MLP_feed.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Trains and Evaluates the IndianPines network using a feed dictionary\n",
    "========================================"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "from __future__ import absolute_import\n",
    "from __future__ import division\n",
    "from __future__ import print_function\n",
    "\n",
    "import time\n",
    "import numpy as np\n",
    "from six.moves import xrange  # pylint: disable=redefined-builtin\n",
    "import tensorflow as tf\n",
    "import os\n",
    "import IndianPinesMLP \n",
    "import patch_size\n",
    "# import IndianPines_data_set as input_data\n",
    "import Spatial_dataset as input_data"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Declare model parameters as external flags\n",
    "-------------------------------------"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "flags = tf.app.flags\n",
    "FLAGS = flags.FLAGS\n",
    "flags.DEFINE_float('learning_rate', 0.01, 'Initial learning rate.')\n",
    "flags.DEFINE_integer('max_steps', 50000, 'Number of steps to run trainer.')\n",
    "flags.DEFINE_integer('hidden1', 500, 'Number of units in hidden layer 1.')\n",
    "flags.DEFINE_integer('hidden2', 350, 'Number of units in hidden layer 2.')\n",
    "flags.DEFINE_integer('hidden3', 150, 'Number of units in hidden layer 3.')\n",
    "flags.DEFINE_integer('batch_size', 200, 'Batch size.  '\n",
    "                     'Must divide evenly into the dataset sizes.')\n",
    "# flags.DEFINE_string('train_dir', '1.mat', 'Directory to put the training data.')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "learning_rate = 0.01\n",
    "num_epochs = 20\n",
    "max_steps = 50000\n",
    "IMAGE_SIZE = patch_size.patch_size\n",
    "fc1 = 500\n",
    "fc2 = 350\n",
    "fc3 = 150\n",
    "batch_size = 200\n",
    "TRAIN_FILES = 8\n",
    "TEST_FILES = 4\n",
    "DATA_PATH = os.path.join(os.getcwd(),\"Data\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def placeholder_inputs(batch_size):\n",
    "    \"\"\"Generate placeholder variables to represent the input tensors.\n",
    "    These placeholders are used as inputs by the rest of the model building\n",
    "    code and will be fed from the downloaded data in the .run() loop, below.\n",
    "    Args:\n",
    "    batch_size: The batch size will be baked into both placeholders.\n",
    "    Returns:\n",
    "    images_placeholder: Images placeholder.\n",
    "    labels_placeholder: Labels placeholder.\n",
    "    \"\"\"\n",
    "    # Note that the shapes of the placeholders match the shapes of the full\n",
    "    # image and label tensors, except the first dimension is now batch_size\n",
    "    # rather than the full size of the train or test data sets.\n",
    "    images_placeholder = tf.placeholder(tf.float32, shape=(batch_size, IndianPinesMLP\n",
    "                                                           .IMAGE_PIXELS))\n",
    "    labels_placeholder = tf.placeholder(tf.int32, shape=(batch_size))\n",
    "    return images_placeholder, labels_placeholder\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def fill_feed_dict(data_set, images_pl, labels_pl):\n",
    "    \"\"\"Fills the feed_dict for training the given step.\n",
    "    A feed_dict takes the form of:\n",
    "    feed_dict = {\n",
    "      <placeholder>: <tensor of values to be passed for placeholder>,\n",
    "      ....\n",
    "    }\n",
    "    Args:\n",
    "    data_set: The set of images and labels, from input_data.read_data_sets()\n",
    "    images_pl: The images placeholder, from placeholder_inputs().\n",
    "    labels_pl: The labels placeholder, from placeholder_inputs().\n",
    "    Returns:\n",
    "    feed_dict: The feed dictionary mapping from placeholders to values.\n",
    "    \"\"\"\n",
    "    # Create the feed_dict for the placeholders filled with the next\n",
    "    # `batch size ` examples.\n",
    "    images_feed, labels_feed = data_set.next_batch(batch_size)\n",
    "    feed_dict = {\n",
    "      images_pl: images_feed,\n",
    "      labels_pl: labels_feed,\n",
    "    }\n",
    "    return feed_dict"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "def do_eval(sess,\n",
    "            eval_correct,\n",
    "            images_placeholder,\n",
    "            labels_placeholder,\n",
    "            data_set):\n",
    "    \"\"\"Runs one evaluation against the full epoch of data.\n",
    "    Args:\n",
    "    sess: The session in which the model has been trained.\n",
    "    eval_correct: The Tensor that returns the number of correct predictions.\n",
    "    images_placeholder: The images placeholder.\n",
    "    labels_placeholder: The labels placeholder.\n",
    "    data_set: The set of images and labels to evaluate, from\n",
    "      input_data.read_data_sets().\n",
    "    \"\"\"\n",
    "    # And run one epoch of eval.\n",
    "    true_count = 0  # Counts the number of correct predictions.\n",
    "    steps_per_epoch = data_set.num_examples // batch_size\n",
    "    num_examples = steps_per_epoch * batch_size\n",
    "    for step in xrange(steps_per_epoch):\n",
    "        feed_dict = fill_feed_dict(data_set,\n",
    "                                   images_placeholder,\n",
    "                                   labels_placeholder)\n",
    "        true_count += sess.run(eval_correct, feed_dict=feed_dict)\n",
    "    precision = true_count / num_examples\n",
    "    print('  Num examples: %d  Num correct: %d  Precision @ 1: %0.04f' %\n",
    "        (num_examples, true_count, precision))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "def add_DataSet(first,second):\n",
    "    temp_image = np.concatenate((first.images,second.images),axis=0)\n",
    "    temp_labels = np.concatenate((first.labels,second.labels),axis=0)\n",
    "    temp_image = temp_image.reshape(temp_image.shape[0],IMAGE_SIZE,IMAGE_SIZE,220)\n",
    "    temp_image = np.transpose(temp_image,(0,3,1,2))\n",
    "    temp_labels = np.transpose(temp_labels)\n",
    "    return input_data.DataSet(temp_image,temp_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "def run_training():\n",
    "    \"\"\"Train MNIST for a number of steps.\"\"\"\n",
    "    # Get the sets of images and labels for training, validation, and\n",
    "    # test on IndianPines.\n",
    "    \n",
    "    \"\"\"Concatenating all the training and test mat files\"\"\"\n",
    "    for i in range(TRAIN_FILES):\n",
    "        data_sets = input_data.read_data_sets(os.path.join(DATA_PATH, 'Train_'+str(IMAGE_SIZE)+'_'+str(i+1)+'.mat'), 'train')\n",
    "        if(i==0):\n",
    "            Training_data = data_sets\n",
    "            continue\n",
    "        else:\n",
    "            Training_data = add_DataSet(Training_data,data_sets)\n",
    "            \n",
    "    for i in range(TEST_FILES):\n",
    "        data_sets = input_data.read_data_sets(os.path.join(DATA_PATH, 'Test_'+str(IMAGE_SIZE)+'_'+str(i+1)+'.mat'),'test')\n",
    "        if(i==0):\n",
    "            Test_data = data_sets\n",
    "            continue\n",
    "        else:\n",
    "            Test_data = add_DataSet(Test_data,data_sets)\n",
    "        \n",
    "    # Tell TensorFlow that the model will be built into the default Graph.\n",
    "    with tf.Graph().as_default():\n",
    "    # Generate placeholders for the images and labels.\n",
    "        images_placeholder, labels_placeholder = placeholder_inputs(FLAGS.batch_size)\n",
    "\n",
    "        # Build a Graph that computes predictions from the inference model.\n",
    "        logits = IndianPinesMLP.inference(images_placeholder,\n",
    "                                 FLAGS.hidden1,\n",
    "                                 FLAGS.hidden2,\n",
    "                                 FLAGS.hidden3)\n",
    "\n",
    "        # Add to the Graph the Ops for loss calculation.\n",
    "        loss = IndianPinesMLP.loss(logits, labels_placeholder)\n",
    "\n",
    "        # Add to the Graph the Ops that calculate and apply gradients.\n",
    "        train_op = IndianPinesMLP.training(loss, FLAGS.learning_rate)\n",
    "\n",
    "        # Add the Op to compare the logits to the labels during evaluation.\n",
