notebooks/gen_balanced2.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import json\n",
    "import exifread\n",
    "import subprocess\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from PIL import Image\n",
    "\n",
    "import env\n",
    "from utils import KaggleCameraDataset, progress_iter, RNG\n",
    "\n",
    "%matplotlib inline\n",
    "%load_ext autoreload\n",
    "%autoreload 2"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## validation set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "N_VAL_PER_CLASS = 48\n",
    "val, no_val = [], []"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### get best images from Gleb's val set (quality $\\geq$ 95)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "with open('../data/glebs_val_stats.json') as f:\n",
    "    glebs_val_stats = json.load(f)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "325 9\n"
     ]
    }
   ],
   "source": [
    "gv_yes, gv_no = [], []\n",
    "for i, d in sorted(glebs_val_stats.items()):\n",
    "    if d['quality'] and d['quality'] >= 95:\n",
    "        gv_yes.append((d['x_fpath'], d['y']))\n",
    "    else:\n",
    "        gv_no.append((d['x_fpath'], d['y'], d['quality'], d['res'], d['res_valid']))\n",
    "print len(gv_yes), len(gv_no)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "val += gv_yes\n",
    "no_val += gv_no"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[45, 37, 14, 28, 30, 18, 44, 42, 36, 31]"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXQAAAEACAYAAACj0I2EAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n1AgDXZIaYaBLUiMMdElqxP8D1WoYiAsidfEAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8cb0044510>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "VAL_CLASSES = map(int, plt.hist(zip(*val)[1])[0])\n",
    "VAL_CLASSES"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### get best images from alex firsov val set (quality $\\geq$ 95)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "with open('../data/alex_firsov_stats.json') as f:\n",
    "    alex_firsov_stats = json.load(f)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "333 11\n"
     ]
    }
   ],
   "source": [
    "af_yes, af_no = [], []\n",
    "for i, d in sorted(alex_firsov_stats.items()):\n",
    "    d['y'] = 4  # !!!!!!!!!!!!\n",
    "    d['x_fpath'] = d['x_fpath'].replace('Motorola-Droid-Maxx', 'Motorola-X')\n",
    "    if d['quality'] and d['quality'] >= 95:\n",
    "        af_yes.append((d['x_fpath'], d['y'], d['quality'], d['res'], d['res_valid']))\n",
    "    else:\n",
    "        af_no.append((d['x_fpath'], d['y'], d['quality'], d['res'], d['res_valid']))\n",
    "print len(af_yes), len(af_no)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "RNG(seed=1337).shuffle(af_yes)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "val += map(lambda (x,y,q,r,rv): (x,y), af_yes[:(N_VAL_PER_CLASS - VAL_CLASSES[4])])\n",
    "no_val += af_no\n",
    "no_val += af_yes[(N_VAL_PER_CLASS - VAL_CLASSES[4]):]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[45, 37, 14, 28, 48, 18, 44, 42, 36, 31]"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXQAAAEACAYAAACj0I2EAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nlKkv5gAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c4792ce50>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "VAL_CLASSES = map(int, plt.hist(zip(*val)[1])[0])\n",
    "VAL_CLASSES"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### fill other classes with $\\geq$ 95 from esato"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "with open('../data/esato_stats.json') as f:\n",
    "    esato_stats = json.load(f)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1063 89\n"
     ]
    }
   ],
   "source": [
    "es_yes, es_no = [], []\n",
    "for i, d in sorted(esato_stats.items()):\n",
    "    if d['x_fpath'].endswith('jpg.1'):\n",
    "        continue\n",
    "    if d['quality'] and d['quality'] >= 95:\n",
    "        es_yes.append((d['x_fpath'], d['y'], d['quality'], d['res'], d['res_valid']))\n",
    "    else:\n",
    "        es_no.append((d['x_fpath'], d['y'], d['quality'], d['res'], d['res_valid']))\n",
    "print len(es_yes), len(es_no)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[2, 0, 15, 0, 362, 0, 6, 0, 546, 132]"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXcAAAEACAYAAABI5zaHAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nkOHM/lwO1QAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c47833290>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "map(int, plt.hist(zip(*es_yes)[1])[0])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### add all with valid resolution first"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[1, 0, 0, 0, 0, 0, 89, 0, 0, 4]"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXUAAAEACAYAAABMEua6AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nJDXEUJekhhjqktSQ/w+E919+ZJ43owAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c47920b90>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "map(int, plt.hist(zip(*filter(lambda (x,y,q,r,rv): rv, es_yes))[1])[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "es_yes_valid = filter(lambda (x,y,q,r,rv): rv, es_yes)\n",
    "es_yes_not_valid = filter(lambda (x,y,q,r,rv): not rv, es_yes)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXQAAAEACAYAAACj0I2EAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c474fc810>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "for x,y,q,r,rv in es_yes_valid:\n",
    "    VAL_CLASSES = map(int, plt.hist(zip(*val)[1])[0])\n",
    "    if VAL_CLASSES[y] < N_VAL_PER_CLASS:\n",
    "        val.append((x, y))\n",
    "    else:\n",
    "        no_val.append((x,y,q,r,rv))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[45, 37, 14, 28, 48, 18, 45, 42, 48, 35]"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXQAAAEACAYAAACj0I2EAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c47313f50>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "VAL_CLASSES = map(int, plt.hist(zip(*val)[1])[0])\n",
    "VAL_CLASSES"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### now add with incorrect resolution (UPD: add incorrect resolution -> train for further filtering)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "RNG(seed=42).shuffle(es_yes_not_valid)\n",
    "for x,y,q,r,rv in es_yes_not_valid:\n",
    "    no_val.append((x,y,q,r,rv))    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[45, 37, 14, 28, 48, 18, 45, 42, 48, 35]"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXQAAAEACAYAAACj0I2EAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c466fa610>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "VAL_CLASSES = map(int, plt.hist(zip(*val)[1])[0])\n",
    "VAL_CLASSES"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "no_val += es_no"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## training set\n",
    "### load Gleb's train and remove $\\lt$ 93"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "with open('../data/orig_stats.json') as f:\n",
    "    train_stats = json.load(f)\n",
    "for x,y,q in train_stats:\n",
    "    no_val.append((x,y,q,None,True))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "QUALITY_THRESHOLD = 93"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "with open('../data/glebs_train_stats.json') as f:\n",
    "    glebs_train_stats = json.load(f)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "9229\n"
     ]
    }
   ],
   "source": [
    "print len(filter(lambda q: q >= QUALITY_THRESHOLD, [v['quality'] for k, v in glebs_train_stats.items()]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1775\n"
     ]
    }
   ],
   "source": [
    "print len(filter(lambda q: not q or q < QUALITY_THRESHOLD, [v['quality'] for k, v in glebs_train_stats.items()]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1\n"
     ]
    }
   ],
   "source": [
    "print len(filter(lambda q: not q or q < QUALITY_THRESHOLD, [q for x,y,q in train_stats]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "train = []\n",
    "for x,y,q,r,rv in no_val:\n",
    "    if q >= QUALITY_THRESHOLD:\n",
    "        train.append((x,y,q,r,rv))\n",
    "for i, d in sorted(glebs_train_stats.items()):\n",
    "    if d['quality'] >= QUALITY_THRESHOLD:\n",
    "        train.append((d['x_fpath'], d['y'], d['quality'], d['res'], d['res_valid']))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### NEW: add artgor data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "with open('../data/artgor_stats.json') as f:\n",
    "    artgor_stats = json.load(f)\n",
    "for i, d in sorted(artgor_stats.items()):\n",
    "    if d['quality'] >= QUALITY_THRESHOLD:\n",
    "        train.append((d['x_fpath'], d['y'], d['quality'], d['res'], d['res_valid']))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "13897"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[1050, 874, 904, 1002, 1989, 1514, 1834, 1385, 2129, 1216]"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYAAAAEACAYAAAC6d6FnAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c460659d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "TRAIN_CLASSES = map(int, plt.hist(zip(*train)[1])[0])\n",
    "TRAIN_CLASSES"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.5894786284640676"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "M = max(TRAIN_CLASSES)\n",
    "m = min(TRAIN_CLASSES)\n",
    "(M - m) / float(M)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### remove the most \"unlikely resolutions\", i.e. with maximal $d(\\text{img}, \\text{class})=\\min_{r \\in R_{\\text{class}}}\\min\\{\\|r_{\\text{img}}-r\\|_2, \\|r_{\\text{img}}-r_{\\text{rotated}}\\|_2\\}$, $r=\\begin{pmatrix} h \\\\ w \\end{pmatrix}$"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def resolutions(): # with all possible rotations\n",
    "    return [\n",
    "        [(1520, 2688), (2688, 1520)],\n",
    "        [(3024, 4032), (4032, 3024)],\n",
    "        [(2432, 4320), (4320, 2432)],\n",
    "        [(1040, 780), (3088, 4130), (4130, 3088), (3120, 4160), (4160, 3120)],\n",
    "        [(3120, 4160), (4160, 3120), (4160, 2340)],\n",
    "        [(4128, 2322)],\n",
    "        [(4128, 2322)],\n",
    "        [(6000, 4000)],\n",
    "        [(3264, 2448)],\n",
    "        [(3264, 2448)],\n",
    "    ]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def distance(res, c):\n",
    "    return min(min([np.sqrt((res[0]-r[0])**2+(res[1]-r[1])**2) for r in resolutions()[c]]),\n",
    "               min([np.sqrt((res[0]-r[1])**2+(res[1]-r[0])**2) for r in resolutions()[c]]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "5314"
      ]
     },
     "execution_count": 35,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(filter(lambda (x,y,q,r,rv): rv, train)) # with correct resolutions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "8583"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(filter(lambda (x,y,q,r,rv): not rv, train))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### e.g., \"incorrect iphone4s resolutions\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYYAAAEACAYAAAC3adEgAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\ndwicAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c46065f10>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "incorrect_iphone4s = filter(lambda (x,y,q,r,rv): not rv and y == 8, train)\n",
    "dists = map(lambda (x,y,q,r,rv): distance(r, 8), incorrect_iphone4s)\n",
    "dists.sort()\n",
    "plt.hist(dists, bins=24);"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[1655.9619560847405,\n",
       " 1631.9571072794774,\n",
       " 1593.5294161075283,\n",
       " 1584.788944938726,\n",
       " 1584.0012626257594,\n",
       " 1580.0,\n",
       " 1580.0,\n",
       " 1580.0,\n",
       " 1580.0,\n",
       " 1580.0]"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dists[::-1][40:50]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.026346604215456676"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sum(np.asarray(dists) > 1581.)/float(len(dists))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.38114754098360654"
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sum(np.asarray(dists) > 1.)/float(len(dists)) # 0 means simply landscape photos"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[(u'../data/glebs-train/train/iPhone-4s/38290361134_d74e19efa3_o.jpg',\n",
