Dynamic Recurrent Neural Network using tf.nn.dynamic_rnn

dynamic_rnn.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import tensorflow as tf"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Input size\n",
    "nin = 100\n",
    "# Hidden layer num\n",
    "n_hidden = 256\n",
    "# Sequence max length\n",
    "seq_max_len = 30"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "# #batch_size * #max_seq_len * #inputsize\n",
    "input = tf.placeholder(tf.float32, [None, seq_max_len, nin])\n",
    "# #batch_size\n",
    "sequence_length = tf.placeholder(tf.int32, [None])\n",
    "\n",
    "batch_size = tf.shape(input)[0]\n",
    "\n",
    "cell = tf.nn.rnn_cell.GRUCell(n_hidden)\n",
    "initial_state = cell.zero_state(batch_size, dtype=tf.float32)\n",
    "\n",
    "output, state = tf.nn.dynamic_rnn(\n",
    "    cell, input, initial_state=initial_state,\n",
    "    sequence_length=sequence_length)\n",
    "\n",
    "#  Retrieve the last output\n",
    "index = tf.range(0, batch_size) * seq_max_len + (sequence_length - 1)\n",
    "output = tf.gather(tf.reshape(output, [-1, n_hidden]), index)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "# Sequence length\n",
    "seqlen = np.random.permutation(np.arange(1, 11))\n",
    "# #batch_size * variable_length * #inputsize\n",
    "x = [[np.random.randint(100, size=nin) / 100\n",
    "      for _ in range(l)] for l in seqlen]\n",
    "# 0 pad, #batch_size * #max_seq_len * #inputsize\n",
    "x = [_x + [[0.] * nin] * (seq_max_len - len(_x)) for _x in x]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[ 0.24374032 -0.26738089  0.19482653 ...,  0.06986387  0.63538063\n",
      "  -0.15972397]\n",
      " [ 0.02840432 -0.16684723  0.11450367 ..., -0.04713225  0.27185273\n",
      "  -0.09963451]\n",
      " [ 0.14622581 -0.19808987  0.29102004 ..., -0.01792343  0.49112913\n",
      "  -0.0518562 ]\n",
      " ..., \n",
      " [ 0.16123015 -0.3246842   0.17073596 ...,  0.12696315  0.40817925\n",
      "   0.05533985]\n",
      " [ 0.24779946 -0.25818256  0.24045026 ...,  0.08095235  0.45237231\n",
      "  -0.03921546]\n",
      " [ 0.13250643 -0.14197701  0.18299626 ...,  0.09092216  0.27430093\n",
      "  -0.12567034]]\n",
      "(10, 256)\n"
     ]
    }
   ],
   "source": [
    "init = tf.initialize_all_variables()\n",
    "\n",
    "with tf.Session() as sess:\n",
    "    sess.run(init)\n",
    "    out = sess.run(output, {input: x, sequence_length: seqlen})\n",
    "    print(out) # #batch_size * #n_hidden"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "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": "3.5.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}