    "        eval_correct = IndianPinesMLP.evaluation(logits, labels_placeholder)\n",
    "\n",
    "        # Build the summary operation based on the TF collection of Summaries.\n",
    "    #    summary_op = tf.merge_all_summaries()\n",
    "\n",
    "        # Add the variable initializer Op.\n",
    "        init = tf.initialize_all_variables()\n",
    "\n",
    "        # Create a saver for writing training checkpoints.\n",
    "        saver = tf.train.Saver()\n",
    "\n",
    "        # Create a session for running Ops on the Graph.\n",
    "        sess = tf.Session()\n",
    "\n",
    "        # Instantiate a SummaryWriter to output summaries and the Graph.\n",
    "    #    summary_writer = tf.train.SummaryWriter(FLAGS.train_dir, sess.graph)\n",
    "\n",
    "        # And then after everything is built:\n",
    "\n",
    "        # Run the Op to initialize the variables.\n",
    "        sess.run(init)\n",
    "\n",
    "        # Start the training loop.\n",
    "        for step in xrange(FLAGS.max_steps):\n",
    "            start_time = time.time()\n",
    "\n",
    "            # Fill a feed dictionary with the actual set of images and labels\n",
    "            # for this particular training step.\n",
    "            feed_dict = fill_feed_dict(Training_data,\n",
    "                                     images_placeholder,\n",
    "                                     labels_placeholder)\n",
    "\n",
    "            # Run one step of the model.  The return values are the activations\n",
    "            # from the `train_op` (which is discarded) and the `loss` Op.  To\n",
    "            # inspect the values of your Ops or variables, you may include them\n",
    "            # in the list passed to sess.run() and the value tensors will be\n",
    "            # returned in the tuple from the call.\n",
    "            _, loss_value = sess.run([train_op, loss],\n",
    "                                   feed_dict=feed_dict)\n",
    "\n",
    "            duration = time.time() - start_time\n",
    "\n",
    "            # Write the summaries and print an overview fairly often.\n",
    "            if step % 50 == 0:\n",
    "            # Print status to stdout.\n",
    "                print('Step %d: loss = %.2f (%.3f sec)' % (step, loss_value, duration))\n",
    "            # Update the events file.\n",
    "    #             summary_str = sess.run(summary_op, feed_dict=feed_dict)\n",
    "    #             summary_writer.add_summary(summary_str, step)\n",
    "    #             summary_writer.flush()\n",
    "\n",
    "            # Save a checkpoint and evaluate the model periodically.\n",
    "            if (step + 1) % 1000 == 0 or (step + 1) == FLAGS.max_steps:\n",
    "                saver.save(sess, 'model-MLP-'+str(IMAGE_SIZE)+'X'+str(IMAGE_SIZE)+'.ckpt', global_step=step)\n",
    "\n",
    "            # Evaluate against the training set.\n",
    "                print('Training Data Eval:')\n",
    "                do_eval(sess,\n",
    "                        eval_correct,\n",
    "                        images_placeholder,\n",
    "                        labels_placeholder,\n",
    "                        Training_data)\n",
    "                print('Test Data Eval:')\n",
    "                do_eval(sess,\n",
    "                        eval_correct,\n",
    "                        images_placeholder,\n",
    "                        labels_placeholder,\n",
    "                        Test_data)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tensor(\"Reshape:0\", shape=(200, 220), dtype=float32)\n",
      "Step 0: loss = 2.77 (0.172 sec)\n",
      "Step 50: loss = 2.76 (0.002 sec)\n",
      "Step 100: loss = 2.75 (0.003 sec)\n",
      "Step 150: loss = 2.74 (0.002 sec)\n",
      "Step 200: loss = 2.73 (0.002 sec)\n",
      "Step 250: loss = 2.70 (0.002 sec)\n",
      "Step 300: loss = 2.66 (0.002 sec)\n",
      "Step 350: loss = 2.59 (0.003 sec)\n",
      "Step 400: loss = 2.52 (0.004 sec)\n",
      "Step 450: loss = 2.46 (0.002 sec)\n",
      "Step 500: loss = 2.34 (0.002 sec)\n",
      "Step 550: loss = 2.22 (0.002 sec)\n",
      "Step 600: loss = 2.11 (0.002 sec)\n",
      "Step 650: loss = 2.02 (0.003 sec)\n",
      "Step 700: loss = 1.99 (0.002 sec)\n",
      "Step 750: loss = 1.85 (0.002 sec)\n",
      "Step 800: loss = 1.82 (0.004 sec)\n",
      "Step 850: loss = 1.69 (0.003 sec)\n",
      "Step 900: loss = 1.72 (0.002 sec)\n",
      "Step 950: loss = 1.55 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 1655  Precision @ 1: 0.5172\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 507  Precision @ 1: 0.3169\n",
      "Step 1000: loss = 1.56 (0.002 sec)\n",
      "Step 1050: loss = 1.43 (0.002 sec)\n",
      "Step 1100: loss = 1.43 (0.002 sec)\n",
      "Step 1150: loss = 1.44 (0.002 sec)\n",
      "Step 1200: loss = 1.27 (0.004 sec)\n",
      "Step 1250: loss = 1.33 (0.002 sec)\n",
      "Step 1300: loss = 1.24 (0.002 sec)\n",
      "Step 1350: loss = 1.24 (0.002 sec)\n",
      "Step 1400: loss = 1.17 (0.002 sec)\n",
      "Step 1450: loss = 1.41 (0.002 sec)\n",
      "Step 1500: loss = 1.15 (0.002 sec)\n",
      "Step 1550: loss = 1.19 (0.002 sec)\n",
      "Step 1600: loss = 1.12 (0.004 sec)\n",
      "Step 1650: loss = 1.27 (0.002 sec)\n",
      "Step 1700: loss = 1.09 (0.002 sec)\n",
      "Step 1750: loss = 1.25 (0.002 sec)\n",
      "Step 1800: loss = 1.11 (0.002 sec)\n",
      "Step 1850: loss = 1.15 (0.002 sec)\n",
      "Step 1900: loss = 1.05 (0.002 sec)\n",
      "Step 1950: loss = 1.07 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 1904  Precision @ 1: 0.5950\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 706  Precision @ 1: 0.4412\n",
      "Step 2000: loss = 1.12 (0.007 sec)\n",
      "Step 2050: loss = 1.07 (0.002 sec)\n",
      "Step 2100: loss = 1.12 (0.002 sec)\n",
      "Step 2150: loss = 1.10 (0.002 sec)\n",
      "Step 2200: loss = 1.13 (0.002 sec)\n",
      "Step 2250: loss = 1.24 (0.002 sec)\n",
      "Step 2300: loss = 1.15 (0.002 sec)\n",
      "Step 2350: loss = 1.18 (0.002 sec)\n",
      "Step 2400: loss = 0.94 (0.004 sec)\n",
      "Step 2450: loss = 1.03 (0.002 sec)\n",
      "Step 2500: loss = 0.99 (0.002 sec)\n",
      "Step 2550: loss = 1.03 (0.002 sec)\n",
      "Step 2600: loss = 1.03 (0.002 sec)\n",
      "Step 2650: loss = 1.08 (0.002 sec)\n",
      "Step 2700: loss = 0.96 (0.002 sec)\n",
      "Step 2750: loss = 0.96 (0.002 sec)\n",
      "Step 2800: loss = 1.03 (0.007 sec)\n",
      "Step 2850: loss = 1.06 (0.002 sec)\n",
      "Step 2900: loss = 1.06 (0.002 sec)\n",
      "Step 2950: loss = 1.12 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2033  Precision @ 1: 0.6353\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 793  Precision @ 1: 0.4956\n",
      "Step 3000: loss = 0.94 (0.003 sec)\n",
      "Step 3050: loss = 1.12 (0.003 sec)\n",
      "Step 3100: loss = 0.92 (0.002 sec)\n",
      "Step 3150: loss = 0.91 (0.002 sec)\n",
      "Step 3200: loss = 0.99 (0.003 sec)\n",
      "Step 3250: loss = 1.08 (0.002 sec)\n",
      "Step 3300: loss = 0.93 (0.002 sec)\n",
      "Step 3350: loss = 1.01 (0.002 sec)\n",
      "Step 3400: loss = 1.02 (0.002 sec)\n",
      "Step 3450: loss = 1.00 (0.003 sec)\n",
      "Step 3500: loss = 0.84 (0.002 sec)\n",
      "Step 3550: loss = 0.99 (0.002 sec)\n",
      "Step 3600: loss = 0.97 (0.004 sec)\n",
      "Step 3650: loss = 0.92 (0.002 sec)\n",
      "Step 3700: loss = 1.05 (0.002 sec)\n",
      "Step 3750: loss = 1.00 (0.002 sec)\n",
      "Step 3800: loss = 1.00 (0.002 sec)\n",
      "Step 3850: loss = 0.86 (0.002 sec)\n",
      "Step 3900: loss = 1.02 (0.002 sec)\n",
      "Step 3950: loss = 0.80 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2158  Precision @ 1: 0.6744\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 860  Precision @ 1: 0.5375\n",
      "Step 4000: loss = 0.98 (0.005 sec)\n",
      "Step 4050: loss = 0.84 (0.003 sec)\n",
      "Step 4100: loss = 0.83 (0.002 sec)\n",
      "Step 4150: loss = 0.86 (0.002 sec)\n",
      "Step 4200: loss = 0.86 (0.002 sec)\n",
      "Step 4250: loss = 0.81 (0.002 sec)\n",
      "Step 4300: loss = 0.84 (0.002 sec)\n",
      "Step 4350: loss = 1.04 (0.002 sec)\n",
      "Step 4400: loss = 1.00 (0.004 sec)\n",