       "  8,\n",
       "  96,\n",
       "  [1698, 1858],\n",
       "  False),\n",
       " (u'../data/esato-com/train/iPhone-4s/201303121858b24wEX.jpg',\n",
       "  8,\n",
       "  96,\n",
       "  [2498, 4848],\n",
       "  False),\n",
       " (u'../data/esato-com/train/iPhone-4s/201212020931H43XZV.jpg',\n",
       "  8,\n",
       "  96,\n",
       "  [2446, 4848],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/iPhone-4s/24694324078_3435e21cd7_o.jpg',\n",
       "  8,\n",
       "  99,\n",
       "  [1500, 2000],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/iPhone-4s/24694337928_76c70409b6_o.jpg',\n",
       "  8,\n",
       "  99,\n",
       "  [1500, 2000],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/iPhone-4s/24694339268_a86b1ce6cd_o.jpg',\n",
       "  8,\n",
       "  99,\n",
       "  [1500, 2000],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/iPhone-4s/24694350208_4cf3f1ea52_o.jpg',\n",
       "  8,\n",
       "  99,\n",
       "  [1500, 2000],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/iPhone-4s/24694359098_9ae4375750_o.jpg',\n",
       "  8,\n",
       "  99,\n",
       "  [1500, 2000],\n",
       "  False)]"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "incorrect_iphone4s.sort(key=lambda (x,y,q,r,rv): -distance(r,8))\n",
    "incorrect_iphone4s[42:50]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "13246\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[1050, 874, 904, 1002, 1989, 1514, 1834, 1385, 1478, 1216]"
      ]
     },
     "execution_count": 42,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYAAAAEACAYAAAC6d6FnAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nkhplAEhSowwASWqUASBJjfp/FGOc+9KPgkYAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c4633e250>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "train = filter(lambda (x,y,q,r,rv): y != 8 or (y == 8 and (rv or (not rv and distance(r, 8) < 1.))), train)\n",
    "print len(train)\n",
    "TRAIN_CLASSES = map(int, plt.hist(zip(*train)[1])[0])\n",
    "TRAIN_CLASSES"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### incorrect Samsung-Galaxy-S4 resolutions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYcAAAEACAYAAABYq7oeAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAFFxJREFUeJzt3H+s5XV95/HnC0as+GMADTNZBhhRxLrZlpoV2Npmzwoo\n0FSabKjoNvzQTcwqq627VtCkzE3/qJoYf7RdXbNThK5KFdtlkqVlJOM1666ALIxQQRh/MsMslxoZ\nNrqJVXjvH+dz8Xg/9zIz99xz7z3wfCQn8/2+v5/v9/v+Xs7M635/kapCkqRRR6x1A5Kk9cdwkCR1\nDAdJUsdwkCR1DAdJUsdwkCR1DhoOSbYnmUty9yLL/mOSJ5IcN1L7WJI9SXYnOX2kfmmSB5Lcn+SS\nlTsESdJKO5Qzh2uA1y0sJtkCnAN8f6R2PvCSqjoVeCvwiVY/Fvgj4FXAmcDVSTaO3b0kaSIOGg5V\n9RXg0UUWfRh494LahcB1bb3bgI1JNjEMl51V9VhVHQB2AueN07gkaXKWdc8hyW8De6vqngWLTgD2\njszva7WF9YdaTZK0Dm043BWSPAd4H3DuYosXma9F6rS6JGkdOuxwAF4CbAW+niTAFuDOJGcwPFM4\ncWTsFmB/qw8W1L+02MaTGBqStAxVtdgv4styqJeV0j5U1d9X1eaqOqWqXszwH/5fq6pHgB3AJQBJ\nzgIOVNUccDNwbpKN7eb0ua22qKqa2s/VV1+95j3Y/9r3Yf/T95nm3qtW/nfqQ3mU9TPA/wJeluTB\nJJcv/LecnwfHTcB3k3wL+M/A21r9UeCPgTuA24CZGt6YliStQwe9rFRVbzrI8lMWzF+xxLhPAZ86\njN4kSWvEN6RX2GAwWOsWxmL/a8v+18409z4JmcS1qnEkqfXWkyStd0moNbghLUl6BjEcJEkdw0GS\n1DEcJEkdw0GS1DEcJEkdw0GS1DEcJEkdw0GS1DEcJEkdw0GS1DEcJEkdw0GS1DEcJEkdw0GS1DEc\nJEkdw0GS1DEcJEkdw0GS1DEcJEkdw0GS1DloOCTZnmQuyd0jtQ8muS/J7iRfSPKCkWVXJdnTlr92\npH5ekm8meSDJe1b+UCRJKyVV9dQDkt8AfgRcV1W/0mrnALuq6okk7weqqq5K8grg08CrgC3ALcCp\nQIAHgLOB/cDXgIur6puL7K/OOOO1C8tPaWbmP3DeeYe3jiQ9nSShqrJS29twsAFV9ZUkJy+o3TIy\neyvwr9v064Hrq+pnwPeS7AHOYBgOe6rq+wBJrgcuBLpwALj99ncdxiHcwE037TQcJGkFHTQcDsGb\ngc+26ROAr44se6jVAuwdqe9jGBpLeN1h7P4e4OHDGC9JOpixwiHJ+4CfVtV8OCx2SlMsfm/jKa5n\nbRuZHrSPJGne7Owss7OzE9v+ssMhyaXABcBrRsr7gBNH5rcwvMcQ4KRF6kvYtty2JOkZYTAYMBgM\nnpyfmZlZ0e0f6qOsYeSsIMl5wB8Cr6+qn4yM2wFcnOSoJC8GXgrczvAG9EuTnJzkKODiNlaStA4d\n9MwhyWcYXtd5YZIHgauB9wJHAV9MAnBrVb2tqu5N8jngXuCnwNtq+DjU40muAHYyDKTtVXXfJA5I\nkjS+Q3la6U2LlK95ivF/AvzJIvW/A047rO4kSWvCN6QlSR3DQZLUMRwkSR3DQZLUMRwkSR3DQZLU\nMRwkSR3DQZLUMRwkSR3DQZLUMRwkSR3DQZLUMRwkSR3DQZLUMRwkSR3DQZLUMRwkSR3DQZLUMRwk\nSR3DQZLUMRwkSR3DQZLUMRwkSZ2DhkOS7Unmktw9Ujs2yc4k9ye5OcnGkWUfS7Inye4kp4/UL03y\nQFvnkpU/FEnSSjmUM4drgNctqF0J3FJVpwG7gKsAkpwPvKSqTgXeCnyi1Y8F/gh4FXAmcPVooEiS\n1peDhkNVfQV4dEH5QuDaNn1tm5+vX9fWuw3YmGQTw3DZWVWPVdUBYCdw3vjtS5ImYbn3HI6vqjmA\nqnoYOL7VTwD2jozb12oL6w+1miRpHdqwwtvLIvO1SJ1WX8K2kelB+0iS5s3OzjI7Ozux7S83HOaS\nbKqquSSbgUdafR9w4si4LcD+Vh8sqH9p6c1vW2ZbkvTMMBgMGAwGT87PzMys6PYP9bJS+MXf/ncA\nl7Xpy4AbR+qXACQ5CzjQLj/dDJybZGO7OX1uq0mS1qGDnjkk+QzD3/pfmORB4Grg/cDnk7wZeBC4\nCKCqbkpyQZJvAT8GLm/1R5P8MXAHw8tJM+3GtCRpHTpoOFTVm5ZYdM4S469Yov4p4FOH2pgkae34\nhrQkqWM4SJI6hoMkqWM4SJI6hoMkqWM4SJI6hoMkqWM4SJI6hoMkqWM4SJI6hoMkqWM4SJI6hoMk\nqWM4SJI6hoMkqWM4SJI6hoMkqWM4SJI6hoMkqWM4SJI6hoMkqWM4SJI6hoMkqTNWOCT5gyR/n+Tu\nJJ9OclSSrUluTXJ/ks8m2dDGHpXk+iR7knw1yUkrcwiSpJW27HBI8k+Afw+8sqp+BdgAvBH4APCh\nqjoNOAC8pa3yFuCHVXUq8BHgg+M0LkmanHEvKx0JPLedHTwH2A/8K+ALbfm1wO+06QvbPMANwNlj\n7luSNCHLDoeq2g98CHgQeAh4DLgTOFBVT7Rh+4AT2vQJwN627uPAgSTHLXf/kqTJ2bDcFZMcw/Bs\n4GSGwfB54PxFhtb8Kgs3MbJsgW0j04P2kSTNm52dZXZ2dmLbX3Y4AOcA36mqHwIk+Rvg14FjkhzR\nzh62MLzUBMOziBOB/UmOBF5QVY8uvultY7QlSU9/g8GAwWDw5PzMzMyKbn+cew4PAmcl+aUkYXgP\n4RvAl4CL2phLgRvb9I42T1u+a4x9S5ImaJx7DrczvLF8F/B1hpeJPglcCbwryQPAccD2tsp24EVJ\n9gC/38ZJktahcS4rUVUzwMJzme8CZy4y9ifA746zP0nS6vANaUlSx3CQJHUMB0lSx3CQJHUMB0lS\nx3CQJHUMB0lSx3CQJHUMB0lSx3CQJHUMB0lSx3CQJHUMB0lSx3CQJHUMB0lSx3CQJHUMB0lSx3CQ\nJHUMB0lSx3CQJHUMB0lSx3CQJHUMB0lSZ6xwSLIxyeeT3JfkG0nOTHJskp1J7k9yc5KNI+M/lmRP\nkt1JTh+/fUnSJIx75vBR4Kaq+mXgV4FvAlcCt1TVacAu4CqAJOcDL6mqU4G3Ap8Yc9+SpAlZdjgk\neT7wm1V1DUBV/ayqHgMuBK5tw65t87Q/r2tjbwM2Jtm03P1LkiZnnDOHU4AfJLkmyZ1JPpnkaGBT\nVc0BVNXDwPFt/AnA3pH1H2o1SdI6s2HMdV8JvL2q7kjyYYaXlGqJ8VmktsTYbSPTg/aRJM2bnZ1l\ndnZ2YtsfJxz2AXur6o42/wWG4TCXZFNVzSXZDDwyMv7EkfW3APsX3/S2MdqSpKe/wWDAYDB4cn5m\nZmZFt7/sy0rt0tHeJC9rpbOBbwA7gMta7TLgxja9A7gEIMlZwIH5y0+SpPVlnDMHgHcAn07yLOA7\nwOXAkcDnkrwZeBC4CKCqbkpyQZJvAT9uYyVJ69BY4VBVXwdetciic5YYf8U4+5MkrQ7fkJYkdQwH\nSVLHcJAkdQwHSVLHcJAkdQwHSVLHcJAkdQwHSVLHcJAkdQwHSVLHcJAkdQwHSVLHcJAkdQwHSVLH\ncJAkdQwHSVLHcJAkdQwHSVLHcJAkdQwHSVLHcJAkdQwHSVLHcJAkdcYOhyRHJLkzyY42vzXJrUnu\nT/LZJBta/agk1yfZk+SrSU4ad9+SpMlYiTOHdwL3jsx/APhQVZ0GHADe0upvAX5YVacCHwE+uAL7\nliRNwFjhkGQLcAHwX0bKrwG+0KavBX6nTV/Y5gFuAM4eZ9+SpMkZ98zhw8C7gQJI8kLg0ap6oi3f\nB5zQpk8A9gJU1ePAgSTHjbl/SdIEbFjuikl+C5irqt1JBvPl9hlVI8t+YRMjyxbYNjI9aB9J0rzZ\n2VlmZ2cntv1lhwPwauD1SS4AngM8n+G9hI1JjmhnD1uA/W38PuBEYH+SI4EXVNWji2962xhtSdLT\n32AwYDAYPDk/MzOzottf9mWlqnpvVZ1UVacAFwO7qur3gC8BF7VhlwI3tukdbZ62fNdy9y1JmqxJ\nvOdwJfCuJA8AxwHbW3078KIke4Dfb+MkSevQOJeVnlRVXwa+3Ka/C5y5yJifAL+7EvuTJE2Wb0hL\nkjqGgySpYzhIkjqGgySpYzhIkjqGgySpYzhIkjqGgySpYzhIkjqGgySpYzhIkjqGgySpYzhIkjqG\ngySpYzhIkjqGgySpYzhIkjqGgySpYzhIkjqGgySp84wNh82bt5LkkD+bN29d65YladVsWOsG1src\n3PeBOozxmVwzkrTOPGPPHCRJS1t2OCTZkmRXknuT3JPkHa1+bJKdSe5PcnOSjSPrfCzJniS7k5y+\nEgcgSVp545w5/Ax4V1W9AvgXwNuTvBy4Erilqk4DdgFXASQ5H3hJVZ0KvBX4xFidS5ImZtnhUFUP\nV9XuNv0j4D5gC3AhcG0bdm2bp/15XRt/G7Axyabl7l+SNDkrcs8hyVbgdOBWYFNVzcEwQIDj27AT\ngL0jqz3UapKkdWbsp5WSPA+4AXhnVf0oyVKPAC32uM8SY7eNTA/aR5I0b3Z2ltnZ2Yltf6xwSLKB\nYTD8ZVXd2MpzSTZV1VySzcAjrb4POHFk9S3A/sW3vG2ctiTpaW8wGDAYDJ6cn5mZWdHtj3tZ6S+A\ne6vqoyO1HcBlbfoy4MaR+iUASc4CDsxffpIkrS/LPnNI8mrg3wD3JLmL4SWi9wIfAD6X5M3Ag8BF\nAFV1U5ILknwL+DFw+bjNS5ImY9nhUFX/EzhyicXnLLHOFcvdnyRp9fiGtCSpYzhIkjqGgySpYzhI\nkjqGgySpYzhIkjqGgySpYzhIkjqGgySpYzhIkjqGgySpYzhIkjqGgySpYzhIkjqGgySpYzhIkjqG\ngySpYzhIkjqGgySpYzgcsmeT5LA+mzdvXeumJWlZNqx1A9PjJ0Ad1hpzc5lMK5I0YZ45SJI6qx4O\nSc5L8s0kDyR5z2rvX5J0cKsaDkmOAP4MeB3wT4E3Jnn5avYwebNr3cBYZmdn17qFsdj/2prm/qe5\n90lY7TOHM4A9VfX9qvopcD1w4Sr3MGGza93AWKb9L4j9r61p7n+ae5+E1Q6HE4C9I/P7Wk3r2ObN\nWw/7Sa0jj3yuT3ZJU2y1n1Za7PGd7hGgF7zgtw95g//4j99m+/ZH+NM//dA4fU3I8PHXQ7Vp08k8\n/PD3DmsPmzdvZW7u+4e1zhFHHM0TT/y/JZfPzMwsUj28J7WeeCKHtc7c3C8d1s8Klj6OxfsfWs7P\nWHomStXh/aUfa2fJWcC2qjqvzV8JVFV9YGTM6jUkSU8jVbViz8+vdjgcCdwPnA38H+B24I1Vdd+q\nNSFJOqhVvaxUVY8nuQLYyfB+x3aDQZLWn1U9c5AkTYd19Yb0en1BLsn2JHNJ7h6pHZtkZ5L7k9yc\nZOPIso8l2ZNkd5LTR+qXtmO7P8klq9T7liS7ktyb5J4k75iy/p+d5LYkd7X+r271rUlubb18NsmG\nVj8qyfWt/68mOWlkW1e1+n1JXrsa/Y/s+4gkdybZMW39J/lekq+3/wa3t9pUfH/afjcm+Xz7uX0j\nyZnT0n+Sl7Wf+53tz8eSvGNV+q+qdfFhGFTfAk4GngXsBl6+1n213n4DOB24e6T2AeAP2/R7gPe3\n6fOB/96mzwRubdPHAt8GNgLHzE+vQu+bgdPb9PMY3vN5+bT03/Z9dPvzSODW1tdfARe1+seBt7bp\nfwf8pzb9BuD6Nv0K4C6Gl1K3tu9aVvE79AfAfwV2tPmp6R/4DnDsgto0fX8+BVzepje0Hqam/5Hj\nOALYD5y4Gv2v2oEdwoGfBfztyPyVwHvWuq+Rfk7mF8Phm8CmNr0ZuK9NfwJ4w8i4+4BNwMXAx0fq\nHx8dt4rH8d+Ac6axf+Bo4A6GL1M+Ahyx8LsD/B1wZps+Enhkse8T8Lfz41ah7y3AF4EBPw+Hf5ii\n/r8LvHBBbSq+P8DzgW8vUp+K/hf0/Frgf6xW/+vpstK0vSB3fFXNAVTVw8Dxrb7UcSysP8QqH1+S\nrQzPgG5l+MWaiv7bJZm7gIcZ/iP7beBAVT2xoEdG+6yqx4HHkhzH2v78Pwy8m/biR5IXAo9OUf8F\n3Jzka0n+batNy/fnFOAHSa5pl2Y+meRopqf/UW8APtOmJ97/egqHQ3pBbgosPI75t8HW9PiSPA+4\nAXhnVf3oKfa97vqvqieq6tcY/gZ+BvDLT9HLUn2uSf9JfguYq6rdIz1kkX7WZf/Nr1fVPwcuAN6e\n5DefYt/r7fuzAXgl8OdV9UrgxwzPwqal/2ETybOA1wOfP8i+V6z/9RQO+4CTRua3MLy+tl7NJdkE\nkGQzw8scMDyOE0fGzR/Hmh1fu9l5A/CXVXVjK09N//Oq6v8CX2Z4GeaYDP9Hjgt7ebL/DN+r2VhV\nj7L0cU3aq4HXJ/kO8FngNcBHgI1T0v/8b6ZU1T8wvCx5BtPz/dkH7K2qO9r8FxiGxbT0P+984H9X\n1Q/a/MT7X0/h8DXgpUlOTnIUw2tkO9a4p1ELf9vbAVzWpi8DbhypXwJPvhF+oJ3+3Qyc256cOBY4\nt9VWw18A91bVR0dqU9F/khfNP4mR5DkM75fcC3wJuKgNu3RB/5e26YuAXSP1i9vTQC8GXsrwJcyJ\nqqr3VtVJVXUKw+/0rqr6vWnpP8nR7ayTJM9leN37Hqbk+9P2vTfJy1rpbOAb09L/iDcy/OVi3uT7\nX80bKodww+U8hk/T7AGuXOt+Rvr6DMOU/QnwIHA5w7v/t7R+vwgcMzL+zxg+TfJ14JUj9cvasT0A\nXLJKvb8aeJzh0193AXe2n/NxU9L/P2s97wbuBt7X6i8Gbmu9/BXwrFZ/NvC51uetwNaRbV3Vjus+\n4LVr8D36l/z8hvRU9N/6nP/u3DP/93Javj9tv7/K8JfP3cBfM3xiZ5r6fw7DBxieP1KbeP++BCdJ\n6qyny0qSpHXCcJAkdQwHSVLHcJAkdQwHSVLHcJAkdQwHSVLHcJAkdf4/rpho0xW0a8EAAAAASUVO\nRK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c46065650>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "incorrect_S4 = filter(lambda (x,y,q,r,rv): not rv and y == 6, train)\n",
    "dists = map(lambda (x,y,q,r,rv): distance(r, 6), incorrect_S4)\n",
    "dists.sort()\n",
    "plt.hist(dists, bins=24);"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "12907\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[1050, 874, 904, 1002, 1989, 1514, 1495, 1385, 1478, 1216]"
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYAAAAEACAYAAAC6d6FnAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nIEmNMgAkqVEGgCQ16v8BgkmoudV26C0AAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c46635f10>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "train = filter(lambda (x,y,q,r,rv): y != 6 or (y == 6 and (rv or (not rv and distance(r, 6) < 1.))), train)\n",
    "print len(train)\n",
    "TRAIN_CLASSES = map(int, plt.hist(zip(*train)[1])[0])\n",
    "TRAIN_CLASSES"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### incorrect Motorola-X