multiclass classification using tensorflow

tf.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "from sklearn import datasets"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false,
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "iris = datasets.load_iris()\n",
    "num_labels = len(set(iris.target))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(150, 4) (150, 3)\n"
     ]
    }
   ],
   "source": [
    "data = iris.data.astype(np.float32)\n",
    "labels = (np.arange(num_labels) == np.array(iris.target)[:,None]).astype(np.float32)\n",
    "print(data.shape, labels.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def accuracy(predictions, labels):\n",
    "    return (100.0 * np.sum(np.argmax(predictions, 1) == np.argmax(labels, 1))\n",
    "            / predictions.shape[0])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "softmax regression:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "feature_size = data.shape[1]\n",
    "\n",
    "graph = tf.Graph()\n",
    "\n",
    "with graph.as_default():\n",
    "    tf_train_dataset = tf.constant(data)\n",
    "    tf_train_labels = tf.constant(labels)\n",
    "\n",
    "    weights = tf.Variable(tf.truncated_normal([feature_size, num_labels]))\n",
    "    biases = tf.Variable(tf.zeros([num_labels]))\n",
    "\n",
    "    logits = tf.matmul(tf_train_dataset, weights) + biases\n",
    "    loss = tf.reduce_mean(\n",
    "        tf.nn.softmax_cross_entropy_with_logits(logits, tf_train_labels))\n",
    "\n",
    "    optimizer = tf.train.AdamOptimizer(0.1).minimize(loss)\n",
    "    train_prediction = tf.nn.softmax(logits)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false,
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "step:0 loss:4.024884 accuracy: 66.67\n",
      "step:500 loss:0.072255 accuracy: 98.00\n",
      "step:1000 loss:0.054472 accuracy: 98.00\n",
      "step:1500 loss:0.048384 accuracy: 98.00\n",
      "step:2000 loss:0.045301 accuracy: 98.00\n",
      "step:2500 loss:0.043418 accuracy: 98.00\n",
      "step:3000 loss:0.042158 accuracy: 98.67\n",
      "step:3500 loss:0.041285 accuracy: 98.67\n",
      "step:4000 loss:0.040682 accuracy: 98.67\n",
      "step:4500 loss:0.040273 accuracy: 98.67\n",
      "step:5000 loss:0.040007 accuracy: 98.67\n",
      "step:5500 loss:0.039843 accuracy: 98.67\n",
      "step:6000 loss:0.039750 accuracy: 98.67\n",
      "step:6500 loss:0.039702 accuracy: 98.67\n",
      "step:7000 loss:0.039680 accuracy: 98.67\n",
      "step:7500 loss:0.039674 accuracy: 98.67\n",
      "step:8000 loss:0.039672 accuracy: 98.67\n",
      "step:8500 loss:0.039667 accuracy: 98.67\n",
      "step:9000 loss:0.039665 accuracy: 98.67\n",
      "step:9500 loss:0.039664 accuracy: 98.67\n",
      "step:10000 loss:0.039664 accuracy: 98.67\n"
     ]
    }
   ],
   "source": [
    "with tf.Session(graph=graph) as session:\n",
    "    tf.initialize_all_variables().run()\n",
    "    for step in range(10001):\n",
    "        _, l, predictions = session.run([optimizer, loss, train_prediction])\n",
    "        if step % 500 == 0:\n",
    "            print('step:{} loss:{:.6f} accuracy: {:.2f}'.format(\n",
    "                    step, l, accuracy(predictions, labels)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "linear multiclass SVM:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "feature_size = data.shape[1]\n",
    "delta = 1.0\n",
    "regulation_rate = 5e-4\n",
    "graph = tf.Graph()\n",
    "\n",
    "with graph.as_default():\n",
    "    tf_train_dataset = tf.constant(data)\n",
    "    tf_train_labels = tf.constant(labels)\n",
    "    \n",
    "    weights = tf.Variable(tf.truncated_normal([feature_size, num_labels]))\n",
    "    biases = tf.Variable(tf.zeros([num_labels]))\n",
    "\n",
    "    logits = tf.matmul(tf_train_dataset, weights) + biases\n",
    "    # TODO better way as numpy's: np.choose(data.target, logits.T)\n",
    "    y = tf.reduce_sum(logits * tf_train_labels, 1, keep_dims=True)\n",
    "    loss = tf.reduce_mean(tf.reduce_sum(tf.maximum(0.0, logits - y + delta), 1)) - delta\n",
    "    loss += regulation_rate * tf.nn.l2_loss(weights)\n",
    "\n",
    "    optimizer = tf.train.AdamOptimizer(0.1).minimize(loss)\n",
    "    train_prediction = tf.nn.softmax(logits)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "step:0 loss:7.959514 accuracy: 33.33\n",
      "step:500 loss:0.065034 accuracy: 97.33\n",
      "step:1000 loss:0.060172 accuracy: 98.00\n",
      "step:1500 loss:0.058802 accuracy: 98.00\n",
      "step:2000 loss:0.057681 accuracy: 98.00\n",
      "step:2500 loss:0.061486 accuracy: 98.00\n",
      "step:3000 loss:0.057971 accuracy: 98.00\n",
      "step:3500 loss:0.058048 accuracy: 98.00\n",
      "step:4000 loss:0.055821 accuracy: 98.00\n",
      "step:4500 loss:0.058429 accuracy: 98.00\n",
      "step:5000 loss:0.056925 accuracy: 98.00\n",
      "step:5500 loss:0.056828 accuracy: 98.00\n",
      "step:6000 loss:0.060888 accuracy: 98.00\n",
      "step:6500 loss:0.058099 accuracy: 98.00\n",
      "step:7000 loss:0.055918 accuracy: 98.00\n",
      "step:7500 loss:0.056860 accuracy: 98.67\n",
      "step:8000 loss:0.056827 accuracy: 98.00\n",
      "step:8500 loss:0.055365 accuracy: 98.00\n",
      "step:9000 loss:0.055880 accuracy: 98.00\n",
      "step:9500 loss:0.056300 accuracy: 98.67\n",
      "step:10000 loss:0.058815 accuracy: 98.00\n"
     ]
    }
   ],
   "source": [
    "with tf.Session(graph=graph) as session:\n",
    "    tf.initialize_all_variables().run()\n",
    "    for step in range(10001):\n",
    "        _, l, predictions = session.run([optimizer, loss, train_prediction])\n",
    "        if step % 500 == 0:\n",
    "            print('step:{} loss:{:.6f} accuracy: {:.2f}'.format(\n",
    "                    step, l, accuracy(predictions, labels)))"
   ]
  },
  {
   "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": "3.5.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}

I think softmax_cross_entropy_with_logits is not supposed for multi-class, it's just for non-one-hot label. Maybe sigmoid_cross_entropy_with_logits is the one.

Can you show how to generate prediction with this trained model?

Please show how to generate prediction with the trained model.