      "Step 4450: loss = 0.86 (0.002 sec)\n",
      "Step 4500: loss = 1.00 (0.002 sec)\n",
      "Step 4550: loss = 0.81 (0.002 sec)\n",
      "Step 4600: loss = 0.92 (0.002 sec)\n",
      "Step 4650: loss = 0.86 (0.002 sec)\n",
      "Step 4700: loss = 0.91 (0.002 sec)\n",
      "Step 4750: loss = 0.91 (0.002 sec)\n",
      "Step 4800: loss = 0.97 (0.004 sec)\n",
      "Step 4850: loss = 0.84 (0.002 sec)\n",
      "Step 4900: loss = 0.85 (0.002 sec)\n",
      "Step 4950: loss = 0.84 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2182  Precision @ 1: 0.6819\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 882  Precision @ 1: 0.5513\n",
      "Step 5000: loss = 0.82 (0.003 sec)\n",
      "Step 5050: loss = 0.90 (0.002 sec)\n",
      "Step 5100: loss = 0.79 (0.002 sec)\n",
      "Step 5150: loss = 0.89 (0.002 sec)\n",
      "Step 5200: loss = 0.67 (0.004 sec)\n",
      "Step 5250: loss = 0.81 (0.002 sec)\n",
      "Step 5300: loss = 0.77 (0.003 sec)\n",
      "Step 5350: loss = 0.76 (0.002 sec)\n",
      "Step 5400: loss = 0.76 (0.004 sec)\n",
      "Step 5450: loss = 0.80 (0.002 sec)\n",
      "Step 5500: loss = 0.82 (0.002 sec)\n",
      "Step 5550: loss = 0.78 (0.002 sec)\n",
      "Step 5600: loss = 0.85 (0.004 sec)\n",
      "Step 5650: loss = 0.78 (0.002 sec)\n",
      "Step 5700: loss = 0.77 (0.002 sec)\n",
      "Step 5750: loss = 0.81 (0.002 sec)\n",
      "Step 5800: loss = 0.78 (0.002 sec)\n",
      "Step 5850: loss = 0.70 (0.002 sec)\n",
      "Step 5900: loss = 0.79 (0.002 sec)\n",
      "Step 5950: loss = 0.72 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2308  Precision @ 1: 0.7212\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 949  Precision @ 1: 0.5931\n",
      "Step 6000: loss = 0.79 (0.005 sec)\n",
      "Step 6050: loss = 0.85 (0.002 sec)\n",
      "Step 6100: loss = 0.69 (0.002 sec)\n",
      "Step 6150: loss = 0.76 (0.002 sec)\n",
      "Step 6200: loss = 0.68 (0.002 sec)\n",
      "Step 6250: loss = 0.85 (0.002 sec)\n",
      "Step 6300: loss = 0.79 (0.002 sec)\n",
      "Step 6350: loss = 0.76 (0.002 sec)\n",
      "Step 6400: loss = 0.77 (0.004 sec)\n",
      "Step 6450: loss = 0.72 (0.003 sec)\n",
      "Step 6500: loss = 0.74 (0.002 sec)\n",
      "Step 6550: loss = 0.68 (0.002 sec)\n",
      "Step 6600: loss = 0.69 (0.002 sec)\n",
      "Step 6650: loss = 0.77 (0.002 sec)\n",
      "Step 6700: loss = 0.74 (0.002 sec)\n",
      "Step 6750: loss = 0.71 (0.002 sec)\n",
      "Step 6800: loss = 0.64 (0.003 sec)\n",
      "Step 6850: loss = 0.80 (0.002 sec)\n",
      "Step 6900: loss = 0.61 (0.002 sec)\n",
      "Step 6950: loss = 0.65 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2369  Precision @ 1: 0.7403\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 973  Precision @ 1: 0.6081\n",
      "Step 7000: loss = 0.65 (0.003 sec)\n",
      "Step 7050: loss = 0.82 (0.002 sec)\n",
      "Step 7100: loss = 0.63 (0.002 sec)\n",
      "Step 7150: loss = 0.75 (0.002 sec)\n",
      "Step 7200: loss = 0.76 (0.003 sec)\n",
      "Step 7250: loss = 0.79 (0.002 sec)\n",
      "Step 7300: loss = 0.66 (0.002 sec)\n",
      "Step 7350: loss = 0.70 (0.002 sec)\n",
      "Step 7400: loss = 0.62 (0.002 sec)\n",
      "Step 7450: loss = 0.65 (0.002 sec)\n",
      "Step 7500: loss = 0.60 (0.003 sec)\n",
      "Step 7550: loss = 0.71 (0.003 sec)\n",
      "Step 7600: loss = 0.62 (0.004 sec)\n",
      "Step 7650: loss = 0.67 (0.002 sec)\n",
      "Step 7700: loss = 0.61 (0.002 sec)\n",
      "Step 7750: loss = 0.60 (0.002 sec)\n",
      "Step 7800: loss = 0.61 (0.002 sec)\n",
      "Step 7850: loss = 0.66 (0.002 sec)\n",
      "Step 7900: loss = 0.75 (0.002 sec)\n",
      "Step 7950: loss = 0.83 (0.003 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2436  Precision @ 1: 0.7612\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 985  Precision @ 1: 0.6156\n",
      "Step 8000: loss = 0.64 (0.005 sec)\n",
      "Step 8050: loss = 0.63 (0.002 sec)\n",
      "Step 8100: loss = 0.69 (0.002 sec)\n",
      "Step 8150: loss = 0.78 (0.002 sec)\n",
      "Step 8200: loss = 0.60 (0.002 sec)\n",
      "Step 8250: loss = 0.60 (0.002 sec)\n",
      "Step 8300: loss = 0.63 (0.002 sec)\n",
      "Step 8350: loss = 0.61 (0.002 sec)\n",
      "Step 8400: loss = 0.64 (0.004 sec)\n",
      "Step 8450: loss = 0.63 (0.002 sec)\n",
      "Step 8500: loss = 0.71 (0.002 sec)\n",
      "Step 8550: loss = 0.58 (0.002 sec)\n",
      "Step 8600: loss = 0.67 (0.002 sec)\n",
      "Step 8650: loss = 0.55 (0.002 sec)\n",
      "Step 8700: loss = 0.55 (0.002 sec)\n",
      "Step 8750: loss = 0.66 (0.002 sec)\n",
      "Step 8800: loss = 0.61 (0.004 sec)\n",
      "Step 8850: loss = 0.59 (0.002 sec)\n",
      "Step 8900: loss = 0.72 (0.002 sec)\n",
      "Step 8950: loss = 0.63 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2486  Precision @ 1: 0.7769\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1005  Precision @ 1: 0.6281\n",
      "Step 9000: loss = 0.54 (0.002 sec)\n",
      "Step 9050: loss = 0.62 (0.003 sec)\n",
      "Step 9100: loss = 0.57 (0.003 sec)\n",
      "Step 9150: loss = 0.69 (0.002 sec)\n",
      "Step 9200: loss = 0.65 (0.003 sec)\n",
      "Step 9250: loss = 0.65 (0.002 sec)\n",
      "Step 9300: loss = 0.59 (0.002 sec)\n",
      "Step 9350: loss = 0.58 (0.002 sec)\n",
      "Step 9400: loss = 0.64 (0.002 sec)\n",
      "Step 9450: loss = 0.61 (0.003 sec)\n",
      "Step 9500: loss = 0.63 (0.002 sec)\n",
      "Step 9550: loss = 0.63 (0.002 sec)\n",
      "Step 9600: loss = 0.60 (0.004 sec)\n",
      "Step 9650: loss = 0.55 (0.002 sec)\n",
      "Step 9700: loss = 0.66 (0.002 sec)\n",
      "Step 9750: loss = 0.56 (0.002 sec)\n",
      "Step 9800: loss = 0.56 (0.002 sec)\n",
      "Step 9850: loss = 0.54 (0.002 sec)\n",
      "Step 9900: loss = 0.61 (0.003 sec)\n",
      "Step 9950: loss = 0.50 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2544  Precision @ 1: 0.7950\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1018  Precision @ 1: 0.6362\n",
      "Step 10000: loss = 0.58 (0.006 sec)\n",
      "Step 10050: loss = 0.65 (0.002 sec)\n",
      "Step 10100: loss = 0.54 (0.002 sec)\n",
      "Step 10150: loss = 0.54 (0.002 sec)\n",
      "Step 10200: loss = 0.61 (0.003 sec)\n",
      "Step 10250: loss = 0.62 (0.002 sec)\n",
      "Step 10300: loss = 0.57 (0.002 sec)\n",
      "Step 10350: loss = 0.68 (0.002 sec)\n",
      "Step 10400: loss = 0.53 (0.004 sec)\n",
      "Step 10450: loss = 0.60 (0.002 sec)\n",
      "Step 10500: loss = 0.63 (0.002 sec)\n",
      "Step 10550: loss = 0.52 (0.002 sec)\n",
      "Step 10600: loss = 0.48 (0.002 sec)\n",
      "Step 10650: loss = 0.57 (0.002 sec)\n",
      "Step 10700: loss = 0.53 (0.002 sec)\n",
      "Step 10750: loss = 0.58 (0.002 sec)\n",
      "Step 10800: loss = 0.57 (0.003 sec)\n",
      "Step 10850: loss = 0.46 (0.002 sec)\n",
      "Step 10900: loss = 0.55 (0.003 sec)\n",
      "Step 10950: loss = 0.54 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2579  Precision @ 1: 0.8059\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1049  Precision @ 1: 0.6556\n",
      "Step 11000: loss = 0.49 (0.002 sec)\n",
      "Step 11050: loss = 0.66 (0.002 sec)\n",
      "Step 11100: loss = 0.50 (0.002 sec)\n",
      "Step 11150: loss = 0.63 (0.002 sec)\n",
      "Step 11200: loss = 0.52 (0.004 sec)\n",
      "Step 11250: loss = 0.52 (0.002 sec)\n",
      "Step 11300: loss = 0.45 (0.002 sec)\n",
      "Step 11350: loss = 0.52 (0.002 sec)\n",
      "Step 11400: loss = 0.59 (0.002 sec)\n",
      "Step 11450: loss = 0.51 (0.002 sec)\n",
      "Step 11500: loss = 0.48 (0.002 sec)\n",
      "Step 11550: loss = 0.55 (0.002 sec)\n",
      "Step 11600: loss = 0.50 (0.003 sec)\n",
      "Step 11650: loss = 0.51 (0.002 sec)\n",
      "Step 11700: loss = 0.53 (0.002 sec)\n",
      "Step 11750: loss = 0.47 (0.002 sec)\n",
      "Step 11800: loss = 0.55 (0.002 sec)\n",
      "Step 11850: loss = 0.54 (0.002 sec)\n",
      "Step 11900: loss = 0.52 (0.002 sec)\n",
      "Step 11950: loss = 0.50 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2648  Precision @ 1: 0.8275\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1072  Precision @ 1: 0.6700\n",
      "Step 12000: loss = 0.58 (0.007 sec)\n",
      "Step 12050: loss = 0.49 (0.002 sec)\n",
      "Step 12100: loss = 0.49 (0.002 sec)\n",