resolutions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1713\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYoAAAEACAYAAACtVTGuAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAFW9JREFUeJzt3WuwXWWd5/HvL8R0ayvhYhGqEyByEdSabgpLYNqemT2C\nClgt/QZFp4eLvKBaUVtraIhd1eRUd7WXKsdLOS1jTbj1iLSg3WRqaIkU7K5yRlAKIzQEEhqFhAzH\nYUyYwqmiFP7zYj9JNsdz1knOPrdwvp+qXWet/3rWXs96ziK/s9baa5OqQpKkqSxb6A5IkhY3g0KS\n1MmgkCR1MigkSZ0MCklSJ4NCktRp2qBIsiHJeJIHJ9Q/muTRJA8l+cxQfV2SbUm2JHnXUP2c1n5r\nkqtmdzckSXMl0z1HkeT3geeBm6rqd1qtB3wKOK+qfpXk9VX1bJI3ATcDbwPWAHcBJwEBtgJnATuB\nHwIXVtWjc7JXkqRZs3y6BlX1vSTHTSj/MfCZqvpVa/Nsq58P3NLqP02yDTidQVBsq6onAZLc0toa\nFJK0yM30HsUbgX+d5N4k9yR5a6uvBrYPtXu61SbWd7SaJGmRm/aMomO9w6rqzCRvA24Fjmdw5jBR\nMXkg+d0hknQQmGlQbAe+DVBVP0zyYpIjGZwpHDvUbg2DexKZov5rkhggkjQDVTXZH+sj299LT+Hl\nZwt/z+DGNEneCKyoqv8DbATen2RFkjcAJwI/YHDz+sQkxyVZAVzY2k6qqnxVcc011yx4HxbLy7Fw\nLByL7tdcmvaMIsnNQA84MslTwDXAdcD1SR4CXgAuAqiqR5J8E3gE+CXw4RrswYtJrgA2MQinDVW1\nZQ72R5I0y/bnU08fnGLRv5+i/aeBT09S/w5w8gH1TpK04HwyexHr9XoL3YVFw7HYx7HYx7GYH9M+\ncDffktRi65MkLXZJqAW+mS1JWqIMCklSJ4NCktTJoJAkdTIoJEmdDApJUieDQpLUyaCQJHUyKCRJ\nnWb6NeNz6rLLrjig9glceeVHOflkv0pKkmbbogyK66475YDaJ7dzwgnfZt26dXPUI0lauhZlUMCB\nnlE8PUf9kCR5j0KS1MmgkCR1MigkSZ0MCklSJ4NCktRp2qBIsiHJeJIHJ1n2H5K8lOSIodqXk2xL\nsjnJqUP1i5NsTfJYkotmbxckSXNpf84orgfePbGYZA1wNvDkUO1c4ISqOgm4HLi21Q8H/hx4G3AG\ncE2SlSP3XpI056YNiqr6HrBrkkVfAK6cUDsfuKmtdx+wMskqBkGzqaqeq6rdwCbgnFE6LkmaHzO6\nR5HkD4DtVfXQhEWrge1D8ztabWL96VaTJC1yB/xkdpJXA38GvHOyxZPM1yR1Wl2StMjN5Cs8TgDW\nAj9OEmAN8ECS0xmcQRwz1HYNsLPVexPq90y9ifVD070Jq0qS+v0+/X5/XraVqun/sE+yFvhvVfUv\nJln2E+C0qtqV5DzgI1X1niRnAl+sqjPbzez7gdMYXO66H3hru18x8f3qQE82li1bx1/+5aF+KaCk\nJSsJVTXZ1ZuR7c/HY28G/ifwxiRPJbl0QpO9l5aq6g7gJ0keB/4z8OFW3wX8BYOAuA8YmywkJEmL\nz7SXnqrqg9MsP37C/KRf/VpVNwA3HEDfJEmLgE9mS5I6GRSSpE4GhSSpk0EhSepkUEiSOhkUkqRO\nBoUkqZNBIUnqZFBIkjoZFJKkTgaFJKmTQSFJ6mRQSJI6GRSSpE4GhSSpk0EhSepkUEiSOhkUkqRO\nBoUkqdO0QZFkQ5LxJA8O1T6XZEuSzUm+leTQoWXrkmxry981VD8nyaNJtia5avZ3RZI0F/bnjOJ6\n4N0TapuAt1TVqcA2YB1AkjcD7wPeBJwL/HUGlgFfae/zFuADSU6ZnV2QJM2laYOiqr4H7JpQu6uq\nXmqz9wJr2vR7gVuq6ldV9VMGIXJ6e22rqier6pfALcD5s7MLkqS5NBv3KD4E3NGmVwPbh5Y93WoT\n6ztaTZK0yC0fZeUkfwb8sqq+sac0SbNi8kCqqd95/dB0r70kSXv0+336/f68bGvGQZHkYuA84B1D\n5R3AMUPza4CdDALk2EnqU1g/025J0pLQ6/Xo9Xp758fGxuZsW/t76SkMnS0kOQf4U+C9VfXCULuN\nwIVJViR5A3Ai8APgh8CJSY5LsgK4sLWVJC1y055RJLmZwbWfI5M8BVwDfApYAXw3CcC9VfXhqnok\nyTeBR4BfAh+uqgJeTHIFg09LLQM2VNWWudghSdLsmjYoquqDk5Sv72j/aeDTk9S/A5x8QL2TJC04\nn8yWJHUyKCRJnQwKSVIng0KS1MmgkCR1MigkSZ0MCklSJ4NCktTJoJAkdTIoJEmdDApJUieDQpLU\nyaCQJHUyKCRJnQwKSVIng0KS1MmgkCR1MigkSZ0MCklSp2mDIsmGJONJHhyqHZ5kU5LHktyZZOXQ\nsi8n2ZZkc5JTh+oXJ9na1rlo9ndFkjQX9ueM4nrg3RNqVwN3VdXJwN3AOoAk5wInVNVJwOXAta1+\nOPDnwNuAM4BrhsNFkrR4TRsUVfU9YNeE8vnAjW36xja/p35TW+8+YGWSVQyCZlNVPVdVu4FNwDmj\nd1+SNNdmeo/iqKoaB6iqZ4CjWn01sH2o3Y5Wm1h/utUkSYvc8ll+v0wyX5PUafUprB+a7rWXJGmP\nfr9Pv9+fl23NNCjGk6yqqvEkRwM/a/UdwDFD7dYAO1u9N6F+z9Rvv36G3ZKkpaHX69Hr9fbOj42N\nzdm29vfSU3j5WcFG4JI2fQlw+1D9IoAkZwK72yWqO4F3JlnZbmy/s9UkSYvctGcUSW5mcDZwZJKn\ngGuAzwC3JvkQ8BRwAUBV3ZHkvCSPA78ALm31XUn+ArifwSWnsXZTW5K0yE0bFFX1wSkWnT1F+yum\nqN8A3LC/HZMkLQ4+mS1J6mRQSJI6GRSSpE4GhSSpk0EhSepkUEiSOhkUkqROBoUkqZNBIUnqZFBI\nkjoZFJKkTgaFJKmTQSFJ6mRQSJI6GRSSpE4GhSSpk0EhSepkUEiSOhkUkqROIwVFkk8k+ackDyb5\nepIVSdYmuTfJY0m+kWR5a7siyS1JtiX5fpJjZ2cXJElzacZBkeS3gY8Cp1XV7wDLgQ8AnwU+X1Un\nA7uBy9oqlwE/r6qTgC8Cnxul45Kk+THqpadDgN9qZw2vBnYC/xb4Vlt+I/CHbfr8Ng9wG3DWiNuW\nJM2DGQdFVe0EPg88BTwNPAc8AOyuqpdasx3A6ja9Gtje1n0R2J3kiJluX5I0P5bPdMUkhzE4SziO\nQUjcCpw7SdPas8rEtxhaNsH6oelee0mS9uj3+/T7/XnZ1oyDAjgbeKKqfg6Q5O+A3wMOS7KsnVWs\nYXA5CgZnF8cAO5McAhxaVbsmf+v1I3RLkl75er0evV5v7/zY2NicbWuUexRPAWcm+c0kYXDP4WHg\nHuCC1uZi4PY2vbHN05bfPcK2JUnzZJR7FD9gcFP6R8CPGVxK+hpwNfDJJFuBI4ANbZUNwOuTbAP+\npLWTJC1yo1x6oqrGgInnOz8Bzpik7QvA+0bZniRp/vlktiSpk0EhSepkUEiSOhkUkqROBoUkqZNB\nIUnqZFBIkjoZFJKkTgaFJKmTQSFJ6mRQSJI6GRSSpE4GhSSpk0EhSepkUEiSOhkUkqROBoUkqZNB\nIUnqZFBIkjqNFBRJVia5NcmWJA8nOSPJ4Uk2JXksyZ1JVg61/3KSbUk2Jzl19O5LkubaqGcUXwLu\nqKo3Ab8LPApcDdxVVScDdwPrAJKcC5xQVScBlwPXjrhtSdI8mHFQJHkd8K+q6nqAqvpVVT0HnA/c\n2Jrd2OZpP29qbe8DViZZNdPtS5LmxyhnFMcDzya5PskDSb6W5DXAqqoaB6iqZ4CjWvvVwPah9Z9u\nNUnSIrZ8xHVPAz5SVfcn+QKDy041RftMUpui7fqh6V57SZL26Pf79Pv9ednWKEGxA9heVfe3+W8x\nCIrxJKuqajzJ0cDPhtofM7T+GmDn5G+9foRuSdIrX6/Xo9fr7Z0fGxubs23N+NJTu7y0PckbW+ks\n4GFgI3BJq10C3N6mNwIXASQ5E9i95xKVJGnxGuWMAuBjwNeTvAp4ArgUOAT4ZpIPAU8BFwBU1R1J\nzkvyOPCL1laStMiNFBRV9WPgbZMsOnuK9leMsj1J0vzzyWxJUieDQpLUyaCQJHUyKCRJnQwKSVIn\ng0KS1MmgkCR1MigkSZ0MCklSJ4NCktTJoJAkdTIoJEmdDApJUieDQpLUyaCQJHUyKCRJnQwKSVIn\ng0KS1MmgkCR1GjkokixL8kCSjW1+bZJ7kzyW5BtJlrf6iiS3JNmW5PtJjh1125KkuTcbZxQfBx4Z\nmv8s8PmqOhnYDVzW6pcBP6+qk4AvAp+bhW1LkubYSEGRZA1wHvBfhsrvAL7Vpm8E/rBNn9/mAW4D\nzhpl25Kk+THqGcUXgCuBAkhyJLCrql5qy3cAq9v0amA7QFW9COxOcsSI25ckzbHlM10xyXuA8ara\nnKS3p9xew2po2cveYmjZBOuHpnvtJUnao9/v0+/352VbMw4K4O3Ae5OcB7waeB2Dew8rkyxrZxVr\ngJ2t/Q7gGGBnkkOAQ6tq1+RvvX6EbknSK1+v16PX6+2dHxsbm7NtzfjSU1V9qqqOrarjgQuBu6vq\nj4B7gAtas4uB29v0xjZPW373TLctSZo/c/EcxdXAJ5NsBY4ANrT6BuD1SbYBf9LaSZIWuVEuPe1V\nVf8I/GOb/glwxiRtXgDeNxvbkyTNH5/MliR1MigkSZ0MCklSJ4NCktTJoJAkdTIoJEmdDApJUieD\nQpLUyaCQJHUyKCRJnQwKSVIng0KS1MmgkCR1MigkSZ0MCklSJ4NCktTJoJAkdTIoJEmdZhwUSdYk\nuTvJI0keSvKxVj88yaYkjyW5M8nKoXW+nGRbks1JTp2NHZAkza1Rzih+BXyyqt4M/EvgI0lOAa4G\n7qqqk4G7gXUASc4FTqiqk4DLgWtH6rkkaV7MOCiq6pmq2tymnwe2AGuA84EbW7Mb2zzt502t/X3A\nyiSrZrp9SdL8mJV7FEnWAqcC9wKrqmocBmECHNWarQa2D632dKtJkhaxkYMiyWuB24CPtzOLmqrp\nJLWp2kqSFonlo6ycZDmDkPibqrq9lceTrKqq8SRHAz9r9R3AMUOrrwF2Tv7O64eme+0lSdqj3+/T\n7/fnZVsjBQVwHfBIVX1pqLYRuAT4bPt5+1D9I8DfJjkT2L3nEtWvWz9ityTpla3X69Hr9fbOj42N\nzdm2ZhwUSd4O/DvgoSQ/YnAZ6VMMAuKbST4EPAVcAFBVdyQ5L8njwC+AS0ftvCRp7s04KKrqfwCH\nTLH47CnWuWKm25MkLQyfzJYkdTIoJEmdDApJUieDQpLUyaCQJHVa0kFx9NFrSXLAr6OPXrvQXZek\neTPqA3cHtfHxJ5nJt4iMj0/2bSSS9Mq0pM8oJEnTMygkSZ0MCklSJ4NCktTJoJAkdTIoJEmdDApJ\nUieD4hVqJg8T+iChpMks6QfuXslm8jChDxJKmoxnFJKkTgaFJKmTQSFJ6jTvQZHknCSPJtma5Kr5\n3r4k6cDMa1AkWQZ8BXg38BbgA0lOmc8+HEz6/f5Cd2HRcCz2cSz2cSzmx3yfUZwObKuqJ6vql8At\nwPnz3IeDhv8R7ONY7ONY7ONYzI/5DorVwPah+R2tNrK/+qv/eMDPDUiSpjffz1FM9q/zr33Y/9BD\n/+CA3vSFF7bw/PPPTvZWM+jO/viNGQXNsmWv4aWX/t8BrXPttTfwzDM/PeBtSdJsSdWB/x/eZryx\n5ExgfVWd0+avBqqqPjvUZv46JEmvIFU1J5dK5jsoDgEeA84C/hfwA+ADVbVl3johSTog83rpqape\nTHIFsInB/ZENhoQkLW7zekYhSTr4LKons5fCw3hJ1iS5O8kjSR5K8rFWPzzJpiSPJbkzycqhdb6c\nZFuSzUlOHapf3MbqsSQXLcT+jCrJsiQPJNnY5tcmubft0zeSLG/1FUluaePw/STHDr3HulbfkuRd\nC7Uvo0qyMsmtbT8eTnLGUjwuknwiyT8leTDJ19vvfskcF0k2JBlP8uBQbdaOgySntbHdmuSL+9Wp\nqloULwah9ThwHPAqYDNwykL3aw7282jg1Db9Wgb3bE4BPgv8aatfBXymTZ8L/Pc2fQZwb5s+HPhn\nYCVw2J7phd6/GYzHJ4D/Cmxs838LXNCmvwpc3qb/GPjrNv1+4JY2/WbgRwwuo65tx1AWer9mOBY3\nAJe26eXtd7ukjgvgt4EngBVDx8PFS+m4AH4fOBV4cKg2a8cBcB9wepu+A3j3tH1a6EEZGogzgX8Y\nmr8auGqh+zUP+/33wNnAo8CqVjsa2NKmrwXeP9R+C7AKuBD46lD9q8PtDoYXsAb4LtBjX1D8b2DZ\nxGMC+A5wRps+BPjZZMcJ8A972h1ML+B1wD9PUl9Sx0ULiifbP3TLgY3AO4GfLaXjgsEfzMNBMSvH\nQVv3kaH6y9pN9VpMl57m7GG8xSrJWgZ/OdzL4CAYB6iqZ4CjWrOpxmVi/WkOvvH6AnAl7QGYJEcC\nu6rqpbZ8+BjYu79V9SLwXJIjeGWMA8DxwLNJrm+X4r6W5DUsseOiqnYCnweeYtD354AHgN1L9LjY\n46hZOg5WtzYT23daTEGxXw/jvVIkeS1wG/Dxqnqeqfd14riktT2oxyvJe4DxqtrMvn0Jv75fNbRs\nooN+HIYsB04D/lNVnQb8gsFfxUvtuDiMwdf6HMfg7OK3GFxemWipHBfTOdDjYEbjspiCYgdw7ND8\nGmDnAvVlTrUbcbcBf1NVt7fyeJJVbfnRDE61YTAuxwytvmdcDvbxejvw3iRPAN8A3gF8EViZwZdH\nwsv3ae84tOdxVlbVLqYen4PNDmB7Vd3f5r/FIDiW2nFxNvBEVf28nSH8HfB7wGFL9LjYY7aOgxmN\ny2IKih8CJyY5LskKBtfONi5wn+bKdQyuE35pqLYRuKRNXwLcPlS/CPY+2b67nYLeCbyzfVLmcAbX\nce+c+67Pjqr6VFUdW1XHM/hd311VfwTcA1zQml3My8fh4jZ9AXD3UP3C9umXNwAnMniQ86DSfqfb\nk7yxlc4CHmaJHRcMLjmdmeQ3k4R947DUjouJZ9ezchy0y1b/N8npbXwvGnqvqS30TZsJN3DOYfAp\noG3A1Qvdnznax7cDLzL4VNePGFx/PQc4Arir7f93gcOG1vkKg09t/Bg4bah+SRurrcBFC71vI4zJ\nv2Hfzew3MPhUxlYGn3R5Vav/BvDNtr/3AmuH1l/XxmcL8K6F3p8RxuF3GfzBtBn4NoNPrCy54wK4\npv0uHwRuZPApyCVzXAA3M/gr/wUGwXkpg5v7s3IcAG8FHmrLvrQ/ffKBO0lSp8V06UmStAgZFJKk\nTgaFJKmTQSFJ6mRQSJI6GRSSpE4GhSSpk0EhSer0/wFrDF0p2URyvAAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c477fc810>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "c = 4\n",
    "incorrect = filter(lambda (x,y,q,r,rv): not rv and y == c, train)\n",
    "print len(incorrect)\n",
    "dists = map(lambda (x,y,q,r,rv): distance(r, c), incorrect)\n",
    "dists.sort()\n",
    "plt.hist(dists, bins=24);"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[184.56435192094924, 184.56435192094924, 184.56435192094924, 184.56435192094924, 184.56435192094924, 184.56435192094924, 184.56435192094924, 184.56435192094924, 184.56435192094924, 184.56435192094924]\n"
     ]
    }
   ],
   "source": [
    "c = 4\n",
    "dists = map(lambda (x,y,q,r,rv): distance(r, c), incorrect)\n",
    "dists.sort()\n",
    "print dists[500:510]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[(u'../data/glebs-train/train/Motorola-X/24800884010_e8a4414391_o.jpg',\n",
       "  4,\n",
       "  95,\n",
       "  [2432, 4320],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Motorola-X/24801148050_c2e5e2bc3b_o.jpg',\n",
       "  4,\n",
       "  95,\n",
       "  [2432, 4320],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Motorola-X/24886935300_260bef7313_o.jpg',\n",
       "  4,\n",
       "  95,\n",
       "  [2432, 4320],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Motorola-X/24908486461_4b67737261_o.jpg',\n",
       "  4,\n",
       "  95,\n",
       "  [2432, 4320],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Motorola-X/24917793486_a84a24dd28_o.jpg',\n",
       "  4,\n",
       "  95,\n",
       "  [4320, 2432],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Motorola-X/24978362412_198f73d5df_o.jpg',\n",
       "  4,\n",
       "  95,\n",
       "  [2432, 4320],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Motorola-X/25003330771_e554aa4e98_o.jpg',\n",
       "  4,\n",
       "  95,\n",
       "  [4320, 2432],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Motorola-X/25003677326_f67c4fb2fa_o.jpg',\n",
       "  4,\n",
       "  95,\n",
       "  [2432, 4320],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Motorola-X/25033310134_a18f64e4e0_o.jpg',\n",
       "  4,\n",
       "  95,\n",
       "  [4320, 2432],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Motorola-X/25033332724_18718bc9cc_o.jpg',\n",
       "  4,\n",
       "  95,\n",
       "  [2432, 4320],\n",
       "  False)]"
      ]
     },
     "execution_count": 47,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "incorrect.sort(key=lambda (x,y,q,r,rv): distance(r,4))\n",
    "incorrect[500:510]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### add images with Droid Maxx resolution to its class\n",
    "### UPD: no"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "new_train = []\n",
    "for x,y,q,r,rv in train:\n",
    "    if y == 4 and r and distance(r, 2) < 0.01:\n",
    "        new_train.append((x,2,q,r,rv))\n",
    "    else:\n",
    "        new_train.append((x,y,q,r,rv))\n",
    "train = new_train"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### filter Moto X"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "12653\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[1050, 874, 2042, 1002, 597, 1514, 1495, 1385, 1478, 1216]"
      ]
     },
     "execution_count": 49,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYAAAAEACAYAAAC6d6FnAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c4742e450>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "c = 4\n",
    "train = filter(lambda (x,y,q,r,rv): y != c or (y == c and (rv or (not rv and distance(r, c) < 1.))), train)\n",
    "print len(train)\n",
    "TRAIN_CLASSES = map(int, plt.hist(zip(*train)[1])[0])\n",
    "TRAIN_CLASSES"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### class 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1240\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYcAAAEACAYAAABYq7oeAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nFQAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c460ae4d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "c = 2\n",
    "incorrect = filter(lambda (x,y,q,r,rv): not rv and y == c, train)\n",
    "print len(incorrect)\n",
    "dists = map(lambda (x,y,q,r,rv): distance(r, c), incorrect)\n",
    "dists.sort()\n",
    "plt.hist(dists, bins=24);"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]\n"
     ]
    }
   ],
   "source": [
    "c = 2\n",
    "dists = map(lambda (x,y,q,r,rv): distance(r, c), incorrect)\n",
    "dists.sort()\n",
    "print dists[:30]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "12550\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[1050, 874, 1939, 1002, 597, 1514, 1495, 1385, 1478, 1216]"
      ]
     },
     "execution_count": 52,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYAAAAEACAYAAAC6d6FnAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c47518990>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "c = 2\n",
    "train = filter(lambda (x,y,q,r,rv): y != c or (y == c and (rv or (not rv and distance(r, c) < 0.1))), train)\n",
    "print len(train)\n",
    "TRAIN_CLASSES = map(int, plt.hist(zip(*train)[1])[0])\n",
    "TRAIN_CLASSES"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### class 9"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "781\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYMAAAEACAYAAABRQBpkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c46e54d10>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "c = 9\n",
    "incorrect = filter(lambda (x,y,q,r,rv): not rv and y == c, train)\n",
    "print len(incorrect)\n",
    "dists = map(lambda (x,y,q,r,rv): distance(r, c), incorrect)\n",
    "dists.sort()\n",
    "plt.hist(dists, bins=24);"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "12428\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[1050, 874, 1939, 1002, 597, 1514, 1495, 1385, 1478, 1094]"
      ]
     },
     "execution_count": 54,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYAAAAEACAYAAAC6d6FnAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nowwASWrU/wOTaxOrlgY6vQAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c46e49250>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "c = 9\n",
    "train = filter(lambda (x,y,q,r,rv): y != c or (y == c and (rv or (not rv and distance(r, c) < 1.))), train)\n",
    "print len(train)\n",
    "TRAIN_CLASSES = map(int, plt.hist(zip(*train)[1])[0])\n",
