caffe_multiclass_hdf5.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Imports"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "import caffe\n",
    "import caffe.draw\n",
    "from caffe.proto import caffe_pb2\n",
    "from google.protobuf import text_format\n",
    "import subprocess"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Print net"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "name: \"multiclass_hdf5\"\n",
      "\n",
      "layer {\n",
      "  name: \"data_hdf5\"\n",
      "  type: \"HDF5Data\"\n",
      "  include {\n",
      "    phase: TRAIN\n",
      "  }\n",
      "  hdf5_data_param {\n",
      "    source: \"multiclass_caffe.hdf5.txt\"\n",
      "    batch_size: 10\n",
      "  }\n",
      "  top: \"data\"\n",
      "  top: \"label\"\n",
      "}\n",
      "\n",
      "layer {\n",
      "  name: \"data_hdf5\"\n",
      "  type: \"HDF5Data\"\n",
      "  include {\n",
      "    phase: TEST\n",
      "  }\n",
      "  hdf5_data_param {\n",
      "    source: \"multiclass_caffe_test.hdf5.txt\"\n",
      "    batch_size: 10\n",
      "  }\n",
      "  top: \"data\"\n",
      "  top: \"label\"\n",
      "}\n",
      "\n",
      "layer {\n",
      "  name: \"hidden_layer01\"\n",
      "  type: \"InnerProduct\"\n",
      "  inner_product_param {\n",
      "    num_output: 20\n",
      "    weight_filler {\n",
      "      type: \"xavier\"\n",
      "    }\n",
      "    bias_filler {\n",
      "      type: \"constant\"\n",
      "    }\n",
      "  }\n",
      "  bottom: \"data\"\n",
      "  top: \"hl01\"\n",
      "}\n",
      "\n",
      "layer {\n",
      "  name: \"relu01\"\n",
      "  type: \"ReLU\"\n",
      "  bottom: \"hl01\"\n",
      "  top: \"r01\"\n",
      "}\n",
      "\n",
      "layer {\n",
      "  name: \"hidden_layer02\"\n",
      "  type: \"InnerProduct\"\n",
      "  inner_product_param {\n",
      "    num_output: 6\n",
      "    weight_filler {\n",
      "      type: \"xavier\"\n",
      "    }\n",
      "    bias_filler {\n",
      "      type: \"constant\"\n",
      "    }\n",
      "  }\n",
      "  bottom: \"r01\"\n",
      "  top: \"hl02\"\n",
      "}\n",
      "\n",
      "layer {\n",
      "  name: \"accuracy\"\n",
      "  type: \"Accuracy\"\n",
      "  bottom: \"hl02\"\n",
      "  bottom: \"label\"\n",
      "  top: \"accuracy\"\n",
      "  include {\n",
      "    phase: TEST\n",
      "  }\n",
      "\n",
      "}\n",
      "layer {\n",
      "  name: \"loss\"\n",
      "  type: \"SoftmaxWithLoss\"\n",
      "  bottom: \"hl02\"\n",
      "  bottom: \"label\"\n",
      "  top: \"loss\"\n",
      "}\n",
      "\n"
     ]
    }
   ],
   "source": [
    "with open(\"multiclass_hdf5.prototxt\", 'r') as f: print f.read()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Draw net"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "net = caffe_pb2.NetParameter()\n",
    "text_format.Merge(open(\"multiclass_hdf5.prototxt\").read(), net)\n",
    "\n",
    "caffe.draw.draw_net_to_file(net, \"multiclass_hdf5.png\", \"TB\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "![](multiclass_hdf5.png)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Train net"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "/data/soft/caffe/build/tools/caffe: /data/soft/anaconda2/lib/libtiff.so.5: no version information available (required by /usr/lib/x86_64-linux-gnu/libopencv_highgui.so.2.4)\n",
      "/data/soft/caffe/build/tools/caffe: /data/soft/anaconda2/lib/liblzma.so.5: no version information available (required by /usr/lib/x86_64-linux-gnu/libunwind.so.8)\n",
      "/data/soft/caffe/build/tools/caffe: /data/soft/anaconda2/lib/liblzma.so.5: no version information available (required by /usr/lib/x86_64-linux-gnu/libavcodec-ffmpeg.so.56)\n",
      "I0419 11:17:58.279649 15680 caffe.cpp:178] Use CPU.\n",
      "I0419 11:17:58.279850 15680 solver.cpp:48] Initializing solver from parameters: \n",
      "test_iter: 100\n",
      "test_interval: 500\n",
      "base_lr: 0.01\n",
      "display: 100\n",
      "max_iter: 4000\n",
      "lr_policy: \"fixed\"\n",
      "momentum: 0.9\n",
      "weight_decay: 0.004\n",
      "snapshot: 4000\n",
      "snapshot_prefix: \"multiclass_hdf5\"\n",
      "solver_mode: CPU\n",
      "net: \"multiclass_hdf5.prototxt\"\n",
      "I0419 11:17:58.279952 15680 solver.cpp:91] Creating training net from net file: multiclass_hdf5.prototxt\n",
