sampled_softmax.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import tensorflow as tf"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "config = tf.ConfigProto()\n",
    "config.log_device_placement = False\n",
    "config.gpu_options.allow_growth = True\n",
    "s = tf.InteractiveSession(config=config)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "input_size = 256\n",
    "vocab_size = 10**5\n",
    "batch_size = 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "x = tf.get_variable(\"logits\",initializer=tf.random_normal([batch_size,input_size]))\n",
    "y = tf.get_variable(\"answers\",initializer=tf.range(batch_size)[:,None])\n",
    "w = tf.get_variable(\"weights\",initializer=tf.ones((vocab_size,input_size)))\n",
    "b = tf.get_variable(\"bias\",initializer=tf.ones((vocab_size,)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "s.run(tf.initialize_all_variables())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### sampled logsoftmax"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "sampled_loss = tf.nn.sampled_softmax_loss(w,b,x,y,num_sampled=100,num_classes=vocab_size)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1000 loops, best of 3: 1.15 ms per loop\n"
     ]
    }
   ],
   "source": [
    "%timeit -n1000 s.run(sampled_loss)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### regular logsoftmax"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "logits = tf.matmul(x,tf.transpose(w)) +b\n",
    "regular_loss = tf.nn.softmax_cross_entropy_with_logits(logits,\n",
    "                                                       tf.one_hot(y,vocab_size))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1000 loops, best of 3: 18.4 ms per loop\n"
     ]
    }
   ],
   "source": [
    "%timeit -n1000 s.run(regular_loss)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python [Root]",
   "language": "python",
   "name": "Python [Root]"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}

slavic_softmax.ipynb