      "Step 12150: loss = 0.51 (0.002 sec)\n",
      "Step 12200: loss = 0.43 (0.002 sec)\n",
      "Step 12250: loss = 0.55 (0.002 sec)\n",
      "Step 12300: loss = 0.47 (0.002 sec)\n",
      "Step 12350: loss = 0.41 (0.002 sec)\n",
      "Step 12400: loss = 0.49 (0.004 sec)\n",
      "Step 12450: loss = 0.49 (0.002 sec)\n",
      "Step 12500: loss = 0.50 (0.002 sec)\n",
      "Step 12550: loss = 0.48 (0.002 sec)\n",
      "Step 12600: loss = 0.48 (0.002 sec)\n",
      "Step 12650: loss = 0.46 (0.003 sec)\n",
      "Step 12700: loss = 0.53 (0.002 sec)\n",
      "Step 12750: loss = 0.45 (0.002 sec)\n",
      "Step 12800: loss = 0.51 (0.003 sec)\n",
      "Step 12850: loss = 0.50 (0.002 sec)\n",
      "Step 12900: loss = 0.42 (0.002 sec)\n",
      "Step 12950: loss = 0.47 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2653  Precision @ 1: 0.8291\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1097  Precision @ 1: 0.6856\n",
      "Step 13000: loss = 0.45 (0.003 sec)\n",
      "Step 13050: loss = 0.44 (0.002 sec)\n",
      "Step 13100: loss = 0.35 (0.002 sec)\n",
      "Step 13150: loss = 0.45 (0.002 sec)\n",
      "Step 13200: loss = 0.37 (0.004 sec)\n",
      "Step 13250: loss = 0.50 (0.002 sec)\n",
      "Step 13300: loss = 0.56 (0.002 sec)\n",
      "Step 13350: loss = 0.47 (0.002 sec)\n",
      "Step 13400: loss = 0.52 (0.002 sec)\n",
      "Step 13450: loss = 0.53 (0.002 sec)\n",
      "Step 13500: loss = 0.44 (0.002 sec)\n",
      "Step 13550: loss = 0.49 (0.002 sec)\n",
      "Step 13600: loss = 0.43 (0.003 sec)\n",
      "Step 13650: loss = 0.44 (0.002 sec)\n",
      "Step 13700: loss = 0.45 (0.002 sec)\n",
      "Step 13750: loss = 0.48 (0.002 sec)\n",
      "Step 13800: loss = 0.52 (0.002 sec)\n",
      "Step 13850: loss = 0.45 (0.002 sec)\n",
      "Step 13900: loss = 0.49 (0.002 sec)\n",
      "Step 13950: loss = 0.51 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2715  Precision @ 1: 0.8484\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1118  Precision @ 1: 0.6987\n",
      "Step 14000: loss = 0.41 (0.005 sec)\n",
      "Step 14050: loss = 0.46 (0.002 sec)\n",
      "Step 14100: loss = 0.41 (0.002 sec)\n",
      "Step 14150: loss = 0.42 (0.002 sec)\n",
      "Step 14200: loss = 0.56 (0.002 sec)\n",
      "Step 14250: loss = 0.48 (0.002 sec)\n",
      "Step 14300: loss = 0.39 (0.002 sec)\n",
      "Step 14350: loss = 0.44 (0.002 sec)\n",
      "Step 14400: loss = 0.46 (0.004 sec)\n",
      "Step 14450: loss = 0.33 (0.002 sec)\n",
      "Step 14500: loss = 0.37 (0.002 sec)\n",
      "Step 14550: loss = 0.48 (0.003 sec)\n",
      "Step 14600: loss = 0.37 (0.002 sec)\n",
      "Step 14650: loss = 0.49 (0.003 sec)\n",
      "Step 14700: loss = 0.34 (0.002 sec)\n",
      "Step 14750: loss = 0.53 (0.002 sec)\n",
      "Step 14800: loss = 0.42 (0.004 sec)\n",
      "Step 14850: loss = 0.40 (0.002 sec)\n",
      "Step 14900: loss = 0.42 (0.002 sec)\n",
      "Step 14950: loss = 0.45 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2748  Precision @ 1: 0.8588\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1132  Precision @ 1: 0.7075\n",
      "Step 15000: loss = 0.37 (0.002 sec)\n",
      "Step 15050: loss = 0.44 (0.002 sec)\n",
      "Step 15100: loss = 0.35 (0.002 sec)\n",
      "Step 15150: loss = 0.42 (0.002 sec)\n",
      "Step 15200: loss = 0.38 (0.003 sec)\n",
      "Step 15250: loss = 0.32 (0.002 sec)\n",
      "Step 15300: loss = 0.38 (0.002 sec)\n",
      "Step 15350: loss = 0.35 (0.002 sec)\n",
      "Step 15400: loss = 0.38 (0.002 sec)\n",
      "Step 15450: loss = 0.35 (0.002 sec)\n",
      "Step 15500: loss = 0.43 (0.002 sec)\n",
      "Step 15550: loss = 0.42 (0.002 sec)\n",
      "Step 15600: loss = 0.49 (0.004 sec)\n",
      "Step 15650: loss = 0.44 (0.002 sec)\n",
      "Step 15700: loss = 0.38 (0.002 sec)\n",
      "Step 15750: loss = 0.39 (0.002 sec)\n",
      "Step 15800: loss = 0.46 (0.002 sec)\n",
      "Step 15850: loss = 0.39 (0.002 sec)\n",
      "Step 15900: loss = 0.47 (0.002 sec)\n",
      "Step 15950: loss = 0.48 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2765  Precision @ 1: 0.8641\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1165  Precision @ 1: 0.7281\n",
      "Step 16000: loss = 0.30 (0.006 sec)\n",
      "Step 16050: loss = 0.48 (0.002 sec)\n",
      "Step 16100: loss = 0.37 (0.002 sec)\n",
      "Step 16150: loss = 0.37 (0.002 sec)\n",
      "Step 16200: loss = 0.41 (0.002 sec)\n",
      "Step 16250: loss = 0.34 (0.002 sec)\n",
      "Step 16300: loss = 0.39 (0.002 sec)\n",
      "Step 16350: loss = 0.42 (0.002 sec)\n",
      "Step 16400: loss = 0.37 (0.004 sec)\n",
      "Step 16450: loss = 0.38 (0.002 sec)\n",
      "Step 16500: loss = 0.46 (0.002 sec)\n",
      "Step 16550: loss = 0.38 (0.002 sec)\n",
      "Step 16600: loss = 0.34 (0.002 sec)\n",
      "Step 16650: loss = 0.36 (0.002 sec)\n",
      "Step 16700: loss = 0.41 (0.002 sec)\n",
      "Step 16750: loss = 0.33 (0.002 sec)\n",
      "Step 16800: loss = 0.30 (0.003 sec)\n",
      "Step 16850: loss = 0.36 (0.002 sec)\n",
      "Step 16900: loss = 0.34 (0.002 sec)\n",
      "Step 16950: loss = 0.41 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2760  Precision @ 1: 0.8625\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1158  Precision @ 1: 0.7238\n",
      "Step 17000: loss = 0.42 (0.003 sec)\n",
      "Step 17050: loss = 0.36 (0.002 sec)\n",
      "Step 17100: loss = 0.37 (0.002 sec)\n",
      "Step 17150: loss = 0.51 (0.002 sec)\n",
      "Step 17200: loss = 0.40 (0.004 sec)\n",
      "Step 17250: loss = 0.35 (0.002 sec)\n",
      "Step 17300: loss = 0.34 (0.002 sec)\n",
      "Step 17350: loss = 0.35 (0.002 sec)\n",
      "Step 17400: loss = 0.34 (0.002 sec)\n",
      "Step 17450: loss = 0.26 (0.002 sec)\n",
      "Step 17500: loss = 0.35 (0.002 sec)\n",
      "Step 17550: loss = 0.39 (0.002 sec)\n",
      "Step 17600: loss = 0.36 (0.004 sec)\n",
      "Step 17650: loss = 0.42 (0.002 sec)\n",
      "Step 17700: loss = 0.29 (0.002 sec)\n",
      "Step 17750: loss = 0.41 (0.002 sec)\n",
      "Step 17800: loss = 0.29 (0.002 sec)\n",
      "Step 17850: loss = 0.34 (0.002 sec)\n",
      "Step 17900: loss = 0.33 (0.002 sec)\n",
      "Step 17950: loss = 0.43 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2788  Precision @ 1: 0.8712\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1161  Precision @ 1: 0.7256\n",
      "Step 18000: loss = 0.39 (0.006 sec)\n",
      "Step 18050: loss = 0.40 (0.002 sec)\n",
      "Step 18100: loss = 0.36 (0.002 sec)\n",
      "Step 18150: loss = 0.31 (0.002 sec)\n",
      "Step 18200: loss = 0.29 (0.002 sec)\n",
      "Step 18250: loss = 0.38 (0.002 sec)\n",
      "Step 18300: loss = 0.36 (0.002 sec)\n",
      "Step 18350: loss = 0.36 (0.002 sec)\n",
      "Step 18400: loss = 0.33 (0.004 sec)\n",
      "Step 18450: loss = 0.39 (0.002 sec)\n",
      "Step 18500: loss = 0.36 (0.002 sec)\n",
      "Step 18550: loss = 0.36 (0.002 sec)\n",
      "Step 18600: loss = 0.35 (0.002 sec)\n",
      "Step 18650: loss = 0.51 (0.002 sec)\n",
      "Step 18700: loss = 0.28 (0.002 sec)\n",
      "Step 18750: loss = 0.38 (0.002 sec)\n",
      "Step 18800: loss = 0.36 (0.004 sec)\n",
      "Step 18850: loss = 0.30 (0.002 sec)\n",
      "Step 18900: loss = 0.39 (0.002 sec)\n",
      "Step 18950: loss = 0.31 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2858  Precision @ 1: 0.8931\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1189  Precision @ 1: 0.7431\n",
      "Step 19000: loss = 0.29 (0.003 sec)\n",
      "Step 19050: loss = 0.33 (0.003 sec)\n",
      "Step 19100: loss = 0.29 (0.002 sec)\n",
      "Step 19150: loss = 0.35 (0.002 sec)\n",
      "Step 19200: loss = 0.30 (0.003 sec)\n",
      "Step 19250: loss = 0.33 (0.003 sec)\n",
      "Step 19300: loss = 0.37 (0.002 sec)\n",
      "Step 19350: loss = 0.33 (0.002 sec)\n",
      "Step 19400: loss = 0.35 (0.002 sec)\n",
      "Step 19450: loss = 0.35 (0.002 sec)\n",
      "Step 19500: loss = 0.36 (0.002 sec)\n",
      "Step 19550: loss = 0.34 (0.002 sec)\n",
      "Step 19600: loss = 0.34 (0.004 sec)\n",
      "Step 19650: loss = 0.35 (0.003 sec)\n",
      "Step 19700: loss = 0.35 (0.002 sec)\n",
      "Step 19750: loss = 0.32 (0.002 sec)\n",
      "Step 19800: loss = 0.26 (0.002 sec)\n",