      "I0419 11:17:58.280109 15680 net.cpp:313] The NetState phase (0) differed from the phase (1) specified by a rule in layer data_hdf5\n",
      "I0419 11:17:58.280139 15680 net.cpp:313] The NetState phase (0) differed from the phase (1) specified by a rule in layer accuracy\n",
      "I0419 11:17:58.280212 15680 net.cpp:49] Initializing net from parameters: \n",
      "name: \"multiclass_hdf5\"\n",
      "state {\n",
      "  phase: TRAIN\n",
      "}\n",
      "layer {\n",
      "  name: \"data_hdf5\"\n",
      "  type: \"HDF5Data\"\n",
      "  top: \"data\"\n",
      "  top: \"label\"\n",
      "  include {\n",
      "    phase: TRAIN\n",
      "  }\n",
      "  hdf5_data_param {\n",
      "    source: \"multiclass_caffe.hdf5.txt\"\n",
      "    batch_size: 10\n",
      "  }\n",
      "}\n",
      "layer {\n",
      "  name: \"hidden_layer01\"\n",
      "  type: \"InnerProduct\"\n",
      "  bottom: \"data\"\n",
      "  top: \"hl01\"\n",
      "  inner_product_param {\n",
      "    num_output: 6\n",
      "    weight_filler {\n",
      "      type: \"xavier\"\n",
      "    }\n",
      "    bias_filler {\n",
      "      type: \"constant\"\n",
      "    }\n",
      "  }\n",
      "}\n",
      "layer {\n",
      "  name: \"relu01\"\n",
      "  type: \"ReLU\"\n",
      "  bottom: \"hl01\"\n",
      "  top: \"r01\"\n",
      "}\n",
      "layer {\n",
      "  name: \"hidden_layer02\"\n",
      "  type: \"InnerProduct\"\n",
      "  bottom: \"r01\"\n",
      "  top: \"hl02\"\n",
      "  inner_product_param {\n",
      "    num_output: 6\n",
      "    weight_filler {\n",
      "      type: \"xavier\"\n",
      "    }\n",
      "    bias_filler {\n",
      "      type: \"constant\"\n",
      "    }\n",
      "  }\n",
      "}\n",
      "layer {\n",
      "  name: \"loss\"\n",
      "  type: \"SoftmaxWithLoss\"\n",
      "  bottom: \"hl02\"\n",
      "  bottom: \"label\"\n",
      "  top: \"loss\"\n",
      "}\n",
      "I0419 11:17:58.280277 15680 layer_factory.hpp:77] Creating layer data_hdf5\n",
      "I0419 11:17:58.280298 15680 net.cpp:91] Creating Layer data_hdf5\n",
      "I0419 11:17:58.280308 15680 net.cpp:399] data_hdf5 -> data\n",
      "I0419 11:17:58.280328 15680 net.cpp:399] data_hdf5 -> label\n",
      "I0419 11:17:58.280336 15680 hdf5_data_layer.cpp:79] Loading list of HDF5 filenames from: multiclass_caffe.hdf5.txt\n",
      "I0419 11:17:58.280359 15680 hdf5_data_layer.cpp:93] Number of HDF5 files: 1\n",
      "I0419 11:17:58.281669 15680 hdf5.cpp:32] Datatype class: H5T_FLOAT\n",
      "I0419 11:17:58.281793 15680 hdf5.cpp:35] Datatype class: H5T_INTEGER\n",
      "I0419 11:17:58.282029 15680 net.cpp:141] Setting up data_hdf5\n",
      "I0419 11:17:58.282053 15680 net.cpp:148] Top shape: 10 2 (20)\n",
      "I0419 11:17:58.282059 15680 net.cpp:148] Top shape: 10 (10)\n",
      "I0419 11:17:58.282064 15680 net.cpp:156] Memory required for data: 120\n",
      "I0419 11:17:58.282071 15680 layer_factory.hpp:77] Creating layer hidden_layer01\n",
      "I0419 11:17:58.282084 15680 net.cpp:91] Creating Layer hidden_layer01\n",
      "I0419 11:17:58.282091 15680 net.cpp:425] hidden_layer01 <- data\n",
      "I0419 11:17:58.282104 15680 net.cpp:399] hidden_layer01 -> hl01\n",
      "I0419 11:17:58.282624 15680 net.cpp:141] Setting up hidden_layer01\n",
      "I0419 11:17:58.282645 15680 net.cpp:148] Top shape: 10 6 (60)\n",
      "I0419 11:17:58.282650 15680 net.cpp:156] Memory required for data: 360\n",
      "I0419 11:17:58.282673 15680 layer_factory.hpp:77] Creating layer relu01\n",
      "I0419 11:17:58.282685 15680 net.cpp:91] Creating Layer relu01\n",
      "I0419 11:17:58.282691 15680 net.cpp:425] relu01 <- hl01\n",
      "I0419 11:17:58.282698 15680 net.cpp:399] relu01 -> r01\n",
      "I0419 11:17:58.282711 15680 net.cpp:141] Setting up relu01\n",
      "I0419 11:17:58.282717 15680 net.cpp:148] Top shape: 10 6 (60)\n",
      "I0419 11:17:58.282721 15680 net.cpp:156] Memory required for data: 600\n",
      "I0419 11:17:58.282727 15680 layer_factory.hpp:77] Creating layer hidden_layer02\n",
      "I0419 11:17:58.282733 15680 net.cpp:91] Creating Layer hidden_layer02\n",
      "I0419 11:17:58.282737 15680 net.cpp:425] hidden_layer02 <- r01\n",
      "I0419 11:17:58.282747 15680 net.cpp:399] hidden_layer02 -> hl02\n",