    "TRAIN_CLASSES"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### class 7"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1082\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYAAAAEACAYAAAC6d6FnAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c460f9050>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "c = 7\n",
    "incorrect = filter(lambda (x,y,q,r,rv): not rv and y == c, train)\n",
    "print len(incorrect)\n",
    "dists = map(lambda (x,y,q,r,rv): distance(r, c), incorrect)\n",
    "dists.sort()\n",
    "plt.hist(dists, bins=24);"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[(u'../data/glebs-train/train/Sony-NEX-7/24513930567_55f14618b1_o.jpg',\n",
       "  7,\n",
       "  100,\n",
       "  [3744, 5617],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/38142165914_cda10d75f7_o.jpg',\n",
       "  7,\n",
       "  99,\n",
       "  [3540, 6000],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/38685299801_bef6ff20fb_o.jpg',\n",
       "  7,\n",
       "  96,\n",
       "  [3750, 5625],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/37541197882_14bca12c58_o.jpg',\n",
       "  7,\n",
       "  99,\n",
       "  [5635, 3737],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/38653248512_b91ed4e845_o.jpg',\n",
       "  7,\n",
       "  96,\n",
       "  [5630, 3753],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/37967530344_a1f8336856_o.jpg',\n",
       "  7,\n",
       "  96,\n",
       "  [5646, 3764],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/37112580226_c98fb4c5c2_o.jpg',\n",
       "  7,\n",
       "  96,\n",
       "  [5600, 4000],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/38110805492_fa9f7b999a_o.jpg',\n",
       "  7,\n",
       "  93,\n",
       "  [3603, 6000],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/37967044094_7f21ef19e5_o.jpg',\n",
       "  7,\n",
       "  96,\n",
       "  [5697, 3798],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/26909773289_d85a255141_o.jpg',\n",
       "  7,\n",
       "  96,\n",
       "  [3809, 5714],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/38653228162_597508f187_o.jpg',\n",
       "  7,\n",
       "  96,\n",
       "  [3813, 5720],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/37875538205_8d31186cf8_o.jpg',\n",
       "  7,\n",
       "  99,\n",
       "  [5710, 3832],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/38653315692_653366e3b9_o.jpg',\n",
       "  7,\n",
       "  96,\n",
       "  [3821, 5731],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/38814039471_3e9b8b18b1_o.jpg',\n",
       "  7,\n",
       "  99,\n",
       "  [3835, 5752],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/24813019838_94e5739e02_o.jpg',\n",
       "  7,\n",
       "  96,\n",
       "  [3854, 5781],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/26908977599_cea849c600_o.jpg',\n",
       "  7,\n",
       "  96,\n",
       "  [3854, 5781],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/38685792691_ab7e0b4847_o.jpg',\n",
       "  7,\n",
       "  96,\n",
       "  [5789, 3859],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/37179667870_fa1a221716_o.jpg',\n",
       "  7,\n",
       "  93,\n",
       "  [3757, 5988],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/37797496805_e318a775bc_o.jpg',\n",
       "  7,\n",
       "  96,\n",
       "  [3865, 5798],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/37797917115_65e841023e_o.jpg',\n",
       "  7,\n",
       "  96,\n",
       "  [5798, 3865],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/38685830931_7ba83a9700_o.jpg',\n",
       "  7,\n",
       "  96,\n",
       "  [3865, 5798],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/37875536925_dfd55732fa_o.jpg',\n",
       "  7,\n",
       "  99,\n",
       "  [3810, 5878],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/37875545235_fac44a3c7d_o.jpg',\n",
       "  7,\n",
       "  99,\n",
       "  [3812, 5908],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/26909041769_f03c692a91_o.jpg',\n",
       "  7,\n",
       "  96,\n",
       "  [3888, 5832],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/24544949827_3b53c0085e_o.jpg',\n",
       "  7,\n",
       "  100,\n",
       "  [3888, 5833],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/26986545869_c0cb61bef1_o.jpg',\n",
       "  7,\n",
       "  99,\n",
       "  [3833, 5912],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/38685813721_c9fc810953_o.jpg',\n",
       "  7,\n",
       "  96,\n",
       "  [5848, 3899],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/26909017229_5c434145b3_o.jpg',\n",
       "  7,\n",
       "  96,\n",
       "  [5848, 3899],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/26909532029_0cfe6aeeb7_o.jpg',\n",
       "  7,\n",
       "  96,\n",
       "  [5848, 3899],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/24812938958_b12e9a1c23_o.jpg',\n",
       "  7,\n",
       "  96,\n",
       "  [5849, 3899],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/24812936948_7574fa719b_o.jpg',\n",
       "  7,\n",
       "  96,\n",
       "  [5856, 3904],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/26986562239_38e8c7b642_o.jpg',\n",
       "  7,\n",
       "  99,\n",
       "  [5860, 3904],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/38533117315_53ba74e244_o.jpg',\n",
       "  7,\n",
       "  100,\n",
       "  [3911, 5867],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/24513794497_ee37c03cd9_o.jpg',\n",
       "  7,\n",
       "  100,\n",
       "  [3922, 5884],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/26626262579_1b3caef0f6_o.jpg',\n",
       "  7,\n",
       "  99,\n",
       "  [4000, 5887],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/24813154588_60f1db51d2_o.jpg',\n",
       "  7,\n",
       "  96,\n",
       "  [5916, 3944],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/38652752822_012f38046e_o.jpg',\n",
       "  7,\n",
       "  96,\n",
       "  [5950, 3967],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/36601955143_61e746acc8_o.jpg',\n",
       "  7,\n",
       "  99,\n",
       "  [5981, 3987],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/38685802261_a64f2ff2e9_o.jpg',\n",
       "  7,\n",
       "  96,\n",
       "  [5992, 3995],\n",
       "  False),\n",
       " (u'../data/esato-com/train/Sony-NEX-7/201308281543a07Z13.jpg',\n",
       "  7,\n",
       "  95,\n",
       "  [4000, 6000],\n",
       "  False),\n",
       " (u'../data/esato-com/train/Sony-NEX-7/201308281542u833tY.jpg',\n",
       "  7,\n",
       "  95,\n",
       "  [4000, 6000],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/37593141095_b2c6bfbf69_o.jpg',\n",
       "  7,\n",
       "  95,\n",
       "  [4000, 6000],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/37593143425_d0f70773d0_o.jpg',\n",
       "  7,\n",
       "  95,\n",
       "  [4000, 6000],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/37593144275_659d2cd101_o.jpg',\n",
       "  7,\n",
       "  95,\n",
       "  [4000, 6000],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/37593154265_f155f028fc_o.jpg',\n",
       "  7,\n",
       "  95,\n",
       "  [4000, 6000],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/37593155425_0cc03bfe4b_o.jpg',\n",
       "  7,\n",
       "  95,\n",
       "  [4000, 6000],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/37593156215_e813738885_o.jpg',\n",
       "  7,\n",
       "  95,\n",
       "  [4000, 6000],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/37593157235_5c58eb5e1a_o.jpg',\n",
       "  7,\n",
       "  95,\n",
       "  [4000, 6000],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/37593159505_63cc78c811_o.jpg',\n",
       "  7,\n",
       "  95,\n",
       "  [4000, 6000],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Sony-NEX-7/37593162095_d092743582_o.jpg',\n",
       "  7,\n",
       "  95,\n",
       "  [4000, 6000],\n",
       "  False)]"
      ]
     },
     "execution_count": 56,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dists[500:600]\n",
    "incorrect.sort(key=lambda (x,y,q,r,rv): -distance(r,c))\n",
    "incorrect[550:600]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "11839\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[1050, 874, 1939, 1002, 597, 1514, 1495, 796, 1478, 1094]"
      ]
     },
     "execution_count": 57,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYAAAAEACAYAAAC6d6FnAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAD0xJREFUeJzt3X+MHGd9x/H3h7jEgYBrgX0nYrCBUEor0RAJSKFVD0GI\nE9qmqpQqFCk4ERJSSUGtVJFQqXZKK0ElWgW1FUKE4CBSfkSt4koRsaKwQlRqQiAuoc4vhILtpD5+\nNEkxiIgf3/6xY7I57nLnO3tm7ef9klaefXZ2nu/Onfcz88w+e6kqJEntecbQBUiShmEASFKjDABJ\napQBIEmNMgAkqVEGgCQ1atkASLIlye1J9ie5J8m7u/aNSfYmuT/JrUk2TDznw0keTLIvyTkT7W9P\n8kD3nMtOzEuSJK1ElpsHkGQWmK2qfUnOBL4CXAxcDnyvqv4uyXuBjVV1VZILgSur6i1JXgtcW1Xn\nJdkI3AWcC6TbzrlV9fiJe3mSpKUsewZQVYeral+3fAS4F9jCOAR2d6vt7u7T/XtDt/4dwIYkM8AF\nwN6qeryqHgP2AtuP42uRJB2DY7oGkGQbcA7wn8BMVc3DOCSAzd1qZwEHJ552qGtb2P5w1yZJGsCK\nA6Ab/rkJeE93JrDU2FEWuV+LtPM025AknWDrVrJSknWM3/w/WVU3d83zSWaqar67TvDtrv0Q8MKJ\np28BHuna5xa0f2GRvgwFSVqFqlrsQHtJKz0D+Diwv6qunWjbA+zolncAN0+0XwaQ5DzgsW6o6Fbg\n/CQbugvC53dtv6CqvFWxc+fOwWuYlpv7wn3hvnj622osewaQ5PXA24B7ktzNeNjmfcAHgc8muQI4\nAFzSvXnfkuSiJN8AfsD400JU1aNJ3s/4k0AFXFPji8GSpAEsGwBV9R/AaUs8/KYlnnPlEu2fAD6x\nwtokSSeQM4Gn2Nzc3NAlTA33xZPcF09yX6zNshPB+pakpq0mSZp2SagTdBFYJ9Ds7DaS9H6bnd02\n9EuXNCDPAKZAcnSqRO89r/rTA5Kmi2cAkqQVMwAkqVEGgCQ1ygCQpEYZAJLUKANAkhplAEhSowwA\nSWqUASBJjTIAJKlRBoAkNcoAkKRGGQCS1CgDQJIaZQBIUqMMAElqlAEgSY0yACSpUQaAJDXKAJCk\nRhkAktQoA0CSGmUASFKjDABJapQBIEmNMgAkqVEGgCQ1ygCQpEYZAJLUKANAkhplAEhSowwASWqU\nASBJjTIAJKlRBoAkNcoAkKRGGQCS1CgDQJIaZQBIUqMMAElqlAEgSY0yACSpUQaAJDVq2QBIcl2S\n+SRfm2jbmeRQkq92t+0Tj12d5MEk9yZ580T79iT3JXkgyXuP/0uRJB2LVNXTr5D8FnAEuKGqXtm1\n7QS+X1V/v2DdVwA3Aq8GtgC3AS8DAjwAvBF4BPgycGlV3bdIf7VcTaeaJMAQrzm0tq+lU1USqirH\n8px1y61QVV9KsnWx/hZpuxj4dFX9BHgoyYPAa7p1H6yqb3WFfrpb9xcCQJLUj7VcA3hXkn1JPpZk\nQ9d2FnBwYp2Hu7aF7Ye6Nqk3s7PbSNL7bXZ229AvXVrUsmcAS/hn4K+rqpL8DfAh4B0sflZQLB40\nS4497Nq16+fLc3NzzM3NrbJM6Unz899iiKG2+fljOiuXVmQ0GjEajda0jWWvAQB0Q0D/fvQawFKP\nJbkKqKr6YPfY54GdjINhV1Vt79qfst6C7XkNoL+em7oGMNx+Xg880XuvMzNbOXz4od771TBWcw1g\npUNAYeLoPsnsxGN/CHy9W94DXJrkmUleDJwN3Mn4ou/ZSbYmeSZwabeu1IAnGAdPv7fxGY+0tGWH\ngJLcCMwBz0tygPER/RuSnAP8DHgIeCdAVe1P8llgP/Bj4E+6w/mfJrkS2Ms4dK6rqnuP/8uRNA1m\nZ7cNEkCe9RybFQ0B9ckhoF57dgion54H63eon6+/0/07kUNAkqRTjAEgSY0yACSpUQaAJDXKAJCk\nRhkAktQoA0CSGmUASFKjDABJapQBIEmNMgAkqVEGgCQ1ygCQpEYZAJLUKANAkhplAEhSowwASWqU\nASBJjTIAJKlRBoAkNcoAkKRGGQCS1CgDQJIaZQBIUqMMAElqlAEgSY0yACSpUQaAJDXKAJCkRhkA\nktQoA0CSGmUASFKjDABJapQBIEmNMgAkqVEGgCQ1ygCQpEYZAJLUKANAkhplAEhSowwASWrUuqEL\nkKTj53SS9N7rzMxWDh9+qPd+1ypVNXQNT5Gkpq2mE238CzvEaw4t7esh93NrP1/39QAVJFTVMaWf\nQ0CS1CgDQJIaNZXXAJ7//K2997l582a+8pUvcsYZZ/TetyQNYSoD4Hvf+2LvfR458iqOHDliAEhq\nxrIBkOQ64HeB+ap6Zde2EfgMsBV4CPijqnq8e+zDwIXAD4AdVbWva3878JeMr9D8bVXdsHSv/Z8B\nJKf13qckDWkl1wCuBy5Y0HYVcFtVvRy4HbgaIMmFwEur6mXAO4GPdO0bgb8CXg28FtiZZMNxeQWS\npFVZNgCq6kvAowuaLwZ2d8u7u/tH22/onncHsCHJDOMA2VtVj1fVY8BeYPvay5ckrdZqPwW0uarm\nAarqMLC5az8LODix3qGubWH7w12bJGkgx/si8MJJCEdnZSw2OeFpZk3smlie6246/pw1KZ2sRqMR\no9FoTdtYbQDMJ5mpqvkks8C3u/ZDwAsn1tsCPNK1zy1o/8LSm9+1yrJ0bJ5giFmT8/P9h450qpmb\nm2Nubu7n96+55ppj3sZKh4DCU4/i9wA7uuUdwM0T7ZcBJDkPeKwbKroVOD/Jhu6C8PldmyRpICv5\nGOiNjI/en5fkALAT+ADwuSRXAAeASwCq6pYkFyX5BuOPgV7etT+a5P3AXYwPOa/pLgZLkgYylV8G\nN8SwxPr1mzhwYD+bNm3qvW+/OKunXt3P/fXsvu6/Ar8MTpK0UgaAJDXKAJCkRhkAktQoA0CSGmUA\nSFKjDABJapQBIEmNMgAkqVHOBO6sX7+J5zznDL7znYPLr3xCOGvyhPfq7NT+enZf91/BKmYCT+Xf\nBB7K+M1/qF9aSeqXQ0CS1CgDQJIaZQBIUqMMAElqlAEgSY0yACSpUQaAJDXKeQDSKev0bkKWtDgD\nQDplPcEwExvByY0nB4eAJKlRBoAkNcoAkKRGGQCS1CgDQJIaZQBIUqMMAElqlAEgSY0yACSpUQaA\nJDXKAJCkRhkAktQoA0CSGmUASFKjDABJapQBIEmNMgAkqVEGgCQ1ygCQpEYZAJLUKP8ovAZwOol/\nNFwamgGgATwB1AD9GjrSJIeAJKlRBoAkNcoAkKRGGQCS1CgDQJIataYASPJQkv9KcneSO7u2jUn2\nJrk/ya1JNkys/+EkDybZl+SctRYvSVq9tZ4B/AyYq6pXVdVrurargNuq6uXA7cDVAEkuBF5aVS8D\n3gl8ZI19S5LWYK0BkEW2cTGwu1ve3d0/2n4DQFXdAWxIMrPG/iVJq7TWACjg1iRfTvKOrm2mquYB\nquowsLlrPws4OPHch7s2SdIA1joT+HVVdTjJJmBvkvtZeornYtMwl1h318TyXHeTJB01Go0YjUZr\n2kaqjs+U/CQ7gSPAOxhfF5hPMgt8oapekeQj3fJnuvXvA37n6NnCxHZqiK8JWL9+Ez/60XcZ7isK\n7Nd+T5V+h+x7uH6P13vpqitIqKpj+r6TVQ8BJXlWkjO75WcDbwbuAfYAO7rVdgA3d8t7gMu69c8D\nHlv45i9J6s9ahoBmgH8bH7GzDvhUVe1Nchfw2SRXAAeASwCq6pYkFyX5BvAD4PI11i5JWoPjNgR0\nvDgEZL/2e7L3O2TfDgEdy3OcCSxJjTIAJKlRBoAkNcoAkKRGGQCS1CgDQJIaZQBIUqMMAElqlAEg\nSY0yACSpUQaAJDXKAJCkRhkAktQoA0CSGmUASFKjDABJapQBIEmNMgAkqVEGgCQ1ygCQpEYZAJLU\nKANAkhplAEhSowwASWqUASBJjTIAJKlRBoAkNcoAkKRGGQCS1CgDQJIaZQBIUqMMAElqlAEgSY0y\nACSpUQaAJDXKAJCkRhkAktQoA0CSGmUASFKjDABJapQBIEmNMgAkqVEGgCQ1ygCQpEYZAJLUKANA\nkhrVewAk2Z7kviQPJHlv3/1LksZ6DYAkzwD+EbgA+HXgrUl+tc8aTi6joQuYIqOhC5gio6ELmCKj\noQs4qfV9BvAa4MGq+lZV/Rj4NHBxzzWcREZDFzBFRkMXMEVGQxcwRUZDF3BS6zsAzgIOTtw/1LVJ\nknq2ruf+skhbLWx47nN/r4dSnuqHP/x+731K0pBS9Qvvvyeus+Q8YFdVbe/uXwVUVX1wYp3+CpKk\nU0hVLXaQvaS+A+A04H7gjcD/AHcCb62qe3srQpIE9DwEVFU/TXIlsJfx9YfrfPOXpGH0egYgSZoe\nUzUT2EliY0m2JLk9yf4k9yR599A1DS3JM5J8NcmeoWsZUpINST6X5N4k/53ktUPXNJQkf5bk60m+\nluRTSZ45dE19SXJdkvkkX5to25hkb5L7k9yaZMNy25maAHCS2FP8BPjzqvo14DeBdzW8L456D7B/\n6CKmwLXALVX1CuA3gCaHUJO8APhT4NyqeiXj4exLh62qV9czfq+cdBVwW1W9HLgduHq5jUxNAOAk\nsZ+rqsNVta9bPsL4P3mz8yWSbAEuAj42dC1DSvIc4Ler6nqAqvpJVf3fwGUN6TTg2UnWAc8CHhm4\nnt5U1ZeARxc0Xwzs7pZ3A3+w3HamKQCcJLaIJNuAc4A7hq1kUP8A/AWLzBlpzEuA7ya5vhsO+2iS\nM4YuaghV9QjwIeAA8DDwWFXdNmxVg9tcVfMwPogENi33hGkKgBVNEmtJkjOBm4D3dGcCzUnyFmC+\nOyMKi/+etGIdcC7wT1V1LvBDxqf9zUnyy4yPeLcCLwDOTPLHw1Z18pmmADgEvGji/hYaOqVbqDut\nvQn4ZFXdPHQ9A3o98PtJvgn8C/CGJDcMXNNQDgEHq+qu7v5NjAOhRW8CvllV/1tVPwX+FXjdwDUN\nbT7JDECSWeDbyz1hmgLgy8DZSbZ2V/MvBVr+xMfHgf1Vde3QhQypqt5XVS+qqpcw/p24vaouG7qu\nIXSn9weT/ErX9EbavTB+ADgvyfokYbwvWrsgvvCMeA+wo1t+O7DsgWPf3wW0JCeJPSnJ64G3Afck\nuZvxUNj7qurzw1amKfBu4FNJfgn4JnD5wPUMoqruTHITcDfw4+7fjw5bVX+S3AjMAc9LcgDYCXwA\n+FySKxgH5CXLbseJYJLUpmkaApIk9cgAkKRGGQCS1CgDQJIaZQBIUqMMAElqlAEgSY0yACSpUf8P\n28wZq4YEFDEAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c4683f9d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "c = 7\n",