      "Step 19850: loss = 0.41 (0.002 sec)\n",
      "Step 19900: loss = 0.31 (0.002 sec)\n",
      "Step 19950: loss = 0.28 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2839  Precision @ 1: 0.8872\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1195  Precision @ 1: 0.7469\n",
      "Step 20000: loss = 0.32 (0.005 sec)\n",
      "Step 20050: loss = 0.28 (0.002 sec)\n",
      "Step 20100: loss = 0.31 (0.002 sec)\n",
      "Step 20150: loss = 0.28 (0.002 sec)\n",
      "Step 20200: loss = 0.31 (0.002 sec)\n",
      "Step 20250: loss = 0.32 (0.002 sec)\n",
      "Step 20300: loss = 0.34 (0.002 sec)\n",
      "Step 20350: loss = 0.36 (0.002 sec)\n",
      "Step 20400: loss = 0.44 (0.005 sec)\n",
      "Step 20450: loss = 0.35 (0.002 sec)\n",
      "Step 20500: loss = 0.28 (0.002 sec)\n",
      "Step 20550: loss = 0.29 (0.002 sec)\n",
      "Step 20600: loss = 0.35 (0.003 sec)\n",
      "Step 20650: loss = 0.32 (0.002 sec)\n",
      "Step 20700: loss = 0.29 (0.002 sec)\n",
      "Step 20750: loss = 0.27 (0.002 sec)\n",
      "Step 20800: loss = 0.30 (0.003 sec)\n",
      "Step 20850: loss = 0.30 (0.002 sec)\n",
      "Step 20900: loss = 0.34 (0.002 sec)\n",
      "Step 20950: loss = 0.32 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2872  Precision @ 1: 0.8975\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1210  Precision @ 1: 0.7562\n",
      "Step 21000: loss = 0.29 (0.003 sec)\n",
      "Step 21050: loss = 0.28 (0.002 sec)\n",
      "Step 21100: loss = 0.23 (0.002 sec)\n",
      "Step 21150: loss = 0.29 (0.002 sec)\n",
      "Step 21200: loss = 0.30 (0.004 sec)\n",
      "Step 21250: loss = 0.26 (0.002 sec)\n",
      "Step 21300: loss = 0.33 (0.002 sec)\n",
      "Step 21350: loss = 0.29 (0.002 sec)\n",
      "Step 21400: loss = 0.23 (0.002 sec)\n",
      "Step 21450: loss = 0.28 (0.002 sec)\n",
      "Step 21500: loss = 0.31 (0.003 sec)\n",
      "Step 21550: loss = 0.30 (0.002 sec)\n",
      "Step 21600: loss = 0.34 (0.004 sec)\n",
      "Step 21650: loss = 0.30 (0.002 sec)\n",
      "Step 21700: loss = 0.28 (0.003 sec)\n",
      "Step 21750: loss = 0.22 (0.002 sec)\n",
      "Step 21800: loss = 0.34 (0.002 sec)\n",
      "Step 21850: loss = 0.33 (0.002 sec)\n",
      "Step 21900: loss = 0.28 (0.002 sec)\n",
      "Step 21950: loss = 0.38 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2868  Precision @ 1: 0.8962\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1214  Precision @ 1: 0.7588\n",
      "Step 22000: loss = 0.27 (0.005 sec)\n",
      "Step 22050: loss = 0.28 (0.002 sec)\n",
      "Step 22100: loss = 0.29 (0.002 sec)\n",
      "Step 22150: loss = 0.26 (0.002 sec)\n",
      "Step 22200: loss = 0.24 (0.003 sec)\n",
      "Step 22250: loss = 0.34 (0.002 sec)\n",
      "Step 22300: loss = 0.26 (0.002 sec)\n",
      "Step 22350: loss = 0.32 (0.002 sec)\n",
      "Step 22400: loss = 0.22 (0.004 sec)\n",
      "Step 22450: loss = 0.29 (0.002 sec)\n",
      "Step 22500: loss = 0.34 (0.002 sec)\n",
      "Step 22550: loss = 0.28 (0.002 sec)\n",
      "Step 22600: loss = 0.30 (0.002 sec)\n",
      "Step 22650: loss = 0.32 (0.002 sec)\n",
      "Step 22700: loss = 0.32 (0.002 sec)\n",
      "Step 22750: loss = 0.32 (0.002 sec)\n",
      "Step 22800: loss = 0.30 (0.003 sec)\n",
      "Step 22850: loss = 0.25 (0.002 sec)\n",
      "Step 22900: loss = 0.32 (0.002 sec)\n",
      "Step 22950: loss = 0.19 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2883  Precision @ 1: 0.9009\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1206  Precision @ 1: 0.7538\n",
      "Step 23000: loss = 0.27 (0.003 sec)\n",
      "Step 23050: loss = 0.24 (0.002 sec)\n",
      "Step 23100: loss = 0.28 (0.002 sec)\n",
      "Step 23150: loss = 0.25 (0.002 sec)\n",
      "Step 23200: loss = 0.25 (0.003 sec)\n",
      "Step 23250: loss = 0.29 (0.002 sec)\n",
      "Step 23300: loss = 0.25 (0.002 sec)\n",
      "Step 23350: loss = 0.28 (0.002 sec)\n",
      "Step 23400: loss = 0.41 (0.002 sec)\n",
      "Step 23450: loss = 0.28 (0.002 sec)\n",
      "Step 23500: loss = 0.25 (0.002 sec)\n",
      "Step 23550: loss = 0.28 (0.002 sec)\n",
      "Step 23600: loss = 0.34 (0.003 sec)\n",
      "Step 23650: loss = 0.29 (0.002 sec)\n",
      "Step 23700: loss = 0.24 (0.002 sec)\n",
      "Step 23750: loss = 0.27 (0.002 sec)\n",
      "Step 23800: loss = 0.23 (0.002 sec)\n",
      "Step 23850: loss = 0.23 (0.002 sec)\n",
      "Step 23900: loss = 0.22 (0.003 sec)\n",
      "Step 23950: loss = 0.24 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2919  Precision @ 1: 0.9122\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1249  Precision @ 1: 0.7806\n",
      "Step 24000: loss = 0.26 (0.005 sec)\n",
      "Step 24050: loss = 0.28 (0.002 sec)\n",
      "Step 24100: loss = 0.30 (0.002 sec)\n",
      "Step 24150: loss = 0.26 (0.003 sec)\n",
      "Step 24200: loss = 0.23 (0.002 sec)\n",
      "Step 24250: loss = 0.33 (0.002 sec)\n",
      "Step 24300: loss = 0.32 (0.002 sec)\n",
      "Step 24350: loss = 0.25 (0.002 sec)\n",
      "Step 24400: loss = 0.25 (0.003 sec)\n",
      "Step 24450: loss = 0.25 (0.002 sec)\n",
      "Step 24500: loss = 0.27 (0.002 sec)\n",
      "Step 24550: loss = 0.31 (0.002 sec)\n",
      "Step 24600: loss = 0.26 (0.002 sec)\n",
      "Step 24650: loss = 0.25 (0.002 sec)\n",
      "Step 24700: loss = 0.22 (0.002 sec)\n",
      "Step 24750: loss = 0.30 (0.002 sec)\n",
      "Step 24800: loss = 0.26 (0.003 sec)\n",
      "Step 24850: loss = 0.20 (0.002 sec)\n",
      "Step 24900: loss = 0.20 (0.002 sec)\n",
      "Step 24950: loss = 0.29 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2916  Precision @ 1: 0.9113\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1236  Precision @ 1: 0.7725\n",
      "Step 25000: loss = 0.26 (0.003 sec)\n",
      "Step 25050: loss = 0.28 (0.002 sec)\n",
      "Step 25100: loss = 0.22 (0.002 sec)\n",
      "Step 25150: loss = 0.28 (0.002 sec)\n",
      "Step 25200: loss = 0.21 (0.003 sec)\n",
      "Step 25250: loss = 0.23 (0.002 sec)\n",
      "Step 25300: loss = 0.27 (0.002 sec)\n",
      "Step 25350: loss = 0.25 (0.002 sec)\n",
      "Step 25400: loss = 0.26 (0.002 sec)\n",
      "Step 25450: loss = 0.23 (0.002 sec)\n",
      "Step 25500: loss = 0.22 (0.002 sec)\n",
      "Step 25550: loss = 0.22 (0.002 sec)\n",
      "Step 25600: loss = 0.25 (0.004 sec)\n",
      "Step 25650: loss = 0.24 (0.002 sec)\n",
      "Step 25700: loss = 0.29 (0.002 sec)\n",
      "Step 25750: loss = 0.29 (0.002 sec)\n",
      "Step 25800: loss = 0.26 (0.002 sec)\n",
      "Step 25850: loss = 0.25 (0.002 sec)\n",
      "Step 25900: loss = 0.23 (0.002 sec)\n",
      "Step 25950: loss = 0.25 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2919  Precision @ 1: 0.9122\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1259  Precision @ 1: 0.7869\n",
      "Step 26000: loss = 0.22 (0.005 sec)\n",
      "Step 26050: loss = 0.29 (0.002 sec)\n",
      "Step 26100: loss = 0.26 (0.003 sec)\n",
      "Step 26150: loss = 0.31 (0.002 sec)\n",
      "Step 26200: loss = 0.25 (0.003 sec)\n",
      "Step 26250: loss = 0.21 (0.002 sec)\n",
      "Step 26300: loss = 0.26 (0.002 sec)\n",
      "Step 26350: loss = 0.20 (0.002 sec)\n",
      "Step 26400: loss = 0.26 (0.003 sec)\n",
      "Step 26450: loss = 0.19 (0.002 sec)\n",
      "Step 26500: loss = 0.27 (0.003 sec)\n",
      "Step 26550: loss = 0.34 (0.002 sec)\n",
      "Step 26600: loss = 0.26 (0.003 sec)\n",
      "Step 26650: loss = 0.23 (0.002 sec)\n",
      "Step 26700: loss = 0.22 (0.002 sec)\n",
      "Step 26750: loss = 0.23 (0.002 sec)\n",
      "Step 26800: loss = 0.26 (0.004 sec)\n",
      "Step 26850: loss = 0.25 (0.002 sec)\n",
      "Step 26900: loss = 0.22 (0.002 sec)\n",
      "Step 26950: loss = 0.21 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2930  Precision @ 1: 0.9156\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1256  Precision @ 1: 0.7850\n",
      "Step 27000: loss = 0.21 (0.003 sec)\n",
      "Step 27050: loss = 0.23 (0.002 sec)\n",
      "Step 27100: loss = 0.26 (0.002 sec)\n",
      "Step 27150: loss = 0.22 (0.003 sec)\n",
      "Step 27200: loss = 0.19 (0.004 sec)\n",
      "Step 27250: loss = 0.27 (0.002 sec)\n",
      "Step 27300: loss = 0.20 (0.004 sec)\n",
      "Step 27350: loss = 0.28 (0.002 sec)\n",