      "I0419 11:17:58.282763 15680 net.cpp:141] Setting up hidden_layer02\n",
      "I0419 11:17:58.282769 15680 net.cpp:148] Top shape: 10 6 (60)\n",
      "I0419 11:17:58.282773 15680 net.cpp:156] Memory required for data: 840\n",
      "I0419 11:17:58.282783 15680 layer_factory.hpp:77] Creating layer loss\n",
      "I0419 11:17:58.282793 15680 net.cpp:91] Creating Layer loss\n",
      "I0419 11:17:58.282798 15680 net.cpp:425] loss <- hl02\n",
      "I0419 11:17:58.282804 15680 net.cpp:425] loss <- label\n",
      "I0419 11:17:58.282811 15680 net.cpp:399] loss -> loss\n",
      "I0419 11:17:58.282847 15680 layer_factory.hpp:77] Creating layer loss\n",
      "I0419 11:17:58.282866 15680 net.cpp:141] Setting up loss\n",
      "I0419 11:17:58.282874 15680 net.cpp:148] Top shape: (1)\n",
      "I0419 11:17:58.282878 15680 net.cpp:151]     with loss weight 1\n",
      "I0419 11:17:58.282892 15680 net.cpp:156] Memory required for data: 844\n",
      "I0419 11:17:58.282898 15680 net.cpp:217] loss needs backward computation.\n",
      "I0419 11:17:58.282907 15680 net.cpp:217] hidden_layer02 needs backward computation.\n",
      "I0419 11:17:58.282912 15680 net.cpp:217] relu01 needs backward computation.\n",
      "I0419 11:17:58.282917 15680 net.cpp:217] hidden_layer01 needs backward computation.\n",
      "I0419 11:17:58.282922 15680 net.cpp:219] data_hdf5 does not need backward computation.\n",
      "I0419 11:17:58.282927 15680 net.cpp:261] This network produces output loss\n",
      "I0419 11:17:58.282937 15680 net.cpp:274] Network initialization done.\n",
      "I0419 11:17:58.283059 15680 solver.cpp:181] Creating test net (#0) specified by net file: multiclass_hdf5.prototxt\n",
      "I0419 11:17:58.283082 15680 net.cpp:313] The NetState phase (1) differed from the phase (0) specified by a rule in layer data_hdf5\n",
      "I0419 11:17:58.283143 15680 net.cpp:49] Initializing net from parameters: \n",
      "name: \"multiclass_hdf5\"\n",
      "state {\n",
      "  phase: TEST\n",
      "}\n",
      "layer {\n",
      "  name: \"data_hdf5\"\n",
      "  type: \"HDF5Data\"\n",
      "  top: \"data\"\n",
      "  top: \"label\"\n",
      "  include {\n",
      "    phase: TEST\n",
      "  }\n",
      "  hdf5_data_param {\n",
      "    source: \"multiclass_caffe_test.hdf5.txt\"\n",
      "    batch_size: 10\n",
      "  }\n",
      "}\n",
      "layer {\n",
      "  name: \"hidden_layer01\"\n",
      "  type: \"InnerProduct\"\n",
      "  bottom: \"data\"\n",
      "  top: \"hl01\"\n",
      "  inner_product_param {\n",
      "    num_output: 6\n",
      "    weight_filler {\n",
      "      type: \"xavier\"\n",
      "    }\n",
      "    bias_filler {\n",
      "      type: \"constant\"\n",
      "    }\n",
      "  }\n",
      "}\n",
      "layer {\n",
      "  name: \"relu01\"\n",
      "  type: \"ReLU\"\n",
      "  bottom: \"hl01\"\n",
      "  top: \"r01\"\n",
      "}\n",
      "layer {\n",
      "  name: \"hidden_layer02\"\n",
      "  type: \"InnerProduct\"\n",
      "  bottom: \"r01\"\n",
      "  top: \"hl02\"\n",
      "  inner_product_param {\n",
      "    num_output: 6\n",
      "    weight_filler {\n",
      "      type: \"xavier\"\n",
      "    }\n",
      "    bias_filler {\n",
      "      type: \"constant\"\n",
      "    }\n",
      "  }\n",
      "}\n",
      "layer {\n",
      "  name: \"accuracy\"\n",
      "  type: \"Accuracy\"\n",
      "  bottom: \"hl02\"\n",
      "  bottom: \"label\"\n",
      "  top: \"accuracy\"\n",
      "  include {\n",
      "    phase: TEST\n",
      "  }\n",
      "}\n",
      "layer {\n",
      "  name: \"loss\"\n",
      "  type: \"SoftmaxWithLoss\"\n",
      "  bottom: \"hl02\"\n",
      "  bottom: \"label\"\n",
      "  top: \"loss\"\n",
      "}\n",
      "I0419 11:17:58.283212 15680 layer_factory.hpp:77] Creating layer data_hdf5\n",
      "I0419 11:17:58.283222 15680 net.cpp:91] Creating Layer data_hdf5\n",
      "I0419 11:17:58.283229 15680 net.cpp:399] data_hdf5 -> data\n",
      "I0419 11:17:58.283239 15680 net.cpp:399] data_hdf5 -> label\n",
      "I0419 11:17:58.283249 15680 hdf5_data_layer.cpp:79] Loading list of HDF5 filenames from: multiclass_caffe_test.hdf5.txt\n",
      "I0419 11:17:58.283268 15680 hdf5_data_layer.cpp:93] Number of HDF5 files: 1\n",
      "I0419 11:17:58.283838 15680 net.cpp:141] Setting up data_hdf5\n",
