Use case of Caffe with cifar10

{
 "metadata": {
  "name": ""
 },
 "nbformat": 3,
 "nbformat_minor": 0,
 "worksheets": [
  {
   "cells": [
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "%matplotlib inline\n",
      "import numpy as np\n",
      "import matplotlib.pyplot as plt\n",
      "import leveldb"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 1
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "MODEL_FILE=\"cifar10_quick.prototxt\"\n",
      "PRETRAINED=\"cifar10_quick_iter_5000.caffemodel\"\n",
      "import sys\n",
      "sys.path.insert(0,\"../../python\")\n",
      "import caffe"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 2
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "caffe.set_phase_test()\n",
      "caffe.set_mode_cpu()"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 3
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "import caffe_pb2\n",
      "MEAN_FILE=\"mean.binaryproto\"\n",
      "with open(MEAN_FILE,\"rb\") as fp:\n",
      "    b=caffe_pb2.BlobProto()\n",
      "    b.ParseFromString(fp.read())\n",
      "mean=np.array(b.data).reshape(3,32,32)"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 4
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "net=caffe.Classifier(MODEL_FILE, PRETRAINED, image_dims=(32,32))"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 5
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "ldb=leveldb.LevelDB(\"cifar10_test_leveldb\")\n",
      "iter=ldb.RangeIter()\n",
      "data=[]\n",
      "for k,v in iter:\n",
      "    a=caffe_pb2.Datum()\n",
      "    a.ParseFromString(v)\n",
      "    data.append((k,a))"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 6
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "from PIL import Image\n",
      "import struct\n",
      "test_image=[]\n",
      "for k,v in data:\n",
      "    d=np.array(list(struct.unpack(\"3072B\",v.data)),dtype=np.uint8).reshape(3,32,32).transpose(1,2,0)\n",
      "    test_image.append(d)"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 7
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "pred=net.predict(test_image)"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 8
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "labels=[]\n",
      "for k,v in data:\n",
      "    labels.append(v.label)"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 9
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "pr=[]\n",
      "for p in pred:\n",
      "    pr.append(p.argmax())"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 10
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "labels=np.array(labels)\n",
      "pr=np.array(pr)\n",
      "m=(labels==pr)"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 11
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "m.sum(), len(m)"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 12,
       "text": [
        "(997, 10000)"
       ]
      }
     ],
     "prompt_number": 12
    }
   ],
   "metadata": {}
  }
 ]
}