    "train = filter(lambda (x,y,q,r,rv): y != c or (y == c and (rv or (not rv and distance(r, c) < 1.))), train)\n",
    "print len(train)\n",
    "TRAIN_CLASSES = map(int, plt.hist(zip(*train)[1])[0])\n",
    "TRAIN_CLASSES"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### class 5"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1229\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYAAAAEACAYAAAC6d6FnAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAEnqlAEgSZ0yACSpU/8PgzFbxEc3Ab0AAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c4683f290>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "c = 5\n",
    "incorrect = filter(lambda (x,y,q,r,rv): not rv and y == c, train)\n",
    "print len(incorrect)\n",
    "dists = map(lambda (x,y,q,r,rv): distance(r, c), incorrect)\n",
    "dists.sort()\n",
    "plt.hist(dists, bins=24);"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]"
      ]
     },
     "execution_count": 59,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dists[:9]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[1783.1850717185807,\n",
       " 2240.1901704989245,\n",
       " 2240.1901704989245,\n",
       " 2240.1901704989245,\n",
       " 2240.1901704989245,\n",
       " 2240.1901704989245,\n",
       " 2240.1901704989245,\n",
       " 2240.1901704989245,\n",
       " 2240.1901704989245,\n",
       " 2240.1901704989245,\n",
       " 2240.1901704989245,\n",
       " 2240.1901704989245,\n",
       " 2240.1901704989245,\n",
       " 2240.1901704989245,\n",
       " 2246.0650035116973,\n",
       " 2377.6433710714482,\n",
       " 2389.5145950590049,\n",
       " 2465.6481906387212,\n",
       " 2533.3432455946431,\n",
       " 2533.3432455946431,\n",
       " 2533.3432455946431,\n",
       " 2533.3432455946431,\n",
       " 2619.1983506408978,\n",
       " 2643.598305340658,\n",
       " 3210.4834526905756,\n",
       " 3915.9975740544069,\n",
       " 4759.2541432455573,\n",
       " 4791.6948984675555,\n",
       " 4861.0744697031741,\n",
       " 5029.1315353647296]"
      ]
     },
     "execution_count": 60,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dists[-30:]#, dists[230:260]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[(u'../data/glebs-train/train/Samsung-Galaxy-Note3/35117221375_929f93ce27_o.jpg',\n",
       "  5,\n",
       "  96,\n",
       "  [1836, 3264],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Samsung-Galaxy-Note3/35117221805_b6d7eaaf7a_o.jpg',\n",
       "  5,\n",
       "  96,\n",
       "  [1836, 3264],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Samsung-Galaxy-Note3/35117222025_ef03a23839_o.jpg',\n",
       "  5,\n",
       "  96,\n",
       "  [1836, 3264],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Samsung-Galaxy-Note3/35117222445_45b64bb3fe_o.jpg',\n",
       "  5,\n",
       "  96,\n",
       "  [1836, 3264],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Samsung-Galaxy-Note3/35117222905_5aa70485e9_o.jpg',\n",
       "  5,\n",
       "  96,\n",
       "  [1836, 3264],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Samsung-Galaxy-Note3/35231019683_321c9f0a34_o.jpg',\n",
       "  5,\n",
       "  93,\n",
       "  [3264, 1836],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Samsung-Galaxy-Note3/35569476706_12ecc6623c_o.jpg',\n",
       "  5,\n",
       "  96,\n",
       "  [1836, 3264],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Samsung-Galaxy-Note3/35754448775_772a29eaca_o.jpg',\n",
       "  5,\n",
       "  96,\n",
       "  [1836, 3264],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Samsung-Galaxy-Note3/35999662456_1cbe0bb2b5_o.jpg',\n",
       "  5,\n",
       "  93,\n",
       "  [1836, 3264],\n",
       "  False),\n",
       " (u'../data/glebs-train/train/Samsung-Galaxy-Note3/36041325785_0c7c5e0d6b_o.jpg',\n",
       "  5,\n",
       "  93,\n",
       "  [1836, 3264],\n",
       "  False)]"
      ]
     },
     "execution_count": 61,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "incorrect.sort(key=lambda (x,y,q,r,rv): -distance(r,5))\n",
    "incorrect[130:140]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### same ratio as in training set, keep"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "11774\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[1050, 874, 1939, 1002, 597, 1449, 1495, 796, 1478, 1094]"
      ]
     },
     "execution_count": 62,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYAAAAEACAYAAAC6d6FnAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c4b9c6c50>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "c = 5\n",
    "train = filter(lambda (x,y,q,r,rv): y != c or (y == c and (rv or (not rv and ((distance(r, c) < 1.) or (773. < distance(r, c) < 992.))))), train)\n",
    "print len(train)\n",
    "TRAIN_CLASSES = map(int, plt.hist(zip(*train)[1])[0])\n",
    "TRAIN_CLASSES"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### class 0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "33\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXcAAAEACAYAAABI5zaHAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c460ae350>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "c = 0\n",
    "incorrect = filter(lambda (x,y,q,r,rv): not rv and y == c, train)\n",
    "print len(incorrect)\n",
    "dists = map(lambda (x,y,q,r,rv): distance(r, c), incorrect)\n",
    "dists.sort()\n",
    "plt.hist(dists, bins=24);"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### simply remove all incorrect"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "11741\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[1017, 874, 1939, 1002, 597, 1449, 1495, 796, 1478, 1094]"
      ]
     },
     "execution_count": 64,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYAAAAEACAYAAAC6d6FnAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nJDXKAJCkRhkAktQoA0CSGvX/wHNHJw6DahcAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c49a15a50>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "c = 0\n",
    "train = filter(lambda (x,y,q,r,rv): y != c or (y == c and rv), train)\n",
    "print len(train)\n",
    "TRAIN_CLASSES = map(int, plt.hist(zip(*train)[1])[0])\n",
    "TRAIN_CLASSES"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### class 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "206\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYAAAAEACAYAAAC6d6FnAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c49a41390>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "c = 1\n",
    "incorrect = filter(lambda (x,y,q,r,rv): not rv and y == c, train)\n",
    "print len(incorrect)\n",
    "dists = map(lambda (x,y,q,r,rv): distance(r, c), incorrect)\n",
    "dists.sort()\n",
    "plt.hist(dists, bins=24);"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[40.0,\n",
       " 45.254833995939045,\n",
       " 45.254833995939045,\n",
       " 45.254833995939045,\n",
       " 45.254833995939045,\n",
       " 45.254833995939045,\n",
       " 45.254833995939045,\n",
       " 45.254833995939045,\n",
       " 45.254833995939045,\n",
       " 45.254833995939045,\n",
       " 45.254833995939045,\n",
       " 45.254833995939045,\n",
       " 45.254833995939045,\n",
       " 45.254833995939045,\n",
       " 45.254833995939045,\n",
       " 45.254833995939045,\n",
       " 45.254833995939045,\n",
       " 45.254833995939045,\n",
       " 45.254833995939045,\n",
       " 45.254833995939045]"
      ]
     },
     "execution_count": 66,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dists[:20]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "11624\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[1017, 757, 1939, 1002, 597, 1449, 1495, 796, 1478, 1094]"
      ]
     },
     "execution_count": 67,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYAAAAEACAYAAAC6d6FnAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c450274d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "c = 1\n",
    "train = filter(lambda (x,y,q,r,rv): y != c or (y == c and (rv or (not rv and (distance(r, c) < 46.)))), train)\n",
    "print len(train)\n",
    "TRAIN_CLASSES = map(int, plt.hist(zip(*train)[1])[0])\n",
    "TRAIN_CLASSES"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### class 3"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "175\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYAAAAEACAYAAAC6d6FnAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAEyNJREFUeJzt3V2wZWV95/HvDxpQfGnACd2lrbTEATVVI8NUAqOT8USI\ntKQGzAUTdVK2kotUoq3RmmijF3RfCTqWCZWXKSuEaR3FCEwCVeMMhIKdKqfSGAtaEDpIBrV5GQ7l\nCFQlU2XE/s/FWqezac+hT++99tmHfr6fql299rPWXut/nrP3/q1nvZxOVSFJas9x8y5AkjQfBoAk\nNcoAkKRGGQCS1CgDQJIaZQBIUqOOGABJrk2ymOTesbZPJ9mfZF+Sm5K8fGzeFUke6ue/fVaFS5Km\ns5oRwHXARYe13Qb8XFWdAzwEXAGQ5I3AvwfeALwD+KMkGa5cSdJQjhgAVfV14KnD2m6vqoP9073A\nln76EuArVfVsVX2PLhx+YbhyJUlDGeIcwOXA1/rpVwGPjM17rG+TJK0zUwVAkk8CP66q65eallnM\nvzUhSevQhklfmGQ7cDHwtrHmR4FXjz3fAjy+wusNBkmaQFUNcm51tSOAMLZ3n2Qb8DHgkqr60dhy\ntwDvSnJiktcCrwO+sdJKq+o5j/POuwj4H3SDhtU9Tjrpg1xzzTU/ta5JHldeeeUg6xn6sR7rsiZr\naqGu9VjTkI44AkjyZWABeEWSA8CVwCeAE4G/7C/y2VtVv11VDyT5KvAA8GPgt2voiiVJgzhiAFTV\ne5Zpvu55lv8U8KlpipIkzZ53Ao9ZWFiYdwnLWo91WdPqWNPqrce61mNNQ8q8jtAk+amjQ+efv427\n7vodYNuq13PSSTv4zGfOYseOHQNXKEnrTxJqjU8CS5KOMQaAJDXKAJCkRhkAktQoA0CSGmUASFKj\nDABJapQBIEmNMgAkqVEGgCQ1ygCQpEYZAJLUKANAkhplAEhSowwASWqUASBJjTIAJKlRBoAkNcoA\nkKRGGQCS1CgDQJIaZQBIUqMMAElqlAEgSY0yACSpUQaAJDXqiAGQ5Noki0nuHWs7NcltSR5McmuS\njWPzrknyUJJ9Sc6ZVeGSpOmsZgRwHXDRYW07gdur6mzgDuAKgCTvAH62qv458JvAfx6wVknSgI4Y\nAFX1deCpw5ovBfb003v650vtX+hfdxewMcmmYUqVJA1p0nMAp1fVIkBVPQGc3re/CnhkbLnH+jZJ\n0joz9EngLNNWA29DkjSADRO+bjHJpqpaTLIZeLJvfxR49dhyW4DHV1rJrl27Dk0vLCxMWIokHbtG\noxGj0Wgm615tAITn7t3fArwPuLr/9+ax9g8Af5bkfODppUNFyxkPgM5VqyxHktqwsLDwnB3k3bt3\nD7buIwZAki8DC8ArkhwArqT7pr4hyeXAAeAygKr6WpKLk/wd8A/A+werVJI0qCMGQFW9Z4VZF66w\n/AenqkiStCa8E1iSGmUASFKjDABJapQBIEmNMgAkqVEGgCQ1ygCQpEYZAJLUKANAkhplAEhSowwA\nSWqUASBJjTIAJKlRBoAkNcoAkKRGGQCS1CgDQJIaZQBIUqMMAElqlAEgSY0yACSpUQaAJDXKAJCk\nRhkAktQoA0CSGmUASFKjDABJapQBIEmNMgAkqVFTBUCSjyT5dpJ7k3wpyYlJtibZm+TBJNcn2TBU\nsZKk4UwcAEleCewAzq2qfwFsAN4NXA18tqrOBp4GfmOIQiVJw5r2ENDxwEv6vfwXA48DvwTc1M/f\nA/zqlNuQJM3AxAFQVY8DnwUOAI8BzwB3A09X1cF+sUeBV05bpCRpeBMfn09yCnApcAbdl/8NwDuW\nWbRWWseuXbsOTS8sLExaiiQds0ajEaPRaCbrnuYE7YXAw1X1Q4Akfw68GTglyXH9KGAL3WGhZY0H\nQOeqKcqRpGPPwsLCc3aQd+/ePdi6pzkHcAA4P8mLkgS4ALgfuBO4rF9mO3DzdCVKkmZhmnMA3wBu\nBO4BvgUE+DywE/hoku8ApwHXDlCnJGlgU12jX1W7gcPHI98FzptmvZKk2fNOYElqlAEgSY0yACSp\nUQaAJDXKAJCkRhkAktQoA0CSGmUASFKjDABJapQBIEmNMgAkqVEGgCQ1ygCQpEYZAJLUKANAkhpl\nAEhSowwASWqUASBJjTIAJKlRBoAkNcoAkKRGGQCS1CgDQJIaZQBIUqMMAElqlAEgSY0yACSpUQaA\nJDVqqgBIsjHJDUn2J7k/yXlJTk1yW5IHk9yaZONQxUqShjPtCOD3ga9V1RuANwF/C+wEbq+qs4E7\ngCum3IYkaQYmDoAkLwN+saquA6iqZ6vqGeBSYE+/2B7gnVNXKUka3DQjgDOBHyS5LsndST6f5GRg\nU1UtAlTVE8DPDFGoJGlYG6Z87bnAB6rqm0k+R3f4p1a7gl27dh2aXlhYmKIUSTo2jUYjRqPRTNY9\nTQA8CjxSVd/sn99EFwCLSTZV1WKSzcCTK61gPAA6V01RjiQdexYWFp6zg7x79+7B1j3xIaD+MM8j\nSc7qmy4A7gduAd7Xt20Hbp6mQEnSbEwzAgD4EPClJCcADwPvB44HvprkcuAAcNmU25AkzcBUAVBV\n3wJ+fplZF06zXknS7HknsCQ1ygCQpEYZAJLUKANAkhplAEhSowwASWqUASBJjTIAJKlRBoAkNcoA\nkKRGGQCS1CgDQJIaZQBIUqMMAElqlAEgSY0yACSpUQaAJDXKAJCkRhkAktQoA0CSGmUASFKjDABJ\napQBIEmNMgAkqVEGgCQ1ygCQpEYZAJLUKANAkho1dQAkOS7J3Ulu6Z9vTbI3yYNJrk+yYfoyJUlD\nG2IE8GHggbHnVwOfraqzgaeB3xhgG5KkgU0VAEm2ABcDfzLW/Dbgpn56D/Cr02xDkjQb044APgf8\nLlAASV4BPFVVB/v5jwKvnHIbkqQZmPj4fJJfARaral+ShaXm/jGuVlrHrl27Dk0vLCystJgkNWs0\nGjEajWay7mlO0L4FuCTJxcCLgZcBvwdsTHJcPwrYAjy+0grGA6Bz1RTlSNKxZ2Fh4Tk7yLt37x5s\n3RMfAqqqT1TVa6rqTOBdwB1V9evAncBl/WLbgZunL1OSNLRZ3AewE/hoku8ApwHXzmAbkqQpDXKN\nflX9FfBX/fR3gfOGWK8kaXa8E1iSGmUASFKjDABJapQBIEmNMgAkqVEGgCQ1ygCQpEYZAJLUKANA\nkhplAEhSowwASWqUASBJjTIAJKlRBoAkNcoAkKRGGQCS1CgDQJIaZQBIUqMMAElqlAEgSY0yACSp\nUQaAJDXKAJCkRhkAktQoA0CSGmUASFKjDABJapQBIEmNmjgAkmxJckeSB5Lcl+RDffupSW5L8mCS\nW5NsHK5cSdJQphkBPAt8tKreCPxr4ANJXg/sBG6vqrOBO4Arpi9TkjS0iQOgqp6oqn399N8D+4Et\nwKXAnn6xPcA7py1SkjS8Qc4BJNkKnAPsBTZV1SJ0IQH8zBDbkCQNa+oASPJS4Ebgw/1IoKauSpI0\ncxumeXGSDXRf/l+sqpv75sUkm6pqMclm4MmVXr9r165D0wsLC9OUIknHpNFoxGg0msm6UzX5DnuS\nLwA/qKqPjrVdDfywqq5O8nHg1Kraucxr6/Btn3/+Nu6663eAbauu4aSTdvCZz5zFjh07Jv0xJOkF\nIwlVlSHWNfEIIMlbgP8A3JfkHrpDP58Arga+muRy4ABw2RCFSpKGNXEAVNX/Ao5fYfaFk65XkrQ2\nvBNYkhplAEhSowwASWqUASBJjTIAJKlRBoAkNcoAkKRGGQCS1CgDQJIaZQBIUqMMAK1bmzdvJclR\nPTZv3jrvsqUXjKn+HLQ0S4uL3+do/3uJxcVB/kii1ARHAJLUKANAkhplAEhSowwASWqUASBJjTIA\nJKlRBoAkNcoAkKRGGQCS1CgDQJIaZQBIxyD/jpJWw78FJB2D/DtKWg1HAOuUe3Drn78jvdA5Apix\nzZu39ntjkzh29uCm64f1yb1svdAZADM2yZdE59j6opisH46tPpDWGw8BSWvqpGPqsNEkh8HW+8/U\nkpkFQJJtSf42yXeSfHxW25FeWH5ENxJa/WM9Hzr7p5HdsfMztWQmAZDkOOAPgIuAnwPeneT1s9jW\nkEaj0bxLWMFo3gX8lPXZV6N5F7CM0bwLWMZo3gUs60jvqUlGG8cf/5KpRierfZ+/UC8ImNUI4BeA\nh6rq+1X1Y+ArwKUz2tZg1ueXGqzHD+z67KvRvAtYxmjeBSxjNO8ClnWk99Qko42DB//fUb9mfHSy\n2vf5JLWth1HQrE4Cvwp4ZOz5o3ShsG6sdFXK7t27V3zNcced3L+h1qvu+PLR2LTpDJ544nuzKWcu\njr4PYL3/bif7mdbvdrRezCoAlnsXHfESkBe96AROPvlKNmz4w1Vv6B//8T5OOGHn0dQGrHRVyq7+\nsbyDB7PMa45kLT9QS8eXV+/Yuyzx6PsA1vvvdpKfaZLa1mo7Wi9SNcklikdYaXI+sKuqtvXPdwJV\nVVePLTP8hiWpAVU1SPLOKgCOBx4ELgD+D/AN4N1VtX/wjUmSJjKTQ0BV9ZMkHwRuozvRfK1f/pK0\nvsxkBCBJWv/mcifwvG4SS7IlyR1JHkhyX5IP9e2nJrktyYNJbk2ycew11yR5KMm+JOfMsLbjktyd\n5Jb++dYke/uark+yoW8/MclX+pr+OslrZlTPxiQ3JNmf5P4k5827n5J8JMm3k9yb5Et9X6x5PyW5\nNsliknvH2o66b5Js7z8DDyZ57wxq+nT/+9uX5KYkLx+bd0Vf0/4kbx9rH+yzuVxNY/P+Y5KDSU4b\na5tLP/XtO/qf+74kV421z7yfVqoryZv69+49Sb6R5OfH5g3TV1W1pg+60Pk74AzgBGAf8Po12vZm\n4Jx++qV05yleD1wNfKxv/zhwVT/9DuC/99PnAXtnWNtHgP8K3NI//zPgsn76j4Hf7Kd/C/ijfvrX\ngK/MqJ7/Ary/n94AbJxnPwGvBB4GThzrn+3z6Cfg3wDnAPeOtR1V3wCnAv+779dTlqYHrulC4Lh+\n+irgU/30G4F7+t/r1v7zmKE/m8vV1LdvAf4n8F3gtHXQTwt0h6s39M//Wf/vG9ain56nrluBt4/1\nz5399MVD9dU8RgBzu0msqp6oqn399N8D++nejJcCe/rF9ozVcynwhX75u4CNSTYNXVeSLXS/1D8Z\na34bcNNYTe8cq2mp1hvpTrQPXc/LgF+squsAqurZqnqGOfcTcDzwkn4v/8XA48Avscb9VFVfB546\nrPlo++Yi4Laqeqaqnqb7Ato2ZE1VdXtVHeyf7qV7rwNcQheIz1bV94CH6D6Xg342V+gngM8Bv3tY\n29z6iW5n4aqqerZf5gdjNc28n56nroN0X+bQfaE/1k9fwkB9NY8AWO4msVetdRFJttIl7l5gU1Ut\nQhcSwOn9YofX+hizqXXpA1F9ba8Anhr78I730aGaquonwNPjw+iBnAn8IMl16Q5LfT7Jycyxn6rq\nceCzwIF+/c8AdwNPz7Gfxp2+yr5ZqnGt3ltLLge+tkJNS9ue+Wczyb8DHqmq+w6bNc9+Ogv4t/2h\nxDuT/KsValqzfup9BPhPSQ4AnwauWKGuiftqHgEw0U1igxaQvJRur/DD/Uhgpe3PvNYkvwIs9iOT\npe1lmW3X2LznrGLomuiGvOcCf1hV5wL/AOx8nu2sRT+dQreXdQbd4aCX0A2LV9ruWvTTaqxUx5p9\nDpJ8EvhxVV2/Qk1L255pTUleDHwSuHK52cs8X6t+2gCcUlXnAx8DblihpqVtr9Xv7rfovqNeQxcG\nf7pCXRP31TwC4FFg/ITcFrqh/JroDx/cCHyxqm7umxeXDlkk2Qw8OVbrq2dc61uAS5I8DFxPd+jn\n9+iGdUu/n/HtHqop3f0WL6+q5YbZ03iUbi/tm/3zm+gCYZ79dCHwcFX9sN+j/3PgzcApc+yncUfb\nN2vyOUiyne7w4nvGmudV08/SHUv/VpLv9uu/O8npc6wJur3m/wZQVX8D/KQfha+07bX6DtteVX/R\n13UjsHQSeLC+mkcA/A3wuiRnJDkReBdwyxpu/0+BB6rq98fabgHe10+/D7h5rP29cOju5qeXhvlD\nqapPVNVrqupMur64o6p+HbgTuKxfbPthNW3vpy8D7hiynr6mReCRJGf1TRcA9zPHfqI79HN+khcl\nyVhN8+qnw0dpR9s3twK/nO5qq1OBX+7bBqspyTa6PdpLqupHh9X6rnRXSr0WeB3dzZqz+Gweqqmq\nvl1Vm6vqzKp6Ld0X1r+sqieZYz8Bf0F/jqh/z59YVf+3r+nX1qiflqvrsSRv7eu6gO4cBAzZV9Oc\nuZ7ijPc2uitwHgJ2ruF23wL8hO6s/T10x5C3AacBt/c1/SXdcHDpNX9Ad8b/W8C5M67vrfzTVUCv\nBe4CvkN3pcsJfftJwFf7vtsLbJ1RLW+ie6Pvo9s72jjvfqI7dLAfuJfuROsJ8+gn4Mt0e1Y/ogum\n99NdgXFUfUMXFA/1tb93BjU9BHy/f5/fTX9VVL/8FX1N++mvNOnbB/tsLlfTYfMfpr8KaM79tAH4\nInAf8E3grWvZT89T15v7eu4B/pouLAftK28Ek6RG+V9CSlKjDABJapQBIEmNMgAkqVEGgCQ1ygCQ\npEYZAJLUKANAkhr1/wG7eGpHl5M12QAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c4741ad90>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "c = 3\n",
    "incorrect = filter(lambda (x,y,q,r,rv): not rv and y == c, train)\n",
    "print len(incorrect)\n",
    "dists = map(lambda (x,y,q,r,rv): distance(r, c), incorrect)\n",
    "dists.sort()\n",
    "plt.hist(dists, bins=24);"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[30.0,\n",
       " 30.0,\n",
       " 30.0,\n",
       " 30.0,\n",
       " 30.0,\n",
       " 30.0,\n",
       " 30.0,\n",
       " 30.0,\n",
       " 30.0,\n",
       " 30.0,\n",
       " 30.0,\n",
       " 30.0,\n",
       " 30.0,\n",
       " 30.0,\n",
       " 30.0,\n",
       " 30.0,\n",
       " 30.0,\n",
       " 30.0,\n",
       " 30.0,\n",
       " 35.777087639996637,\n",
       " 185.77674773770801,\n",
       " 282.85155117128136,\n",
       " 316.42534664593478,\n",
       " 329.13371143047624,\n",
       " 338.00147928670373,\n",
       " 364.9438312946254,\n",
       " 403.27162062312294,\n",
       " 416.20307543313515,\n",
       " 430.83639586274512,\n",
       " 449.44410108488466]"
      ]
     },
     "execution_count": 69,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dists[100:130]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "11449\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[1017, 757, 1939, 827, 597, 1449, 1495, 796, 1478, 1094]"
      ]
     },
     "execution_count": 70,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYAAAAEACAYAAAC6d6FnAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nDetxAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c462be350>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "c = 3\n",
    "train = filter(lambda (x,y,q,r,rv): y != c or (y == c and (rv or (not rv and (distance(r, c) < 1.)))), train)\n",
    "print len(train)\n",
    "TRAIN_CLASSES = map(int, plt.hist(zip(*train)[1])[0])\n",
    "TRAIN_CLASSES"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.6921093347086127"
      ]
     },
     "execution_count": 71,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "M = max(TRAIN_CLASSES)\n",
    "m = min(TRAIN_CLASSES)\n",
    "(M - m) / float(M)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "11449"
      ]
     },
     "execution_count": 72,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(train)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## NEW: filter using EXIF"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "a28625b1b71f42f598b13de2b15b6554",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "A Jupyter Widget"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "train_with_exif = []\n",
    "for x,y,q,r,rv in progress_iter(train, True):\n",
    "    tags = exifread.process_file(open(x, 'rb'))\n",
    "    train_with_exif.append((x,y,q,r,rv,tags))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "train = train_with_exif"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'Lens Make'"
      ]
     },
     "execution_count": 75,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "'Image Make'\n",
    "'Image Model'\n",
    "'Lens Make'\n",
    "# distance((int(str(t['EXIF ExifImageWidth'])), int(str(t['EXIF ExifImageLength']))), c) > 0.1\n",
    "# 'Image Orientation'"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### class 0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1017\n"
     ]
    }
   ],
   "source": [
    "c = 0\n",
    "c_train = filter(lambda (x,y,q,r,rv,tags): y == c, train)\n",
    "print len(c_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 77,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'Horizontal (normal)'"
      ]
     },
     "execution_count": 77,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "str(c_train[0][5]['Image Orientation'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "994\n"
     ]
    }
   ],
   "source": [
    "c1 = lambda t: 'EXIF ExifImageWidth' in t\n",
    "c2 = lambda t: 'EXIF ExifImageLength' in t\n",
    "c3 = lambda t: distance((int(str(t['EXIF ExifImageWidth'])), int(str(t['EXIF ExifImageLength']))), c) < 0.1\n",
    "cond = [\n",
    "    lambda t: str(t['Image Make']) == 'HTC',\n",
    "    lambda t: str(t['Image Model']) == 'HTC One',\n",
    "    lambda t: c1(t) and c2(t) and c3(t),\n",
    "    lambda t: 'Image Orientation' in t and str(t['Image Orientation']) == 'Horizontal (normal)'\n",
    "]\n",
    "all_cond = lambda (x,y,q,r,rv,t): all(map(lambda c: c(t), cond))\n",
    "print len(filter(all_cond, c_train))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 79,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "11449\n",
      "11426\n"
     ]
    }
   ],
   "source": [
    "print len(train)\n",
    "train = filter(lambda (x,y,q,r,rv,tags): y != c or all_cond((x,y,q,r,rv,tags)), train)\n",
    "print len(train)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### class 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 80,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "757\n"
     ]
    }
   ],
   "source": [
    "c = 1\n",
    "c_train = filter(lambda (x,y,q,r,rv,tags): y == c, train)\n",
    "print len(c_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 81,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "662\n"
     ]
    }
   ],
   "source": [
    "c1 = lambda t: 'EXIF ExifImageWidth' in t\n",
    "c2 = lambda t: 'EXIF ExifImageLength' in t\n",
    "c3 = lambda t: distance((int(str(t['EXIF ExifImageWidth'])), int(str(t['EXIF ExifImageLength']))), c) < 0.1\n",
    "cond = [\n",
    "    lambda t: 'Image Make' in t and str(t['Image Make']) == 'LGE',\n",
    "    lambda t: 'Image Model' in t and str(t['Image Model']) == 'Nexus 5X',\n",
    "    lambda t: c1(t) and c2(t) and c3(t),\n",
    "#     lambda t: 'Image Orientation' in t and str(t['Image Orientation']) == 'Horizontal (normal)'\n",
    "]\n",
    "all_cond = lambda (x,y,q,r,rv,t): all(map(lambda c: c(t), cond))\n",
    "print len(filter(all_cond, c_train))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 82,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "11426\n",
      "11331\n"
     ]
    }
   ],
   "source": [
    "print len(train)\n",
    "train = filter(lambda (x,y,q,r,rv,tags): y != c or all_cond((x,y,q,r,rv,tags)), train)\n",
    "print len(train)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### class 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1939\n"
     ]
    }
   ],
   "source": [
    "c = 2\n",
    "c_train = filter(lambda (x,y,q,r,rv,tags): y == c, train)\n",
    "print len(c_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 84,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['XT1080']\n"
     ]
    }
   ],
   "source": [
    "print list(set([str(t['Image Model']) for x,y,q,r,rv,t in c_train if 'data/train' in x]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 85,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "801\n"
     ]
    }
   ],
   "source": [
    "c1 = lambda t: 'EXIF ExifImageWidth' in t\n",
    "c2 = lambda t: 'EXIF ExifImageLength' in t\n",
    "c3 = lambda t: distance((float(str(t['EXIF ExifImageWidth'])), float(str(t['EXIF ExifImageLength']))), c) < 0.1\n",
    "cond = [\n",
    "    lambda t: 'Image Make' in t and str(t['Image Make']) == 'Motorola',\n",
    "    lambda t: 'Image Model' in t and str(t['Image Model']) == 'XT1080',\n",
    "    lambda t: c1(t) and c2(t) and c3(t),\n",
    "#     lambda t: 'Image Orientation' in t and str(t['Image Orientation']) == 'Horizontal (normal)'\n",
    "]\n",
    "all_cond = lambda (x,y,q,r,rv,t): all(map(lambda c: c(t), cond))\n",
    "print len(filter(all_cond, c_train))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 86,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "11331\n",
      "10193\n"
     ]
    }
   ],
   "source": [
    "print len(train)\n",
    "train = filter(lambda (x,y,q,r,rv,tags): y != c or all_cond((x,y,q,r,rv,tags)), train)\n",
    "print len(train)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### class 3"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 87,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "827\n"