      "I0419 11:17:58.283860 15680 net.cpp:148] Top shape: 10 2 (20)\n",
      "I0419 11:17:58.283872 15680 net.cpp:148] Top shape: 10 (10)\n",
      "I0419 11:17:58.283881 15680 net.cpp:156] Memory required for data: 120\n",
      "I0419 11:17:58.283890 15680 layer_factory.hpp:77] Creating layer label_data_hdf5_1_split\n",
      "I0419 11:17:58.283902 15680 net.cpp:91] Creating Layer label_data_hdf5_1_split\n",
      "I0419 11:17:58.283911 15680 net.cpp:425] label_data_hdf5_1_split <- label\n",
      "I0419 11:17:58.283922 15680 net.cpp:399] label_data_hdf5_1_split -> label_data_hdf5_1_split_0\n",
      "I0419 11:17:58.283948 15680 net.cpp:399] label_data_hdf5_1_split -> label_data_hdf5_1_split_1\n",
      "I0419 11:17:58.283965 15680 net.cpp:141] Setting up label_data_hdf5_1_split\n",
      "I0419 11:17:58.283975 15680 net.cpp:148] Top shape: 10 (10)\n",
      "I0419 11:17:58.283985 15680 net.cpp:148] Top shape: 10 (10)\n",
      "I0419 11:17:58.283993 15680 net.cpp:156] Memory required for data: 200\n",
      "I0419 11:17:58.284000 15680 layer_factory.hpp:77] Creating layer hidden_layer01\n",
      "I0419 11:17:58.284013 15680 net.cpp:91] Creating Layer hidden_layer01\n",
      "I0419 11:17:58.284020 15680 net.cpp:425] hidden_layer01 <- data\n",
      "I0419 11:17:58.284031 15680 net.cpp:399] hidden_layer01 -> hl01\n",
      "I0419 11:17:58.284054 15680 net.cpp:141] Setting up hidden_layer01\n",
      "I0419 11:17:58.284063 15680 net.cpp:148] Top shape: 10 6 (60)\n",
      "I0419 11:17:58.284068 15680 net.cpp:156] Memory required for data: 440\n",
      "I0419 11:17:58.284103 15680 layer_factory.hpp:77] Creating layer relu01\n",
      "I0419 11:17:58.284112 15680 net.cpp:91] Creating Layer relu01\n",
      "I0419 11:17:58.284118 15680 net.cpp:425] relu01 <- hl01\n",
      "I0419 11:17:58.284123 15680 net.cpp:399] relu01 -> r01\n",
      "I0419 11:17:58.284133 15680 net.cpp:141] Setting up relu01\n",
      "I0419 11:17:58.284140 15680 net.cpp:148] Top shape: 10 6 (60)\n",
      "I0419 11:17:58.284145 15680 net.cpp:156] Memory required for data: 680\n",
      "I0419 11:17:58.284150 15680 layer_factory.hpp:77] Creating layer hidden_layer02\n",
      "I0419 11:17:58.284160 15680 net.cpp:91] Creating Layer hidden_layer02\n",
      "I0419 11:17:58.284165 15680 net.cpp:425] hidden_layer02 <- r01\n",
      "I0419 11:17:58.284173 15680 net.cpp:399] hidden_layer02 -> hl02\n",
      "I0419 11:17:58.284188 15680 net.cpp:141] Setting up hidden_layer02\n",
      "I0419 11:17:58.284196 15680 net.cpp:148] Top shape: 10 6 (60)\n",
      "I0419 11:17:58.284200 15680 net.cpp:156] Memory required for data: 920\n",
      "I0419 11:17:58.284210 15680 layer_factory.hpp:77] Creating layer hl02_hidden_layer02_0_split\n",
      "I0419 11:17:58.284219 15680 net.cpp:91] Creating Layer hl02_hidden_layer02_0_split\n",
      "I0419 11:17:58.284224 15680 net.cpp:425] hl02_hidden_layer02_0_split <- hl02\n",
      "I0419 11:17:58.284231 15680 net.cpp:399] hl02_hidden_layer02_0_split -> hl02_hidden_layer02_0_split_0\n",
      "I0419 11:17:58.284240 15680 net.cpp:399] hl02_hidden_layer02_0_split -> hl02_hidden_layer02_0_split_1\n",
      "I0419 11:17:58.284251 15680 net.cpp:141] Setting up hl02_hidden_layer02_0_split\n",
      "I0419 11:17:58.284258 15680 net.cpp:148] Top shape: 10 6 (60)\n",
      "I0419 11:17:58.284265 15680 net.cpp:148] Top shape: 10 6 (60)\n",
      "I0419 11:17:58.284270 15680 net.cpp:156] Memory required for data: 1400\n",
      "I0419 11:17:58.284274 15680 layer_factory.hpp:77] Creating layer accuracy\n",
      "I0419 11:17:58.284282 15680 net.cpp:91] Creating Layer accuracy\n",
      "I0419 11:17:58.284287 15680 net.cpp:425] accuracy <- hl02_hidden_layer02_0_split_0\n",
      "I0419 11:17:58.284293 15680 net.cpp:425] accuracy <- label_data_hdf5_1_split_0\n",
      "I0419 11:17:58.284301 15680 net.cpp:399] accuracy -> accuracy\n",
      "I0419 11:17:58.284312 15680 net.cpp:141] Setting up accuracy\n",
      "I0419 11:17:58.284317 15680 net.cpp:148] Top shape: (1)\n",
      "I0419 11:17:58.284322 15680 net.cpp:156] Memory required for data: 1404\n",
      "I0419 11:17:58.284327 15680 layer_factory.hpp:77] Creating layer loss\n",
      "I0419 11:17:58.284334 15680 net.cpp:91] Creating Layer loss\n",
      "I0419 11:17:58.284339 15680 net.cpp:425] loss <- hl02_hidden_layer02_0_split_1\n",