      "Step 27400: loss = 0.21 (0.002 sec)\n",
      "Step 27450: loss = 0.23 (0.003 sec)\n",
      "Step 27500: loss = 0.29 (0.002 sec)\n",
      "Step 27550: loss = 0.16 (0.002 sec)\n",
      "Step 27600: loss = 0.14 (0.004 sec)\n",
      "Step 27650: loss = 0.20 (0.002 sec)\n",
      "Step 27700: loss = 0.27 (0.002 sec)\n",
      "Step 27750: loss = 0.20 (0.002 sec)\n",
      "Step 27800: loss = 0.27 (0.002 sec)\n",
      "Step 27850: loss = 0.22 (0.002 sec)\n",
      "Step 27900: loss = 0.20 (0.002 sec)\n",
      "Step 27950: loss = 0.23 (0.003 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2924  Precision @ 1: 0.9137\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1245  Precision @ 1: 0.7781\n",
      "Step 28000: loss = 0.29 (0.005 sec)\n",
      "Step 28050: loss = 0.32 (0.002 sec)\n",
      "Step 28100: loss = 0.25 (0.002 sec)\n",
      "Step 28150: loss = 0.23 (0.002 sec)\n",
      "Step 28200: loss = 0.27 (0.002 sec)\n",
      "Step 28250: loss = 0.23 (0.002 sec)\n",
      "Step 28300: loss = 0.28 (0.002 sec)\n",
      "Step 28350: loss = 0.21 (0.002 sec)\n",
      "Step 28400: loss = 0.21 (0.003 sec)\n",
      "Step 28450: loss = 0.29 (0.002 sec)\n",
      "Step 28500: loss = 0.29 (0.003 sec)\n",
      "Step 28550: loss = 0.23 (0.002 sec)\n",
      "Step 28600: loss = 0.17 (0.002 sec)\n",
      "Step 28650: loss = 0.21 (0.002 sec)\n",
      "Step 28700: loss = 0.19 (0.002 sec)\n",
      "Step 28750: loss = 0.21 (0.002 sec)\n",
      "Step 28800: loss = 0.21 (0.003 sec)\n",
      "Step 28850: loss = 0.20 (0.002 sec)\n",
      "Step 28900: loss = 0.22 (0.002 sec)\n",
      "Step 28950: loss = 0.20 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2955  Precision @ 1: 0.9234\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1273  Precision @ 1: 0.7956\n",
      "Step 29000: loss = 0.15 (0.003 sec)\n",
      "Step 29050: loss = 0.23 (0.002 sec)\n",
      "Step 29100: loss = 0.21 (0.002 sec)\n",
      "Step 29150: loss = 0.29 (0.002 sec)\n",
      "Step 29200: loss = 0.31 (0.004 sec)\n",
      "Step 29250: loss = 0.25 (0.002 sec)\n",
      "Step 29300: loss = 0.21 (0.002 sec)\n",
      "Step 29350: loss = 0.26 (0.002 sec)\n",
      "Step 29400: loss = 0.22 (0.002 sec)\n",
      "Step 29450: loss = 0.20 (0.002 sec)\n",
      "Step 29500: loss = 0.19 (0.002 sec)\n",
      "Step 29550: loss = 0.25 (0.002 sec)\n",
      "Step 29600: loss = 0.25 (0.004 sec)\n",
      "Step 29650: loss = 0.25 (0.002 sec)\n",
      "Step 29700: loss = 0.22 (0.002 sec)\n",
      "Step 29750: loss = 0.21 (0.002 sec)\n",
      "Step 29800: loss = 0.23 (0.003 sec)\n",
      "Step 29850: loss = 0.29 (0.002 sec)\n",
      "Step 29900: loss = 0.22 (0.002 sec)\n",
      "Step 29950: loss = 0.16 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2925  Precision @ 1: 0.9141\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1246  Precision @ 1: 0.7788\n",
      "Step 30000: loss = 0.25 (0.006 sec)\n",
      "Step 30050: loss = 0.24 (0.002 sec)\n",
      "Step 30100: loss = 0.24 (0.002 sec)\n",
      "Step 30150: loss = 0.19 (0.003 sec)\n",
      "Step 30200: loss = 0.18 (0.002 sec)\n",
      "Step 30250: loss = 0.19 (0.002 sec)\n",
      "Step 30300: loss = 0.23 (0.002 sec)\n",
      "Step 30350: loss = 0.16 (0.002 sec)\n",
      "Step 30400: loss = 0.20 (0.003 sec)\n",
      "Step 30450: loss = 0.26 (0.002 sec)\n",
      "Step 30500: loss = 0.24 (0.002 sec)\n",
      "Step 30550: loss = 0.23 (0.002 sec)\n",
      "Step 30600: loss = 0.24 (0.002 sec)\n",
      "Step 30650: loss = 0.25 (0.002 sec)\n",
      "Step 30700: loss = 0.25 (0.002 sec)\n",
      "Step 30750: loss = 0.22 (0.002 sec)\n",
      "Step 30800: loss = 0.18 (0.003 sec)\n",
      "Step 30850: loss = 0.16 (0.002 sec)\n",
      "Step 30900: loss = 0.26 (0.002 sec)\n",
      "Step 30950: loss = 0.21 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2959  Precision @ 1: 0.9247\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1281  Precision @ 1: 0.8006\n",
      "Step 31000: loss = 0.25 (0.003 sec)\n",
      "Step 31050: loss = 0.21 (0.002 sec)\n",
      "Step 31100: loss = 0.14 (0.002 sec)\n",
      "Step 31150: loss = 0.19 (0.002 sec)\n",
      "Step 31200: loss = 0.22 (0.004 sec)\n",
      "Step 31250: loss = 0.19 (0.002 sec)\n",
      "Step 31300: loss = 0.21 (0.002 sec)\n",
      "Step 31350: loss = 0.26 (0.002 sec)\n",
      "Step 31400: loss = 0.15 (0.002 sec)\n",
      "Step 31450: loss = 0.18 (0.002 sec)\n",
      "Step 31500: loss = 0.18 (0.002 sec)\n",
      "Step 31550: loss = 0.19 (0.002 sec)\n",
      "Step 31600: loss = 0.23 (0.004 sec)\n",
      "Step 31650: loss = 0.18 (0.002 sec)\n",
      "Step 31700: loss = 0.22 (0.002 sec)\n",
      "Step 31750: loss = 0.17 (0.002 sec)\n",
      "Step 31800: loss = 0.25 (0.002 sec)\n",
      "Step 31850: loss = 0.26 (0.002 sec)\n",
      "Step 31900: loss = 0.17 (0.003 sec)\n",
      "Step 31950: loss = 0.18 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2969  Precision @ 1: 0.9278\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1279  Precision @ 1: 0.7994\n",
      "Step 32000: loss = 0.23 (0.006 sec)\n",
      "Step 32050: loss = 0.19 (0.002 sec)\n",
      "Step 32100: loss = 0.21 (0.002 sec)\n",
      "Step 32150: loss = 0.20 (0.002 sec)\n",
      "Step 32200: loss = 0.17 (0.002 sec)\n",
      "Step 32250: loss = 0.17 (0.002 sec)\n",
      "Step 32300: loss = 0.20 (0.002 sec)\n",
      "Step 32350: loss = 0.24 (0.002 sec)\n",
      "Step 32400: loss = 0.21 (0.004 sec)\n",
      "Step 32450: loss = 0.19 (0.002 sec)\n",
      "Step 32500: loss = 0.23 (0.002 sec)\n",
      "Step 32550: loss = 0.16 (0.002 sec)\n",
      "Step 32600: loss = 0.16 (0.002 sec)\n",
      "Step 32650: loss = 0.17 (0.002 sec)\n",
      "Step 32700: loss = 0.23 (0.002 sec)\n",
      "Step 32750: loss = 0.22 (0.003 sec)\n",
      "Step 32800: loss = 0.17 (0.003 sec)\n",
      "Step 32850: loss = 0.26 (0.002 sec)\n",
      "Step 32900: loss = 0.18 (0.002 sec)\n",
      "Step 32950: loss = 0.14 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2968  Precision @ 1: 0.9275\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1286  Precision @ 1: 0.8037\n",
      "Step 33000: loss = 0.21 (0.003 sec)\n",
      "Step 33050: loss = 0.21 (0.002 sec)\n",
      "Step 33100: loss = 0.23 (0.002 sec)\n",
      "Step 33150: loss = 0.21 (0.002 sec)\n",
      "Step 33200: loss = 0.16 (0.004 sec)\n",
      "Step 33250: loss = 0.25 (0.002 sec)\n",
      "Step 33300: loss = 0.22 (0.002 sec)\n",
      "Step 33350: loss = 0.20 (0.003 sec)\n",
      "Step 33400: loss = 0.22 (0.002 sec)\n",
      "Step 33450: loss = 0.20 (0.002 sec)\n",
      "Step 33500: loss = 0.23 (0.002 sec)\n",
      "Step 33550: loss = 0.22 (0.002 sec)\n",
      "Step 33600: loss = 0.15 (0.004 sec)\n",
      "Step 33650: loss = 0.22 (0.002 sec)\n",
      "Step 33700: loss = 0.20 (0.002 sec)\n",
      "Step 33750: loss = 0.22 (0.002 sec)\n",
      "Step 33800: loss = 0.23 (0.002 sec)\n",
      "Step 33850: loss = 0.22 (0.003 sec)\n",
      "Step 33900: loss = 0.22 (0.002 sec)\n",
      "Step 33950: loss = 0.17 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2982  Precision @ 1: 0.9319\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1291  Precision @ 1: 0.8069\n",
      "Step 34000: loss = 0.16 (0.006 sec)\n",
      "Step 34050: loss = 0.19 (0.002 sec)\n",
      "Step 34100: loss = 0.24 (0.002 sec)\n",
      "Step 34150: loss = 0.19 (0.002 sec)\n",
      "Step 34200: loss = 0.23 (0.002 sec)\n",
      "Step 34250: loss = 0.20 (0.002 sec)\n",
      "Step 34300: loss = 0.23 (0.002 sec)\n",
      "Step 34350: loss = 0.14 (0.002 sec)\n",
      "Step 34400: loss = 0.15 (0.004 sec)\n",
      "Step 34450: loss = 0.19 (0.003 sec)\n",
      "Step 34500: loss = 0.19 (0.002 sec)\n",
      "Step 34550: loss = 0.17 (0.002 sec)\n",
      "Step 34600: loss = 0.17 (0.002 sec)\n",
      "Step 34650: loss = 0.20 (0.002 sec)\n",
      "Step 34700: loss = 0.18 (0.002 sec)\n",
      "Step 34750: loss = 0.21 (0.002 sec)\n",
      "Step 34800: loss = 0.27 (0.003 sec)\n",
      "Step 34850: loss = 0.18 (0.002 sec)\n",
      "Step 34900: loss = 0.22 (0.002 sec)\n",
      "Step 34950: loss = 0.18 (0.003 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2985  Precision @ 1: 0.9328\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1295  Precision @ 1: 0.8094\n",