     ]
    }
   ],
   "source": [
    "c = 3\n",
    "c_train = filter(lambda (x,y,q,r,rv,tags): y == c, train)\n",
    "print len(c_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 88,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['motorola']\n"
     ]
    }
   ],
   "source": [
    "print list(set([str(t['Image Make']) for x,y,q,r,rv,t in c_train if 'data/train' in x]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 89,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['Nexus 6']\n"
     ]
    }
   ],
   "source": [
    "print list(set([str(t['Image Model']) for x,y,q,r,rv,t in c_train if 'data/train' in x]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 90,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "793\n"
     ]
    }
   ],
   "source": [
    "c1 = lambda t: 'EXIF ExifImageWidth' in t\n",
    "c2 = lambda t: 'EXIF ExifImageLength' in t\n",
    "c3 = lambda t: distance((float(str(t['EXIF ExifImageWidth'])), float(str(t['EXIF ExifImageLength']))), c) < 0.1\n",
    "cond = [\n",
    "    lambda t: 'Image Make' in t and str(t['Image Make']) == 'motorola',\n",
    "    lambda t: 'Image Model' in t and str(t['Image Model']) == 'Nexus 6',\n",
    "    lambda t: c1(t) and c2(t) and c3(t),\n",
    "#     lambda t: 'Image Orientation' in t and str(t['Image Orientation']) == 'Horizontal (normal)'\n",
    "]\n",
    "all_cond = lambda (x,y,q,r,rv,t): all(map(lambda c: c(t), cond))\n",
    "print len(filter(all_cond, c_train))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 91,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "10193\n",
      "10159\n"
     ]
    }
   ],
   "source": [
    "print len(train)\n",
    "train = filter(lambda (x,y,q,r,rv,tags): y != c or all_cond((x,y,q,r,rv,tags)), train)\n",
    "print len(train)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### class 4"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 92,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "597\n"
     ]
    }
   ],
   "source": [
    "c = 4\n",
    "c_train = filter(lambda (x,y,q,r,rv,tags): y == c, train)\n",
    "print len(c_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 93,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['Motorola']\n"
     ]
    }
   ],
   "source": [
    "print list(set([str(t['Image Make']) for x,y,q,r,rv,t in c_train if 'data/train' in x]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 94,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['XT1096']\n"
     ]
    }
   ],
   "source": [
    "print list(set([str(t['Image Model']) for x,y,q,r,rv,t in c_train if 'data/train' in x]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "595\n"
     ]
    }
   ],
   "source": [
    "c1 = lambda t: 'EXIF ExifImageWidth' in t\n",
    "c2 = lambda t: 'EXIF ExifImageLength' in t\n",
    "c3 = lambda t: distance((float(str(t['EXIF ExifImageWidth'])), float(str(t['EXIF ExifImageLength']))), c) < 0.1\n",
    "cond = [\n",
    "    lambda t: 'Image Make' in t and str(t['Image Make']) == 'Motorola',\n",
    "    lambda t: 'Image Model' in t and str(t['Image Model']) == 'XT1096',\n",
    "    lambda t: c1(t) and c2(t) and c3(t),\n",
    "#     lambda t: 'Image Orientation' in t and str(t['Image Orientation']) == 'Horizontal (normal)'\n",
    "]\n",
    "all_cond = lambda (x,y,q,r,rv,t): all(map(lambda c: c(t), cond))\n",
    "print len(filter(all_cond, c_train))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 96,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "10159\n",
      "10157\n"
     ]
    }
   ],
   "source": [
    "print len(train)\n",
    "train = filter(lambda (x,y,q,r,rv,tags): y != c or all_cond((x,y,q,r,rv,tags)), train)\n",
    "print len(train)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### class 5"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 97,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1449\n"
     ]
    }
   ],
   "source": [
    "c = 5\n",
    "c_train = filter(lambda (x,y,q,r,rv,tags): y == c, train)\n",
    "print len(c_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 98,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['samsung', 'SAMSUNG']\n",
      "['SAMSUNG-SM-N900A']\n"
     ]
    }
   ],
   "source": [
    "print list(set([str(t['Image Make']) for x,y,q,r,rv,t in c_train if 'data/train' in x]))\n",
    "print list(set([str(t['Image Model']) for x,y,q,r,rv,t in c_train if 'data/train' in x]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 99,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['SAMSUNG-SM-N900A', 'SM-N9005']\n"
     ]
    }
   ],
   "source": [
    "print list(set([str(t['Image Model']) for x,y,q,r,rv,t in c_train]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 100,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['[4601, 0]', '2322', '[3264, 0]', '4128', '3210', '3399', '3142', '3264', '1836']\n"
     ]
    }
   ],
   "source": [
    "print list(set([str(t['EXIF ExifImageWidth']) for x,y,q,r,rv,t in c_train]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 101,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def f(v):\n",
    "    return float(str(v).replace(', 0]', '').replace('[', ''))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 102,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "903\n"
     ]
    }
   ],
   "source": [
    "c1 = lambda t: 'EXIF ExifImageWidth' in t\n",
    "c2 = lambda t: 'EXIF ExifImageLength' in t\n",
    "c3 = lambda t: distance((f(t['EXIF ExifImageWidth']), f(t['EXIF ExifImageLength'])), c) < 0.1\n",
    "cond = [\n",
    "    lambda t: 'Image Make' in t and str(t['Image Make']).lower() == 'samsung',\n",
    "    lambda t: 'Image Model' in t and str(t['Image Model']) in ['SAMSUNG-SM-N900A', 'SM-N9005'],\n",
    "    lambda t: c1(t) and c2(t) and c3(t),\n",
    "#     lambda t: 'Image Orientation' in t and str(t['Image Orientation']) == 'Horizontal (normal)'\n",
    "]\n",
    "all_cond = lambda (x,y,q,r,rv,t): all(map(lambda c: c(t), cond))\n",
    "print len(filter(all_cond, c_train))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 103,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "10157\n",
      "9611\n"
     ]
    }
   ],
   "source": [
    "print len(train)\n",
    "train = filter(lambda (x,y,q,r,rv,tags): y != c or all_cond((x,y,q,r,rv,tags)), train)\n",
    "print len(train)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### class 6"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 104,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1495\n"
     ]
    }
   ],
   "source": [
    "c = 6\n",
    "c_train = filter(lambda (x,y,q,r,rv,tags): y == c, train)\n",
    "print len(c_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 105,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['samsung']\n",
      "['SCH-I545']\n"
     ]
    }
   ],
   "source": [
    "print list(set([str(t['Image Make']) for x,y,q,r,rv,t in c_train if 'data/train' in x]))\n",
    "print list(set([str(t['Image Model']) for x,y,q,r,rv,t in c_train if 'data/train' in x]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 106,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['GT-I9505', 'SCH-I545', 'None', 'GT-I9500']\n"
     ]
    }
   ],
   "source": [
    "print list(set([str(t.get('Image Model', None)) for x,y,q,r,rv,t in c_train]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 107,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1493\n"
     ]
    }
   ],
   "source": [
    "c1 = lambda t: 'EXIF ExifImageWidth' in t\n",
    "c2 = lambda t: 'EXIF ExifImageLength' in t\n",
    "c3 = lambda t: distance((f(t['EXIF ExifImageWidth']), f(t['EXIF ExifImageLength'])), c) < 0.1\n",
    "cond = [\n",
    "    lambda t: 'Image Make' in t and str(t['Image Make']).lower() == 'samsung',\n",
    "    lambda t: 'Image Model' in t and str(t['Image Model']) in ['GT-I9505', 'SCH-I545', 'GT-I9500'],\n",
    "    lambda t: c1(t) and c2(t) and c3(t),\n",
    "#     lambda t: 'Image Orientation' in t and str(t['Image Orientation']) == 'Horizontal (normal)'\n",
    "]\n",
    "all_cond = lambda (x,y,q,r,rv,t): all(map(lambda c: c(t), cond))\n",
    "print len(filter(all_cond, c_train))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 108,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "9611\n",
      "9609\n"
     ]
    }
   ],
   "source": [
    "print len(train)\n",
    "train = filter(lambda (x,y,q,r,rv,tags): y != c or all_cond((x,y,q,r,rv,tags)), train)\n",
    "print len(train)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### class 7"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 109,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "796\n"
     ]
    }
   ],
   "source": [
    "c = 7\n",
    "c_train = filter(lambda (x,y,q,r,rv,tags): y == c, train)\n",
    "print len(c_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 110,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['SONY']\n",
      "['NEX-7']\n"
     ]
    }
   ],
   "source": [
    "print list(set([str(t['Image Make']) for x,y,q,r,rv,t in c_train if 'data/train' in x]))\n",
    "print list(set([str(t['Image Model']) for x,y,q,r,rv,t in c_train if 'data/train' in x]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 111,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "757\n"
     ]
    }
   ],
   "source": [
    "c1 = lambda t: 'EXIF ExifImageWidth' in t\n",
    "c2 = lambda t: 'EXIF ExifImageLength' in t\n",
    "c3 = lambda t: distance((f(t['EXIF ExifImageWidth']), f(t['EXIF ExifImageLength'])), c) < 0.1\n",
    "cond = [\n",
    "    lambda t: 'Image Make' in t and str(t['Image Make']) == 'SONY',\n",
    "    lambda t: 'Image Model' in t and str(t['Image Model']) == 'NEX-7',\n",
    "    lambda t: c1(t) and c2(t) and c3(t),\n",
    "#     lambda t: 'Image Orientation' in t and str(t['Image Orientation']) == 'Horizontal (normal)'\n",
    "]\n",
    "all_cond = lambda (x,y,q,r,rv,t): all(map(lambda c: c(t), cond))\n",
    "print len(filter(all_cond, c_train))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 112,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "9609\n",
      "9570\n"
     ]
    }
   ],
   "source": [
    "print len(train)\n",
    "train = filter(lambda (x,y,q,r,rv,tags): y != c or all_cond((x,y,q,r,rv,tags)), train)\n",
    "print len(train)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### class 8"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 113,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1478\n"
     ]
    }
   ],
   "source": [
    "c = 8\n",
    "c_train = filter(lambda (x,y,q,r,rv,tags): y == c, train)\n",
    "print len(c_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 114,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['Apple']\n",
      "['iPhone 4S']\n",
      "['Apple']\n",
      "['iPhone 4S back camera 4.28mm f/2.4']\n"
     ]
    }
   ],
   "source": [
    "print list(set([str(t['Image Make']) for x,y,q,r,rv,t in c_train if 'data/train' in x]))\n",
    "print list(set([str(t['Image Model']) for x,y,q,r,rv,t in c_train if 'data/train' in x]))\n",
    "print list(set([str(t['EXIF LensMake']) for x,y,q,r,rv,t in c_train if 'data/train' in x]))\n",
    "print list(set([str(t['EXIF LensModel']) for x,y,q,r,rv,t in c_train if 'data/train' in x]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 115,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "888\n"
     ]
    }
   ],
   "source": [
    "c1 = lambda t: 'EXIF ExifImageWidth' in t\n",
    "c2 = lambda t: 'EXIF ExifImageLength' in t\n",
    "c3 = lambda t: distance((f(t['EXIF ExifImageWidth']), f(t['EXIF ExifImageLength'])), c) < 0.1\n",
    "cond = [\n",
    "    lambda t: 'Image Make' in t and str(t['Image Make']) == 'Apple',\n",
    "    lambda t: 'Image Model' in t and str(t['Image Model']) == 'iPhone 4S',\n",
    "    lambda t: 'EXIF LensMake' in t and str(t['EXIF LensMake']) == 'Apple',\n",
    "    lambda t: 'EXIF LensModel' in t and str(t['EXIF LensModel']) == 'iPhone 4S back camera 4.28mm f/2.4',\n",
    "    lambda t: c1(t) and c2(t) and c3(t),\n",
    "#     lambda t: 'Image Orientation' in t and str(t['Image Orientation']) == 'Horizontal (normal)'\n",
    "]\n",
    "all_cond = lambda (x,y,q,r,rv,t): all(map(lambda c: c(t), cond))\n",
    "print len(filter(all_cond, c_train))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 116,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "9570\n",