      "I0419 11:17:58.284345 15680 net.cpp:425] loss <- label_data_hdf5_1_split_1\n",
      "I0419 11:17:58.284353 15680 net.cpp:399] loss -> loss\n",
      "I0419 11:17:58.284361 15680 layer_factory.hpp:77] Creating layer loss\n",
      "I0419 11:17:58.284376 15680 net.cpp:141] Setting up loss\n",
      "I0419 11:17:58.284384 15680 net.cpp:148] Top shape: (1)\n",
      "I0419 11:17:58.284387 15680 net.cpp:151]     with loss weight 1\n",
      "I0419 11:17:58.284396 15680 net.cpp:156] Memory required for data: 1408\n",
      "I0419 11:17:58.284401 15680 net.cpp:217] loss needs backward computation.\n",
      "I0419 11:17:58.284409 15680 net.cpp:219] accuracy does not need backward computation.\n",
      "I0419 11:17:58.284421 15680 net.cpp:217] hl02_hidden_layer02_0_split needs backward computation.\n",
      "I0419 11:17:58.284430 15680 net.cpp:217] hidden_layer02 needs backward computation.\n",
      "I0419 11:17:58.284438 15680 net.cpp:217] relu01 needs backward computation.\n",
      "I0419 11:17:58.284446 15680 net.cpp:217] hidden_layer01 needs backward computation.\n",
      "I0419 11:17:58.284454 15680 net.cpp:219] label_data_hdf5_1_split does not need backward computation.\n",
      "I0419 11:17:58.284463 15680 net.cpp:219] data_hdf5 does not need backward computation.\n",
      "I0419 11:17:58.284471 15680 net.cpp:261] This network produces output accuracy\n",
      "I0419 11:17:58.284478 15680 net.cpp:261] This network produces output loss\n",
      "I0419 11:17:58.284495 15680 net.cpp:274] Network initialization done.\n",
      "I0419 11:17:58.284534 15680 solver.cpp:60] Solver scaffolding done.\n",
      "I0419 11:17:58.284556 15680 caffe.cpp:219] Starting Optimization\n",
      "I0419 11:17:58.284571 15680 solver.cpp:279] Solving multiclass_hdf5\n",
      "I0419 11:17:58.284579 15680 solver.cpp:280] Learning Rate Policy: fixed\n",
      "I0419 11:17:58.284593 15680 solver.cpp:337] Iteration 0, Testing net (#0)\n",
      "I0419 11:17:58.287269 15680 solver.cpp:404]     Test net output #0: accuracy = 0.13\n",
      "I0419 11:17:58.287303 15680 solver.cpp:404]     Test net output #1: loss = 1.77034 (* 1 = 1.77034 loss)\n",
      "I0419 11:17:58.287369 15680 solver.cpp:228] Iteration 0, loss = 1.84579\n",
      "I0419 11:17:58.287392 15680 solver.cpp:244]     Train net output #0: loss = 1.84579 (* 1 = 1.84579 loss)\n",
      "I0419 11:17:58.287400 15680 sgd_solver.cpp:106] Iteration 0, lr = 0.01\n",
      "I0419 11:17:58.289834 15680 solver.cpp:228] Iteration 100, loss = 1.50241\n",
      "I0419 11:17:58.289855 15680 solver.cpp:244]     Train net output #0: loss = 1.50241 (* 1 = 1.50241 loss)\n",
      "I0419 11:17:58.289862 15680 sgd_solver.cpp:106] Iteration 100, lr = 0.01\n",
      "I0419 11:17:58.292460 15680 solver.cpp:228] Iteration 200, loss = 1.16122\n",
      "I0419 11:17:58.292489 15680 solver.cpp:244]     Train net output #0: loss = 1.16122 (* 1 = 1.16122 loss)\n",
      "I0419 11:17:58.292497 15680 sgd_solver.cpp:106] Iteration 200, lr = 0.01\n",
      "I0419 11:17:58.295024 15680 solver.cpp:228] Iteration 300, loss = 0.863732\n",
      "I0419 11:17:58.295047 15680 solver.cpp:244]     Train net output #0: loss = 0.863732 (* 1 = 0.863732 loss)\n",
      "I0419 11:17:58.295053 15680 sgd_solver.cpp:106] Iteration 300, lr = 0.01\n",
      "I0419 11:17:58.296537 15680 solver.cpp:228] Iteration 400, loss = 0.687389\n",
      "I0419 11:17:58.296548 15680 solver.cpp:244]     Train net output #0: loss = 0.687389 (* 1 = 0.687389 loss)\n",
      "I0419 11:17:58.296561 15680 sgd_solver.cpp:106] Iteration 400, lr = 0.01\n",
      "I0419 11:17:58.297930 15680 solver.cpp:337] Iteration 500, Testing net (#0)\n",
      "I0419 11:17:58.299322 15680 solver.cpp:404]     Test net output #0: accuracy = 0.795\n",
      "I0419 11:17:58.299329 15680 solver.cpp:404]     Test net output #1: loss = 0.742539 (* 1 = 0.742539 loss)\n",
      "I0419 11:17:58.299366 15680 solver.cpp:228] Iteration 500, loss = 0.579946\n",
      "I0419 11:17:58.299371 15680 solver.cpp:244]     Train net output #0: loss = 0.579946 (* 1 = 0.579946 loss)\n",
      "I0419 11:17:58.299384 15680 sgd_solver.cpp:106] Iteration 500, lr = 0.01\n",
      "I0419 11:17:58.300743 15680 solver.cpp:228] Iteration 600, loss = 0.508246\n",
      "I0419 11:17:58.300748 15680 solver.cpp:244]     Train net output #0: loss = 0.508246 (* 1 = 0.508246 loss)\n",