      "Step 35000: loss = 0.18 (0.003 sec)\n",
      "Step 35050: loss = 0.14 (0.002 sec)\n",
      "Step 35100: loss = 0.22 (0.002 sec)\n",
      "Step 35150: loss = 0.17 (0.002 sec)\n",
      "Step 35200: loss = 0.12 (0.003 sec)\n",
      "Step 35250: loss = 0.15 (0.002 sec)\n",
      "Step 35300: loss = 0.17 (0.002 sec)\n",
      "Step 35350: loss = 0.16 (0.002 sec)\n",
      "Step 35400: loss = 0.23 (0.002 sec)\n",
      "Step 35450: loss = 0.21 (0.002 sec)\n",
      "Step 35500: loss = 0.16 (0.002 sec)\n",
      "Step 35550: loss = 0.14 (0.003 sec)\n",
      "Step 35600: loss = 0.13 (0.004 sec)\n",
      "Step 35650: loss = 0.21 (0.002 sec)\n",
      "Step 35700: loss = 0.20 (0.002 sec)\n",
      "Step 35750: loss = 0.13 (0.002 sec)\n",
      "Step 35800: loss = 0.19 (0.002 sec)\n",
      "Step 35850: loss = 0.28 (0.003 sec)\n",
      "Step 35900: loss = 0.23 (0.002 sec)\n",
      "Step 35950: loss = 0.17 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2969  Precision @ 1: 0.9278\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1270  Precision @ 1: 0.7937\n",
      "Step 36000: loss = 0.15 (0.007 sec)\n",
      "Step 36050: loss = 0.19 (0.002 sec)\n",
      "Step 36100: loss = 0.16 (0.002 sec)\n",
      "Step 36150: loss = 0.16 (0.002 sec)\n",
      "Step 36200: loss = 0.16 (0.002 sec)\n",
      "Step 36250: loss = 0.24 (0.002 sec)\n",
      "Step 36300: loss = 0.23 (0.002 sec)\n",
      "Step 36350: loss = 0.17 (0.002 sec)\n",
      "Step 36400: loss = 0.23 (0.005 sec)\n",
      "Step 36450: loss = 0.14 (0.002 sec)\n",
      "Step 36500: loss = 0.23 (0.003 sec)\n",
      "Step 36550: loss = 0.19 (0.003 sec)\n",
      "Step 36600: loss = 0.18 (0.002 sec)\n",
      "Step 36650: loss = 0.23 (0.002 sec)\n",
      "Step 36700: loss = 0.13 (0.002 sec)\n",
      "Step 36750: loss = 0.18 (0.002 sec)\n",
      "Step 36800: loss = 0.21 (0.003 sec)\n",
      "Step 36850: loss = 0.18 (0.002 sec)\n",
      "Step 36900: loss = 0.19 (0.002 sec)\n",
      "Step 36950: loss = 0.16 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 2980  Precision @ 1: 0.9313\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1290  Precision @ 1: 0.8063\n",
      "Step 37000: loss = 0.11 (0.003 sec)\n",
      "Step 37050: loss = 0.25 (0.003 sec)\n",
      "Step 37100: loss = 0.17 (0.002 sec)\n",
      "Step 37150: loss = 0.22 (0.002 sec)\n",
      "Step 37200: loss = 0.16 (0.004 sec)\n",
      "Step 37250: loss = 0.25 (0.002 sec)\n",
      "Step 37300: loss = 0.18 (0.002 sec)\n",
      "Step 37350: loss = 0.18 (0.002 sec)\n",
      "Step 37400: loss = 0.19 (0.002 sec)\n",
      "Step 37450: loss = 0.19 (0.002 sec)\n",
      "Step 37500: loss = 0.19 (0.003 sec)\n",
      "Step 37550: loss = 0.15 (0.003 sec)\n",
      "Step 37600: loss = 0.16 (0.003 sec)\n",
      "Step 37650: loss = 0.20 (0.002 sec)\n",
      "Step 37700: loss = 0.15 (0.002 sec)\n",
      "Step 37750: loss = 0.14 (0.002 sec)\n",
      "Step 37800: loss = 0.22 (0.002 sec)\n",
      "Step 37850: loss = 0.16 (0.002 sec)\n",
      "Step 37900: loss = 0.21 (0.002 sec)\n",
      "Step 37950: loss = 0.16 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 3010  Precision @ 1: 0.9406\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1300  Precision @ 1: 0.8125\n",
      "Step 38000: loss = 0.16 (0.005 sec)\n",
      "Step 38050: loss = 0.22 (0.002 sec)\n",
      "Step 38100: loss = 0.20 (0.002 sec)\n",
      "Step 38150: loss = 0.18 (0.002 sec)\n",
      "Step 38200: loss = 0.20 (0.002 sec)\n",
      "Step 38250: loss = 0.17 (0.002 sec)\n",
      "Step 38300: loss = 0.17 (0.003 sec)\n",
      "Step 38350: loss = 0.15 (0.002 sec)\n",
      "Step 38400: loss = 0.16 (0.003 sec)\n",
      "Step 38450: loss = 0.16 (0.003 sec)\n",
      "Step 38500: loss = 0.17 (0.002 sec)\n",
      "Step 38550: loss = 0.14 (0.002 sec)\n",
      "Step 38600: loss = 0.15 (0.002 sec)\n",
      "Step 38650: loss = 0.18 (0.002 sec)\n",
      "Step 38700: loss = 0.20 (0.002 sec)\n",
      "Step 38750: loss = 0.17 (0.002 sec)\n",
      "Step 38800: loss = 0.15 (0.003 sec)\n",
      "Step 38850: loss = 0.19 (0.002 sec)\n",
      "Step 38900: loss = 0.21 (0.002 sec)\n",
      "Step 38950: loss = 0.22 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 3002  Precision @ 1: 0.9381\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1302  Precision @ 1: 0.8137\n",
      "Step 39000: loss = 0.15 (0.003 sec)\n",
      "Step 39050: loss = 0.16 (0.002 sec)\n",
      "Step 39100: loss = 0.14 (0.002 sec)\n",
      "Step 39150: loss = 0.19 (0.002 sec)\n",
      "Step 39200: loss = 0.26 (0.003 sec)\n",
      "Step 39250: loss = 0.19 (0.002 sec)\n",
      "Step 39300: loss = 0.17 (0.002 sec)\n",
      "Step 39350: loss = 0.16 (0.003 sec)\n",
      "Step 39400: loss = 0.17 (0.002 sec)\n",
      "Step 39450: loss = 0.17 (0.002 sec)\n",
      "Step 39500: loss = 0.17 (0.002 sec)\n",
      "Step 39550: loss = 0.24 (0.002 sec)\n",
      "Step 39600: loss = 0.23 (0.003 sec)\n",
      "Step 39650: loss = 0.13 (0.002 sec)\n",
      "Step 39700: loss = 0.17 (0.003 sec)\n",
      "Step 39750: loss = 0.15 (0.002 sec)\n",
      "Step 39800: loss = 0.14 (0.002 sec)\n",
      "Step 39850: loss = 0.18 (0.002 sec)\n",
      "Step 39900: loss = 0.14 (0.002 sec)\n",
      "Step 39950: loss = 0.17 (0.003 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 3019  Precision @ 1: 0.9434\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1303  Precision @ 1: 0.8144\n",
      "Step 40000: loss = 0.14 (0.008 sec)\n",
      "Step 40050: loss = 0.24 (0.003 sec)\n",
      "Step 40100: loss = 0.18 (0.002 sec)\n",
      "Step 40150: loss = 0.13 (0.002 sec)\n",
      "Step 40200: loss = 0.14 (0.002 sec)\n",
      "Step 40250: loss = 0.22 (0.002 sec)\n",
      "Step 40300: loss = 0.20 (0.002 sec)\n",
      "Step 40350: loss = 0.20 (0.002 sec)\n",
      "Step 40400: loss = 0.12 (0.004 sec)\n",
      "Step 40450: loss = 0.18 (0.002 sec)\n",
      "Step 40500: loss = 0.19 (0.002 sec)\n",
      "Step 40550: loss = 0.12 (0.002 sec)\n",
      "Step 40600: loss = 0.15 (0.002 sec)\n",
      "Step 40650: loss = 0.13 (0.002 sec)\n",
      "Step 40700: loss = 0.16 (0.002 sec)\n",
      "Step 40750: loss = 0.17 (0.002 sec)\n",
      "Step 40800: loss = 0.16 (0.004 sec)\n",
      "Step 40850: loss = 0.19 (0.002 sec)\n",
      "Step 40900: loss = 0.19 (0.002 sec)\n",
      "Step 40950: loss = 0.12 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 3018  Precision @ 1: 0.9431\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1288  Precision @ 1: 0.8050\n",
      "Step 41000: loss = 0.18 (0.003 sec)\n",
      "Step 41050: loss = 0.13 (0.002 sec)\n",
      "Step 41100: loss = 0.18 (0.002 sec)\n",
      "Step 41150: loss = 0.19 (0.002 sec)\n",
      "Step 41200: loss = 0.14 (0.003 sec)\n",
      "Step 41250: loss = 0.13 (0.002 sec)\n",
      "Step 41300: loss = 0.14 (0.002 sec)\n",
      "Step 41350: loss = 0.14 (0.002 sec)\n",
      "Step 41400: loss = 0.17 (0.002 sec)\n",
      "Step 41450: loss = 0.12 (0.002 sec)\n",
      "Step 41500: loss = 0.18 (0.002 sec)\n",
      "Step 41550: loss = 0.15 (0.002 sec)\n",
      "Step 41600: loss = 0.15 (0.004 sec)\n",
      "Step 41650: loss = 0.21 (0.002 sec)\n",
      "Step 41700: loss = 0.18 (0.002 sec)\n",
      "Step 41750: loss = 0.22 (0.002 sec)\n",
      "Step 41800: loss = 0.12 (0.002 sec)\n",
      "Step 41850: loss = 0.12 (0.003 sec)\n",
      "Step 41900: loss = 0.21 (0.002 sec)\n",
      "Step 41950: loss = 0.13 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 3029  Precision @ 1: 0.9466\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1307  Precision @ 1: 0.8169\n",
      "Step 42000: loss = 0.14 (0.005 sec)\n",
      "Step 42050: loss = 0.16 (0.003 sec)\n",
      "Step 42100: loss = 0.12 (0.002 sec)\n",
      "Step 42150: loss = 0.15 (0.002 sec)\n",
      "Step 42200: loss = 0.18 (0.002 sec)\n",
      "Step 42250: loss = 0.12 (0.002 sec)\n",
      "Step 42300: loss = 0.16 (0.002 sec)\n",
      "Step 42350: loss = 0.17 (0.002 sec)\n",
      "Step 42400: loss = 0.13 (0.004 sec)\n",
      "Step 42450: loss = 0.17 (0.003 sec)\n",
      "Step 42500: loss = 0.21 (0.002 sec)\n",
      "Step 42550: loss = 0.13 (0.002 sec)\n",
      "Step 42600: loss = 0.16 (0.002 sec)\n",
