
    "# Initialize all variables\n",
    "init_op = tf.global_variables_initializer()\n",
    "sess.run(init_op)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 0: 1.28651762008667\n",
      "Epoch 100: 0.566359281539917\n",
      "Epoch 200: 0.4066397249698639\n",
      "Epoch 300: 0.33803483843803406\n",
      "Epoch 400: 0.30299827456474304\n",
      "Epoch 500: 0.2755032181739807\n",
      "Epoch 600: 0.25843241810798645\n",
      "Epoch 700: 0.25172367691993713\n",
      "Epoch 800: 0.24871711432933807\n",
      "Epoch 900: 0.25110912322998047\n",
      "Epoch 1000: 0.2475467175245285\n",
      "Epoch 1100: 0.2458878606557846\n",
      "Epoch 1200: 0.24449209868907928\n",
      "Epoch 1300: 0.24454854428768158\n",
      "Epoch 1400: 0.24357259273529053\n",
      "Epoch 1500: 0.2434576004743576\n",
      "Epoch 1600: 0.24177248775959015\n",
      "Epoch 1700: 0.24332064390182495\n",
      "Epoch 1800: 0.2407195121049881\n",
      "Epoch 1900: 0.23958206176757812\n"
     ]
    }
   ],
   "source": [
    "# Run training\n",
    "epochs = 2000\n",
    "model.fit(times, accel, epochs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x116758630>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Generate the range of data we'd like to predict\n",
    "test_times = np.expand_dims(np.linspace(times.min(),times.max(),200), 1)\n",
    "predictions = model.predict(test_times)\n",
    "\n",
    "plt.scatter(times, accel)\n",
    "for i, prediction in enumerate(predictions):\n",
    "    plt.plot(test_times, prediction, label='{}th Quantile'.format(int(model.quantiles[i]*100)))\n",
    "    \n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<tf.Tensor 'q_model_5/0_q10/BiasAdd:0' shape=(?, 1) dtype=float32>,\n",
       " <tf.Tensor 'q_model_5/1_q50/BiasAdd:0' shape=(?, 1) dtype=float32>,\n",
       " <tf.Tensor 'q_model_5/2_q90/BiasAdd:0' shape=(?, 1) dtype=float32>]"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.outputs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "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": "3.6.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}