      "I0419 11:17:58.300762 15680 sgd_solver.cpp:106] Iteration 600, lr = 0.01\n",
      "I0419 11:17:58.302146 15680 solver.cpp:228] Iteration 700, loss = 0.458491\n",
      "I0419 11:17:58.302153 15680 solver.cpp:244]     Train net output #0: loss = 0.458491 (* 1 = 0.458491 loss)\n",
      "I0419 11:17:58.302166 15680 sgd_solver.cpp:106] Iteration 700, lr = 0.01\n",
      "I0419 11:17:58.303524 15680 solver.cpp:228] Iteration 800, loss = 0.421955\n",
      "I0419 11:17:58.303530 15680 solver.cpp:244]     Train net output #0: loss = 0.421955 (* 1 = 0.421955 loss)\n",
      "I0419 11:17:58.303544 15680 sgd_solver.cpp:106] Iteration 800, lr = 0.01\n",
      "I0419 11:17:58.304929 15680 solver.cpp:228] Iteration 900, loss = 0.394106\n",
      "I0419 11:17:58.304935 15680 solver.cpp:244]     Train net output #0: loss = 0.394106 (* 1 = 0.394106 loss)\n",
      "I0419 11:17:58.304939 15680 sgd_solver.cpp:106] Iteration 900, lr = 0.01\n",
      "I0419 11:17:58.306354 15680 solver.cpp:337] Iteration 1000, Testing net (#0)\n",
      "I0419 11:17:58.307798 15680 solver.cpp:404]     Test net output #0: accuracy = 0.852\n",
      "I0419 11:17:58.307806 15680 solver.cpp:404]     Test net output #1: loss = 0.539963 (* 1 = 0.539963 loss)\n",
      "I0419 11:17:58.307835 15680 solver.cpp:228] Iteration 1000, loss = 0.372439\n",
      "I0419 11:17:58.307842 15680 solver.cpp:244]     Train net output #0: loss = 0.372439 (* 1 = 0.372439 loss)\n",
      "I0419 11:17:58.307854 15680 sgd_solver.cpp:106] Iteration 1000, lr = 0.01\n",
      "I0419 11:17:58.309281 15680 solver.cpp:228] Iteration 1100, loss = 0.355192\n",
      "I0419 11:17:58.309286 15680 solver.cpp:244]     Train net output #0: loss = 0.355192 (* 1 = 0.355192 loss)\n",
      "I0419 11:17:58.309300 15680 sgd_solver.cpp:106] Iteration 1100, lr = 0.01\n",
      "I0419 11:17:58.310688 15680 solver.cpp:228] Iteration 1200, loss = 0.341336\n",
      "I0419 11:17:58.310693 15680 solver.cpp:244]     Train net output #0: loss = 0.341336 (* 1 = 0.341336 loss)\n",
      "I0419 11:17:58.310708 15680 sgd_solver.cpp:106] Iteration 1200, lr = 0.01\n",
      "I0419 11:17:58.312086 15680 solver.cpp:228] Iteration 1300, loss = 0.3301\n",
      "I0419 11:17:58.312110 15680 solver.cpp:244]     Train net output #0: loss = 0.3301 (* 1 = 0.3301 loss)\n",
      "I0419 11:17:58.312124 15680 sgd_solver.cpp:106] Iteration 1300, lr = 0.01\n",
      "I0419 11:17:58.313483 15680 solver.cpp:228] Iteration 1400, loss = 0.320714\n",
      "I0419 11:17:58.313489 15680 solver.cpp:244]     Train net output #0: loss = 0.320714 (* 1 = 0.320714 loss)\n",
      "I0419 11:17:58.313503 15680 sgd_solver.cpp:106] Iteration 1400, lr = 0.01\n",
      "I0419 11:17:58.314853 15680 solver.cpp:337] Iteration 1500, Testing net (#0)\n",
      "I0419 11:17:58.316249 15680 solver.cpp:404]     Test net output #0: accuracy = 0.864\n",
      "I0419 11:17:58.316256 15680 solver.cpp:404]     Test net output #1: loss = 0.478197 (* 1 = 0.478197 loss)\n",
      "I0419 11:17:58.316284 15680 solver.cpp:228] Iteration 1500, loss = 0.313002\n",
      "I0419 11:17:58.316289 15680 solver.cpp:244]     Train net output #0: loss = 0.313002 (* 1 = 0.313002 loss)\n",
      "I0419 11:17:58.316303 15680 sgd_solver.cpp:106] Iteration 1500, lr = 0.01\n",
      "I0419 11:17:58.317791 15680 solver.cpp:228] Iteration 1600, loss = 0.306575\n",
      "I0419 11:17:58.317819 15680 solver.cpp:244]     Train net output #0: loss = 0.306575 (* 1 = 0.306575 loss)\n",
      "I0419 11:17:58.317826 15680 sgd_solver.cpp:106] Iteration 1600, lr = 0.01\n",
      "I0419 11:17:58.319290 15680 solver.cpp:228] Iteration 1700, loss = 0.301454\n",
      "I0419 11:17:58.319299 15680 solver.cpp:244]     Train net output #0: loss = 0.301454 (* 1 = 0.301454 loss)\n",
      "I0419 11:17:58.319306 15680 sgd_solver.cpp:106] Iteration 1700, lr = 0.01\n",
      "I0419 11:17:58.320708 15680 solver.cpp:228] Iteration 1800, loss = 0.296637\n",
      "I0419 11:17:58.320715 15680 solver.cpp:244]     Train net output #0: loss = 0.296637 (* 1 = 0.296637 loss)\n",
      "I0419 11:17:58.320721 15680 sgd_solver.cpp:106] Iteration 1800, lr = 0.01\n",
      "I0419 11:17:58.322098 15680 solver.cpp:228] Iteration 1900, loss = 0.29287\n",
      "I0419 11:17:58.322106 15680 solver.cpp:244]     Train net output #0: loss = 0.29287 (* 1 = 0.29287 loss)\n",