      "Step 42650: loss = 0.15 (0.002 sec)\n",
      "Step 42700: loss = 0.14 (0.002 sec)\n",
      "Step 42750: loss = 0.14 (0.002 sec)\n",
      "Step 42800: loss = 0.14 (0.004 sec)\n",
      "Step 42850: loss = 0.14 (0.002 sec)\n",
      "Step 42900: loss = 0.16 (0.002 sec)\n",
      "Step 42950: loss = 0.11 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 3018  Precision @ 1: 0.9431\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1309  Precision @ 1: 0.8181\n",
      "Step 43000: loss = 0.16 (0.003 sec)\n",
      "Step 43050: loss = 0.16 (0.002 sec)\n",
      "Step 43100: loss = 0.13 (0.002 sec)\n",
      "Step 43150: loss = 0.19 (0.002 sec)\n",
      "Step 43200: loss = 0.16 (0.004 sec)\n",
      "Step 43250: loss = 0.15 (0.003 sec)\n",
      "Step 43300: loss = 0.17 (0.003 sec)\n",
      "Step 43350: loss = 0.15 (0.002 sec)\n",
      "Step 43400: loss = 0.17 (0.002 sec)\n",
      "Step 43450: loss = 0.15 (0.002 sec)\n",
      "Step 43500: loss = 0.17 (0.002 sec)\n",
      "Step 43550: loss = 0.13 (0.002 sec)\n",
      "Step 43600: loss = 0.14 (0.003 sec)\n",
      "Step 43650: loss = 0.09 (0.002 sec)\n",
      "Step 43700: loss = 0.17 (0.002 sec)\n",
      "Step 43750: loss = 0.10 (0.002 sec)\n",
      "Step 43800: loss = 0.14 (0.002 sec)\n",
      "Step 43850: loss = 0.18 (0.002 sec)\n",
      "Step 43900: loss = 0.10 (0.002 sec)\n",
      "Step 43950: loss = 0.13 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 3039  Precision @ 1: 0.9497\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1295  Precision @ 1: 0.8094\n",
      "Step 44000: loss = 0.17 (0.008 sec)\n",
      "Step 44050: loss = 0.13 (0.002 sec)\n",
      "Step 44100: loss = 0.13 (0.002 sec)\n",
      "Step 44150: loss = 0.16 (0.002 sec)\n",
      "Step 44200: loss = 0.16 (0.002 sec)\n",
      "Step 44250: loss = 0.14 (0.002 sec)\n",
      "Step 44300: loss = 0.10 (0.002 sec)\n",
      "Step 44350: loss = 0.13 (0.003 sec)\n",
      "Step 44400: loss = 0.21 (0.003 sec)\n",
      "Step 44450: loss = 0.11 (0.002 sec)\n",
      "Step 44500: loss = 0.13 (0.002 sec)\n",
      "Step 44550: loss = 0.16 (0.003 sec)\n",
      "Step 44600: loss = 0.21 (0.002 sec)\n",
      "Step 44650: loss = 0.10 (0.002 sec)\n",
      "Step 44700: loss = 0.13 (0.002 sec)\n",
      "Step 44750: loss = 0.16 (0.002 sec)\n",
      "Step 44800: loss = 0.14 (0.004 sec)\n",
      "Step 44850: loss = 0.12 (0.002 sec)\n",
      "Step 44900: loss = 0.15 (0.002 sec)\n",
      "Step 44950: loss = 0.14 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 3037  Precision @ 1: 0.9491\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1319  Precision @ 1: 0.8244\n",
      "Step 45000: loss = 0.15 (0.003 sec)\n",
      "Step 45050: loss = 0.15 (0.002 sec)\n",
      "Step 45100: loss = 0.14 (0.003 sec)\n",
      "Step 45150: loss = 0.20 (0.002 sec)\n",
      "Step 45200: loss = 0.16 (0.004 sec)\n",
      "Step 45250: loss = 0.12 (0.002 sec)\n",
      "Step 45300: loss = 0.11 (0.002 sec)\n",
      "Step 45350: loss = 0.14 (0.002 sec)\n",
      "Step 45400: loss = 0.18 (0.002 sec)\n",
      "Step 45450: loss = 0.15 (0.002 sec)\n",
      "Step 45500: loss = 0.16 (0.002 sec)\n",
      "Step 45550: loss = 0.11 (0.002 sec)\n",
      "Step 45600: loss = 0.14 (0.004 sec)\n",
      "Step 45650: loss = 0.12 (0.002 sec)\n",
      "Step 45700: loss = 0.16 (0.002 sec)\n",
      "Step 45750: loss = 0.18 (0.002 sec)\n",
      "Step 45800: loss = 0.12 (0.002 sec)\n",
      "Step 45850: loss = 0.13 (0.003 sec)\n",
      "Step 45900: loss = 0.17 (0.002 sec)\n",
      "Step 45950: loss = 0.13 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 3044  Precision @ 1: 0.9513\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1315  Precision @ 1: 0.8219\n",
      "Step 46000: loss = 0.13 (0.006 sec)\n",
      "Step 46050: loss = 0.15 (0.002 sec)\n",
      "Step 46100: loss = 0.17 (0.002 sec)\n",
      "Step 46150: loss = 0.14 (0.002 sec)\n",
      "Step 46200: loss = 0.17 (0.002 sec)\n",
      "Step 46250: loss = 0.12 (0.002 sec)\n",
      "Step 46300: loss = 0.20 (0.002 sec)\n",
      "Step 46350: loss = 0.16 (0.002 sec)\n",
      "Step 46400: loss = 0.18 (0.004 sec)\n",
      "Step 46450: loss = 0.13 (0.003 sec)\n",
      "Step 46500: loss = 0.18 (0.002 sec)\n",
      "Step 46550: loss = 0.11 (0.003 sec)\n",
      "Step 46600: loss = 0.12 (0.003 sec)\n",
      "Step 46650: loss = 0.10 (0.002 sec)\n",
      "Step 46700: loss = 0.14 (0.002 sec)\n",
      "Step 46750: loss = 0.13 (0.002 sec)\n",
      "Step 46800: loss = 0.10 (0.004 sec)\n",
      "Step 46850: loss = 0.17 (0.002 sec)\n",
      "Step 46900: loss = 0.14 (0.002 sec)\n",
      "Step 46950: loss = 0.15 (0.003 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 3063  Precision @ 1: 0.9572\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1307  Precision @ 1: 0.8169\n",
      "Step 47000: loss = 0.13 (0.002 sec)\n",
      "Step 47050: loss = 0.16 (0.002 sec)\n",
      "Step 47100: loss = 0.15 (0.003 sec)\n",
      "Step 47150: loss = 0.16 (0.002 sec)\n",
      "Step 47200: loss = 0.12 (0.003 sec)\n",
      "Step 47250: loss = 0.16 (0.002 sec)\n",
      "Step 47300: loss = 0.16 (0.002 sec)\n",
      "Step 47350: loss = 0.14 (0.003 sec)\n",
      "Step 47400: loss = 0.11 (0.002 sec)\n",
      "Step 47450: loss = 0.13 (0.002 sec)\n",
      "Step 47500: loss = 0.13 (0.002 sec)\n",
      "Step 47550: loss = 0.16 (0.003 sec)\n",
      "Step 47600: loss = 0.20 (0.003 sec)\n",
      "Step 47650: loss = 0.16 (0.002 sec)\n",
      "Step 47700: loss = 0.22 (0.002 sec)\n",
      "Step 47750: loss = 0.15 (0.002 sec)\n",
      "Step 47800: loss = 0.13 (0.002 sec)\n",
      "Step 47850: loss = 0.13 (0.002 sec)\n",
      "Step 47900: loss = 0.13 (0.002 sec)\n",
      "Step 47950: loss = 0.12 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 3064  Precision @ 1: 0.9575\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1317  Precision @ 1: 0.8231\n",
      "Step 48000: loss = 0.11 (0.006 sec)\n",
      "Step 48050: loss = 0.14 (0.002 sec)\n",
      "Step 48100: loss = 0.11 (0.002 sec)\n",
      "Step 48150: loss = 0.12 (0.002 sec)\n",
      "Step 48200: loss = 0.14 (0.002 sec)\n",
      "Step 48250: loss = 0.20 (0.002 sec)\n",
      "Step 48300: loss = 0.13 (0.002 sec)\n",
      "Step 48350: loss = 0.16 (0.002 sec)\n",
      "Step 48400: loss = 0.17 (0.004 sec)\n",
      "Step 48450: loss = 0.12 (0.002 sec)\n",
      "Step 48500: loss = 0.16 (0.002 sec)\n",
      "Step 48550: loss = 0.18 (0.002 sec)\n",
      "Step 48600: loss = 0.17 (0.002 sec)\n",
      "Step 48650: loss = 0.15 (0.002 sec)\n",
      "Step 48700: loss = 0.14 (0.002 sec)\n",
      "Step 48750: loss = 0.16 (0.002 sec)\n",
      "Step 48800: loss = 0.10 (0.003 sec)\n",
      "Step 48850: loss = 0.17 (0.002 sec)\n",
      "Step 48900: loss = 0.13 (0.003 sec)\n",
      "Step 48950: loss = 0.12 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 3070  Precision @ 1: 0.9594\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1321  Precision @ 1: 0.8256\n",
      "Step 49000: loss = 0.16 (0.003 sec)\n",
      "Step 49050: loss = 0.15 (0.002 sec)\n",
      "Step 49100: loss = 0.13 (0.002 sec)\n",
      "Step 49150: loss = 0.09 (0.002 sec)\n",
      "Step 49200: loss = 0.09 (0.003 sec)\n",
      "Step 49250: loss = 0.09 (0.002 sec)\n",
      "Step 49300: loss = 0.13 (0.002 sec)\n",
      "Step 49350: loss = 0.14 (0.002 sec)\n",
      "Step 49400: loss = 0.08 (0.002 sec)\n",
      "Step 49450: loss = 0.16 (0.002 sec)\n",
      "Step 49500: loss = 0.13 (0.002 sec)\n",
      "Step 49550: loss = 0.09 (0.002 sec)\n",
      "Step 49600: loss = 0.13 (0.003 sec)\n",
      "Step 49650: loss = 0.11 (0.002 sec)\n",
      "Step 49700: loss = 0.12 (0.002 sec)\n",
      "Step 49750: loss = 0.11 (0.002 sec)\n",
      "Step 49800: loss = 0.14 (0.002 sec)\n",
      "Step 49850: loss = 0.15 (0.002 sec)\n",
      "Step 49900: loss = 0.10 (0.002 sec)\n",
      "Step 49950: loss = 0.14 (0.002 sec)\n",
      "Training Data Eval:\n",
      "  Num examples: 3200  Num correct: 3060  Precision @ 1: 0.9563\n",
      "Test Data Eval:\n",
      "  Num examples: 1600  Num correct: 1312  Precision @ 1: 0.8200\n"
     ]
    }
   ],
   "source": [
    "run_training()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}