      "I0419 11:17:58.322113 15680 sgd_solver.cpp:106] Iteration 1900, lr = 0.01\n",
      "I0419 11:17:58.323475 15680 solver.cpp:337] Iteration 2000, Testing net (#0)\n",
      "I0419 11:17:58.324918 15680 solver.cpp:404]     Test net output #0: accuracy = 0.869\n",
      "I0419 11:17:58.324928 15680 solver.cpp:404]     Test net output #1: loss = 0.452761 (* 1 = 0.452761 loss)\n",
      "I0419 11:17:58.324949 15680 solver.cpp:228] Iteration 2000, loss = 0.289254\n",
      "I0419 11:17:58.324956 15680 solver.cpp:244]     Train net output #0: loss = 0.289254 (* 1 = 0.289254 loss)\n",
      "I0419 11:17:58.324961 15680 sgd_solver.cpp:106] Iteration 2000, lr = 0.01\n",
      "I0419 11:17:58.326380 15680 solver.cpp:228] Iteration 2100, loss = 0.286309\n",
      "I0419 11:17:58.326388 15680 solver.cpp:244]     Train net output #0: loss = 0.286309 (* 1 = 0.286309 loss)\n",
      "I0419 11:17:58.326393 15680 sgd_solver.cpp:106] Iteration 2100, lr = 0.01\n",
      "I0419 11:17:58.327744 15680 solver.cpp:228] Iteration 2200, loss = 0.284148\n",
      "I0419 11:17:58.327751 15680 solver.cpp:244]     Train net output #0: loss = 0.284148 (* 1 = 0.284148 loss)\n",
      "I0419 11:17:58.327755 15680 sgd_solver.cpp:106] Iteration 2200, lr = 0.01\n",
      "I0419 11:17:58.329097 15680 solver.cpp:228] Iteration 2300, loss = 0.281603\n",
      "I0419 11:17:58.329105 15680 solver.cpp:244]     Train net output #0: loss = 0.281603 (* 1 = 0.281603 loss)\n",
      "I0419 11:17:58.329110 15680 sgd_solver.cpp:106] Iteration 2300, lr = 0.01\n",
      "I0419 11:17:58.330488 15680 solver.cpp:228] Iteration 2400, loss = 0.279773\n",
      "I0419 11:17:58.330497 15680 solver.cpp:244]     Train net output #0: loss = 0.279773 (* 1 = 0.279773 loss)\n",
      "I0419 11:17:58.330502 15680 sgd_solver.cpp:106] Iteration 2400, lr = 0.01\n",
      "I0419 11:17:58.331878 15680 solver.cpp:337] Iteration 2500, Testing net (#0)\n",
      "I0419 11:17:58.333276 15680 solver.cpp:404]     Test net output #0: accuracy = 0.873\n",
      "I0419 11:17:58.333284 15680 solver.cpp:404]     Test net output #1: loss = 0.440488 (* 1 = 0.440488 loss)\n",
      "I0419 11:17:58.333305 15680 solver.cpp:228] Iteration 2500, loss = 0.27818\n",
      "I0419 11:17:58.333312 15680 solver.cpp:244]     Train net output #0: loss = 0.27818 (* 1 = 0.27818 loss)\n",
      "I0419 11:17:58.333318 15680 sgd_solver.cpp:106] Iteration 2500, lr = 0.01\n",
      "I0419 11:17:58.334707 15680 solver.cpp:228] Iteration 2600, loss = 0.276987\n",
      "I0419 11:17:58.334715 15680 solver.cpp:244]     Train net output #0: loss = 0.276987 (* 1 = 0.276987 loss)\n",
      "I0419 11:17:58.334722 15680 sgd_solver.cpp:106] Iteration 2600, lr = 0.01\n",
      "I0419 11:17:58.336071 15680 solver.cpp:228] Iteration 2700, loss = 0.275279\n",
      "I0419 11:17:58.336077 15680 solver.cpp:244]     Train net output #0: loss = 0.275279 (* 1 = 0.275279 loss)\n",
      "I0419 11:17:58.336082 15680 sgd_solver.cpp:106] Iteration 2700, lr = 0.01\n",
      "I0419 11:17:58.337426 15680 solver.cpp:228] Iteration 2800, loss = 0.274181\n",
      "I0419 11:17:58.337435 15680 solver.cpp:244]     Train net output #0: loss = 0.274181 (* 1 = 0.274181 loss)\n",
      "I0419 11:17:58.337438 15680 sgd_solver.cpp:106] Iteration 2800, lr = 0.01\n",
      "I0419 11:17:58.338810 15680 solver.cpp:228] Iteration 2900, loss = 0.27318\n",
      "I0419 11:17:58.338819 15680 solver.cpp:244]     Train net output #0: loss = 0.27318 (* 1 = 0.27318 loss)\n",
      "I0419 11:17:58.338824 15680 sgd_solver.cpp:106] Iteration 2900, lr = 0.01\n",
      "I0419 11:17:58.340158 15680 solver.cpp:337] Iteration 3000, Testing net (#0)\n",
      "I0419 11:17:58.341559 15680 solver.cpp:404]     Test net output #0: accuracy = 0.877\n",
      "I0419 11:17:58.341567 15680 solver.cpp:404]     Test net output #1: loss = 0.434081 (* 1 = 0.434081 loss)\n",
      "I0419 11:17:58.341589 15680 solver.cpp:228] Iteration 3000, loss = 0.272267\n",
      "I0419 11:17:58.341596 15680 solver.cpp:244]     Train net output #0: loss = 0.272267 (* 1 = 0.272267 loss)\n",
      "I0419 11:17:58.341601 15680 sgd_solver.cpp:106] Iteration 3000, lr = 0.01\n",
      "I0419 11:17:58.343116 15680 solver.cpp:228] Iteration 3100, loss = 0.271455\n",
      "I0419 11:17:58.343142 15680 solver.cpp:244]     Train net output #0: loss = 0.271455 (* 1 = 0.271455 loss)\n",
      "I0419 11:17:58.343161 15680 sgd_solver.cpp:106] Iteration 3100, lr = 0.01\n",
      "I0419 11:17:58.344684 15680 solver.cpp:228] Iteration 3200, loss = 0.2707\n",
      "I0419 11:17:58.344699 15680 solver.cpp:244]     Train net output #0: loss = 0.2707 (* 1 = 0.2707 loss)\n",
      "I0419 11:17:58.344707 15680 sgd_solver.cpp:106] Iteration 3200, lr = 0.01\n",
      "I0419 11:17:58.346230 15680 solver.cpp:228] Iteration 3300, loss = 0.270018\n",
      "I0419 11:17:58.346246 15680 solver.cpp:244]     Train net output #0: loss = 0.270018 (* 1 = 0.270018 loss)\n",
      "I0419 11:17:58.346253 15680 sgd_solver.cpp:106] Iteration 3300, lr = 0.01\n",
      "I0419 11:17:58.347726 15680 solver.cpp:228] Iteration 3400, loss = 0.269394\n",
      "I0419 11:17:58.347734 15680 solver.cpp:244]     Train net output #0: loss = 0.269394 (* 1 = 0.269394 loss)\n",
      "I0419 11:17:58.347743 15680 sgd_solver.cpp:106] Iteration 3400, lr = 0.01\n",
      "I0419 11:17:58.349104 15680 solver.cpp:337] Iteration 3500, Testing net (#0)\n",
      "I0419 11:17:58.350683 15680 solver.cpp:404]     Test net output #0: accuracy = 0.878\n",
      "I0419 11:17:58.350697 15680 solver.cpp:404]     Test net output #1: loss = 0.430557 (* 1 = 0.430557 loss)\n",
      "I0419 11:17:58.350723 15680 solver.cpp:228] Iteration 3500, loss = 0.26884\n",
      "I0419 11:17:58.350730 15680 solver.cpp:244]     Train net output #0: loss = 0.26884 (* 1 = 0.26884 loss)\n",
      "I0419 11:17:58.350738 15680 sgd_solver.cpp:106] Iteration 3500, lr = 0.01\n",
      "I0419 11:17:58.352162 15680 solver.cpp:228] Iteration 3600, loss = 0.268506\n",
      "I0419 11:17:58.352170 15680 solver.cpp:244]     Train net output #0: loss = 0.268506 (* 1 = 0.268506 loss)\n",
      "I0419 11:17:58.352176 15680 sgd_solver.cpp:106] Iteration 3600, lr = 0.01\n",
      "I0419 11:17:58.353525 15680 solver.cpp:228] Iteration 3700, loss = 0.268152\n",
      "I0419 11:17:58.353533 15680 solver.cpp:244]     Train net output #0: loss = 0.268152 (* 1 = 0.268152 loss)\n",
      "I0419 11:17:58.353536 15680 sgd_solver.cpp:106] Iteration 3700, lr = 0.01\n",
      "I0419 11:17:58.354885 15680 solver.cpp:228] Iteration 3800, loss = 0.267797\n",
      "I0419 11:17:58.354893 15680 solver.cpp:244]     Train net output #0: loss = 0.267797 (* 1 = 0.267797 loss)\n",
      "I0419 11:17:58.354898 15680 sgd_solver.cpp:106] Iteration 3800, lr = 0.01\n",
      "I0419 11:17:58.356355 15680 solver.cpp:228] Iteration 3900, loss = 0.267471\n",
      "I0419 11:17:58.356366 15680 solver.cpp:244]     Train net output #0: loss = 0.267471 (* 1 = 0.267471 loss)\n",
      "I0419 11:17:58.356372 15680 sgd_solver.cpp:106] Iteration 3900, lr = 0.01\n",
      "I0419 11:17:58.357841 15680 solver.cpp:454] Snapshotting to binary proto file multiclass_hdf5_iter_4000.caffemodel\n",
      "I0419 11:17:58.358055 15680 sgd_solver.cpp:273] Snapshotting solver state to binary proto file multiclass_hdf5_iter_4000.solverstate\n",
      "I0419 11:17:58.358188 15680 solver.cpp:317] Iteration 4000, loss = 0.266962\n",
      "I0419 11:17:58.358201 15680 solver.cpp:337] Iteration 4000, Testing net (#0)\n",
      "I0419 11:17:58.360735 15680 solver.cpp:404]     Test net output #0: accuracy = 0.878\n",
      "I0419 11:17:58.360774 15680 solver.cpp:404]     Test net output #1: loss = 0.428501 (* 1 = 0.428501 loss)\n",
      "I0419 11:17:58.360785 15680 solver.cpp:322] Optimization Done.\n",
      "I0419 11:17:58.360791 15680 caffe.cpp:222] Optimization Done.\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# could not train using pycaffe...\n",
    "\n",
    "proc = subprocess.Popen(\n",
    "    [\"/data/soft/caffe/build/tools/caffe\", \"train\", \"--solver=multiclass_hdf5_solver.prototxt\"],\n",
    "    stderr=subprocess.PIPE)\n",
    "\n",
    "out = proc.communicate()[1]\n",
    "\n",
    "print out"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Test model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy:  0.899999976158\n"
     ]
    }
   ],
   "source": [
    "net = caffe.Net('multiclass_hdf5.prototxt',\n",
    "                'multiclass_hdf5_iter_4000.caffemodel',\n",
    "                caffe.TEST)\n",
    "\n",
    "out = net.forward()\n",
    "\n",
    "print 'Accuracy: ', out['accuracy']"
   ]
  }
 ],
 "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": "2.7.11"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}