      "8980\n"
     ]
    }
   ],
   "source": [
    "print len(train)\n",
    "train = filter(lambda (x,y,q,r,rv,tags): y != c or all_cond((x,y,q,r,rv,tags)), train)\n",
    "print len(train)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### class 9"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 117,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1094\n"
     ]
    }
   ],
   "source": [
    "c = 9\n",
    "c_train = filter(lambda (x,y,q,r,rv,tags): y == c, train)\n",
    "print len(c_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 118,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['Apple']\n",
      "['iPhone 6']\n",
      "['Apple']\n",
      "['iPhone 6 back camera 4.15mm f/2.2']\n"
     ]
    }
   ],
   "source": [
    "print list(set([str(t['Image Make']) for x,y,q,r,rv,t in c_train if 'data/train' in x]))\n",
    "print list(set([str(t['Image Model']) for x,y,q,r,rv,t in c_train if 'data/train' in x]))\n",
    "print list(set([str(t['EXIF LensMake']) for x,y,q,r,rv,t in c_train if 'data/train' in x]))\n",
    "print list(set([str(t['EXIF LensModel']) for x,y,q,r,rv,t in c_train if 'data/train' in x]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 119,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1081\n"
     ]
    }
   ],
   "source": [
    "c1 = lambda t: 'EXIF ExifImageWidth' in t\n",
    "c2 = lambda t: 'EXIF ExifImageLength' in t\n",
    "c3 = lambda t: distance((f(t['EXIF ExifImageWidth']), f(t['EXIF ExifImageLength'])), c) < 0.1\n",
    "cond = [\n",
    "    lambda t: 'Image Make' in t and str(t['Image Make']) == 'Apple',\n",
    "    lambda t: 'Image Model' in t and str(t['Image Model']) == 'iPhone 6',\n",
    "    lambda t: 'EXIF LensMake' in t and str(t['EXIF LensMake']) == 'Apple',\n",
    "    lambda t: 'EXIF LensModel' in t and str(t['EXIF LensModel']) == 'iPhone 6 back camera 4.15mm f/2.2',\n",
    "    lambda t: c1(t) and c2(t) and c3(t),\n",
    "#     lambda t: 'Image Orientation' in t and str(t['Image Orientation']) == 'Horizontal (normal)'\n",
    "]\n",
    "all_cond = lambda (x,y,q,r,rv,t): all(map(lambda c: c(t), cond))\n",
    "print len(filter(all_cond, c_train))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 120,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "8980\n",
      "8967\n"
     ]
    }
   ],
   "source": [
    "print len(train)\n",
    "train = filter(lambda (x,y,q,r,rv,tags): y != c or all_cond((x,y,q,r,rv,tags)), train)\n",
    "print len(train)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### TODO: finally fill validation set up from filtered train (q $\\geq$ 95)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 121,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[48, 48, 48, 48, 48, 48, 48, 48, 48, 48]"
      ]
     },
     "execution_count": 121,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXQAAAEACAYAAACj0I2EAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAADWVJREFUeJzt3X2sZHV9x/H3Z7mtiihSi0vpRohpRG1ikT8ANU3WSn1M\nC8bQ4EOKD/3PFlIS4sI/kDRpIMEYkzZtjJSs1gcUH6CJKZSsQ2IbfIhQtoWuTZEipXspUTTEtHLd\nb/+4s/G6ubszc++dOZfvfb+SCWfOzpnfNxN4c+bceyBVhSTp2W/X0ANIkraGQZekJgy6JDVh0CWp\nCYMuSU0YdElqYmmaFyV5BPgRcAR4pqrOT3IacCtwFvAI8AdV9aM5zSlJmmDaM/QjwN6qek1VnT/e\ntw+4u6rOAQ4A18xjQEnSdKYNetZ57cXA/vH2fuCSrRpKkjS7aYNewJ1JvpXkj8b7dlfVMkBVHQZO\nn8eAkqTpTHUNHXhdVR1OcjpwV5JDrEZekrRNTBX08Rk4VfU/Sb4CnA8sJ9ldVctJzgCeWO/YJIZf\nkjagqjLL6ydecklycpJTxtvPB94EHATuAN43ftnlwO0nGMpHFdddd93gM2yXh5+Fn4WfxYkfGzHN\nGfpu4MvjM+0l4NNVdVeSbwOfT/IB4FHg0g1NIEnaEhODXlXfA85dZ/8PgIvmMZQkaXbeKbpAe/fu\nHXqEbcPP4uf8LH7Oz2JzstFrNVMvkNS815CkbpJQW/1DUUnSs4NBl6Qmpr2xaHOLLD1nEcv8gltv\n/QzvfOc7F76uJA1lIdfQ4X/nusaxlpb28dzn/i1PP/3kQtcF2LXrZI4c+YnrNl13yLVddzF27z6L\nw4cfWfi6x9rINfSFnKHDos/Ql8YxX/wPY48cies2XnfItV13MZaXZ2rotuI1dElqwqBLUhMGXZKa\nMOiS1IRBl6QmDLokNWHQJakJgy5JTRh0SWrCoEtSEwZdkpow6JLUhEGXpCYMuiQ1YdAlqQmDLklN\nGHRJasKgS1ITBl2SmjDoktSEQZekJgy6JDVh0CWpCYMuSU0YdElqwqBLUhMGXZKaMOiS1IRBl6Qm\npg56kl1JvpPkjvHzs5Pcm+RQks8mWZrfmJKkSWY5Q78SeHDN8xuBj1TVOcBTwAe3cjBJ0mymCnqS\nPcDbgE+s2f07wBfH2/uBd2ztaJKkWUx7hv5R4GqgAJK8GPhhVR0Z//ljwJlbP54kaVoTr3sneTuw\nXFX3J9l7dPf4sVYd/12uX7O9d/yQJB01Go0YjUabeo9UnaDDQJI/B94LrADPA14AfAV4E3BGVR1J\nciFwXVW9dZ3j64Stn4OlpatZWbmJRa+7Kq7bet0h13bdRa07qYsLmSKhqo49cT6hiZdcquraqnpp\nVb0MuAw4UFXvBb4GXDp+2eXA7bMOLEnaOpv5PfR9wFVJvgv8CnDz1owkSdqImX53vKruAe4Zb38P\nuGAeQ0mSZuedopLUhEGXpCYMuiQ1YdAlqQmDLklNGHRJasKgS1ITBl2SmjDoktSEQZekJgy6JDVh\n0CWpCYMuSU0YdElqwqBLUhMGXZKaMOiS1IRBl6QmDLokNWHQJakJgy5JTRh0SWrCoEtSEwZdkpow\n6JLUhEGXpCYMuiQ1YdAlqQmDLklNGHRJasKgS1ITBl2SmjDoktSEQZekJgy6JDUxMehJnpPkG0nu\nS3IwyXXj/WcnuTfJoSSfTbI0/3ElScczMehV9X/AG6rqNcC5wFuTXADcCHykqs4BngI+ONdJJUkn\nNNUll6r6yXjzOcASUMAbgC+O9+8H3rHl00mSpjZV0JPsSnIfcBj4B+A/gKeq6sj4JY8BZ85nREnS\nNKY9Qz8yvuSyBzgfeOV6L9vKwSRJs5npB5lV9eMk9wAXAi9Ksmt8lr4HePz4R16/Znvv+CFJOmo0\nGjEajTb1Hqk68Yl1kl8FnqmqHyV5HnAncANwOfClqro1yV8B/1xVf73O8bXok/elpatZWbmJYb40\nxHVbrzvk2q67qHUndXEhUyRUVWY5Zpoz9F8D9ifZxeolmlur6qtJHgI+l+TPgPuAm2eeWJK0ZSYG\nvaoOAuets/97wAXzGEqSNDvvFJWkJgy6JDVh0CWpCYMuSU0YdElqwqBLUhMGXZKaMOiS1IRBl6Qm\nDLokNWHQJakJgy5JTRh0SWrCoEtSEwZdkpow6JLUhEGXpCYMuiQ1YdAlqQmDLklNGHRJasKgS1IT\nBl2SmjDoktSEQZekJgy6JDVh0CWpCYMuSU0YdElqwqBLUhMGXZKaMOiS1IRBl6QmDLokNWHQJakJ\ngy5JTUwMepI9SQ4keTDJwSRXjPefluSuJIeS3Jnk1PmPK0k6nmnO0FeAq6rqVcBrgQ8leQWwD7i7\nqs4BDgDXzG9MSdIkE4NeVYer6v7x9tPAQ8Ae4GJg//hl+4FL5jWkJGmyma6hJzkbOBe4F9hdVcuw\nGn3g9K0eTpI0vamDnuQU4DbgyvGZes1tKknSzJameVGSJVZj/qmqun28eznJ7qpaTnIG8MTx3+H6\nNdt7xw9J0lGj0YjRaLSp90jV5BPtJJ8Enqyqq9bsuxH4QVXdmOTDwGlVtW+dY2vRJ/NLS1ezsnIT\nw3yJiOu2XnfItV13UetO08W5T5FQVZnlmIln6EleD7wHOJjkPlY/4WuBG4HPJ/kA8Chw6ewjS5K2\nysSgV9U/Aicd548v2tpxJEkb5Z2iktSEQZekJgy6JDVh0CWpCYMuSU0YdElqwqBLUhMGXZKaMOiS\n1IRBl6QmDLokNWHQJakJgy5JTRh0SWrCoEtSEwZdkpow6JLUhEGXpCYMuiQ1YdAlqQmDLklNGHRJ\nasKgS1ITBl2SmjDoktSEQZekJgy6JDVh0CWpCYMuSU0YdElqwqBLUhMGXZKaMOiS1IRBl6QmDLok\nNWHQJamJiUFPcnOS5SQPrNl3WpK7khxKcmeSU+c7piRpkmnO0G8B3nzMvn3A3VV1DnAAuGarB5Mk\nzWZi0Kvq68APj9l9MbB/vL0fuGSL55IkzWij19BfUlXLAFV1GDh960aSJG3E0mKWuX7N9t7xQ5J0\n1Gg0YjQabeo9UlWTX5ScBfxdVb16/PwhYG9VLSc5A/haVb3yOMcWTF5jKy0tXc3Kyk0set1Vcd3W\n6w65tusuat1pujj3KRKqKrMcM+0ll4wfR90BvG+8fTlw+yyLSpK23jS/tvgZ4J+Alyd5NMn7gRuA\n301yCLho/FySNKCJ19Cr6t3H+aOLtngWSdImeKeoJDVh0CWpCYMuSU0YdElqwqBLUhMGXZKaMOiS\n1IRBl6QmDLokNWHQJakJgy5JTRh0SWrCoEtSEwZdkpow6JLUhEGXpCYMuiQ1YdAlqQmDLklNGHRJ\nasKgS1ITBl2SmjDoktSEQZekJgy6JDVh0CWpCYMuSU0YdElqwqBLUhMGXZKaMOiS1IRBl6QmDLok\nNWHQJakJgy5JTWwq6EnekuTfknw3yYe3aihJ0uw2HPQku4C/AN4M/CbwriSv2KrBehoNPcA2Mhp6\ngG1kNPQA28ho6AGe1TZzhn4+8O9V9Z9V9QzwOeDirRmrq9HQA2wjo6EH2EZGQw+wjYyGHuBZbTNB\n/3Xg+2uePzbeJ0kawNImjs06+2q9F77whb+3iWVm99OfPsjKykKXlKTBpWrdBk8+MLkQuL6q3jJ+\nvg+oqrrxmNdtbAFJ2uGqar0T5+PaTNBPAg4BbwT+G/gm8K6qemhDbyhJ2pQNX3Kpqp8l+WPgLlav\nxd9szCVpOBs+Q5ckbS9zu1PUm45WJdmT5ECSB5McTHLF0DMNLcmuJN9JcsfQswwpyalJvpDkoST/\nmuSCoWcaSpI/TfIvSR5I8ukkvzz0TIuS5OYky0keWLPvtCR3JTmU5M4kp07zXnMJujcd/YIV4Kqq\nehXwWuBDO/izOOpK4MGhh9gGPgZ8tapeCfwWsCMvWSY5E/gT4LyqejWrl4IvG3aqhbqF1VautQ+4\nu6rOAQ4A10zzRvM6Q/emo7GqOlxV94+3n2b1H9od+/v6SfYAbwM+MfQsQ0ryAuC3q+oWgKpaqaof\nDzzWkE4Cnp9kCTgZeHzgeRamqr4O/PCY3RcD+8fb+4FLpnmveQXdm47WkeRs4FzgG8NOMqiPAldz\nnHsWdpCXAU8muWV8+enjSZ439FBDqKrHgY8AjwL/BTxVVXcPO9XgXlJVy7B6UgicPs1B8wr61Dcd\n7RRJTgFuA64cn6nvOEneDiyPv7GE9f8+2SmWgPOAv6yq84CfsPo1e8dJ8iJWz0jPAs4ETkny7mGn\nenaaV9AfA1665vkedtBXqGONv0beBnyqqm4fep4BvR74/SQPA58F3pDkkwPPNJTHgO9X1bfHz29j\nNfA70UXAw1X1g6r6GfAl4HUDzzS05SS7AZKcATwxzUHzCvq3gN9Ictb4p9WXATv5Nxr+Bniwqj42\n9CBDqqprq+qlVfUyVv+eOFBVfzj0XEMYf53+fpKXj3e9kZ37g+JHgQuTPDdJWP0sdtoPiI/9xnoH\n8L7x9uXAVCeCm/lvuRyXNx39XJLXA+8BDia5j9VLT9dW1d8PO5m2gSuATyf5JeBh4P0DzzOIqvpm\nktuA+4Bnxn/9+LBTLU6SzwB7gRcneRS4DrgB+EKSD7D6L7xLp3ovbyySpB78X9BJUhMGXZKaMOiS\n1IRBl6QmDLokNWHQJakJgy5JTRh0SWri/wEDiUDoRWBSCQAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c29440cd0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "RNG(seed=777).shuffle(train)\n",
    "new_train = []\n",
    "for x,y,q,r,rv,_ in train:\n",
    "    VAL_CLASSES = np.bincount(zip(*val)[1])\n",
    "    if q >= 95 and VAL_CLASSES[y] < N_VAL_PER_CLASS:\n",
    "        val.append((x, y))\n",
    "    else:\n",
    "        new_train.append((x,y,q,r,rv))\n",
    "\n",
    "train = new_train\n",
    "        \n",
    "VAL_CLASSES = map(int, plt.hist(zip(*val)[1])[0])\n",
    "VAL_CLASSES"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 122,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.692109334709\n",
      "8847\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[991, 651, 767, 773, 595, 873, 1490, 751, 888, 1068]"
      ]
     },
     "execution_count": 122,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYAAAAEACAYAAAC6d6FnAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nlAEgSY0yACSpUQaAJDXKAJCkRv1/5eNZF/MsMAEAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8c29473d50>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "M = max(TRAIN_CLASSES)\n",
    "m = min(TRAIN_CLASSES)\n",
    "print (M - m) / float(M)\n",
    "print len(train)\n",
    "\n",
    "TRAIN_CLASSES = map(int, plt.hist(zip(*train)[1])[0])\n",
    "TRAIN_CLASSES"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## now need to crop and to save images to disk, but now cleverer:\n",
    "### $\\approx$ undersampling, save separate blocks (of $\\approx$ same size) for each class"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 446,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[21, 14, 16, 16, 12, 18, 31, 16, 18, 22]"
      ]
     },
     "execution_count": 446,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "BLOCK_SIZE = 48\n",
    "N_BLOCKS = [int(np.round(t/float(BLOCK_SIZE))) for t in TRAIN_CLASSES]\n",
    "N_BLOCKS"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "crop_size = 1024\n",
    "for c in xrange(10):\n",
    "    if c == 3:\n",
    "        continue\n",
    "    class_train = filter(lambda (x,y,q,r,rv): y == c, train)\n",
    "    RNG(seed=7331 + c).shuffle(class_train)\n",
    "    class_blocks = []\n",
    "    for _ in xrange(N_BLOCKS[c]):\n",
    "        class_blocks.append([])\n",
    "    pos = 0\n",
    "    for x,_,_,_,_ in progress_iter(class_train, True):\n",
    "        x = Image.open(x)\n",
    "        w = x.size[0]\n",
    "        h = x.size[1]\n",
    "        x = x.crop((w/2-crop_size/2, h/2-crop_size/2,\n",
    "                    w/2+crop_size/2, h/2+crop_size/2))\n",
    "        class_blocks[pos % N_BLOCKS[c]].append(np.asarray(x, dtype=np.uint8))\n",
    "        pos += 1\n",
    "    for b in xrange(N_BLOCKS[c]):\n",
    "        np.save('../data/X_{0}_{1}.npy'.format(c, b), np.asarray(class_blocks[b], dtype=np.uint8))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### save validation set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 123,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "480"
      ]
     },
     "execution_count": 123,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(val)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 124,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "RNG(seed=9999).shuffle(val)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "y_val = np.asarray(zip(*val)[1])\n",
    "crop_size = c = 1024\n",
    "X_val = np.zeros((len(y_val), crop_size, crop_size, 3), dtype=np.uint8)\n",
    "pos = 0\n",
    "for x, _ in progress_iter(val, True):\n",
    "    x = Image.open(x)\n",
    "    w = x.size[0]\n",
    "    h = x.size[1]\n",
    "    x = x.crop((w/2-c/2, h/2-c/2,\n",
    "                w/2+c/2, h/2+c/2))\n",
    "    X_val[pos] = np.asarray(x, dtype=np.uint8)\n",
    "    pos += 1\n",
    "np.save('../data/X_val.npy', X_val)\n",
    "np.save('../data/y_val.npy', y_val)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### [for pseudo-labels] split validation for training, validation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "N_VAL_TRAIN_PER_CLASS = 24\n",
    "N_VAL_TRAIN = N_VAL_TRAIN_PER_CLASS * 10\n",
    "D = 1024\n",
    "d = 512\n",
    "X_val_train = np.zeros((N_VAL_TRAIN, D, D, 3), dtype=np.uint8)\n",
    "X_val_val = np.zeros((len(y_val) - N_VAL_TRAIN, d, d, 3), dtype=np.uint8)\n",
    "y_val_train = []\n",
    "y_val_val = []\n",
    "pos_val_train = 0\n",
    "pos_val_val = 0\n",
    "for i in xrange(len(y_val)):\n",
    "    c = y_val[i]\n",
    "    if y_val_train.count(c) < N_VAL_TRAIN_PER_CLASS:\n",
    "        X_val_train[pos_val_train] = X_val[i]\n",
    "        pos_val_train += 1\n",
    "        y_val_train.append(c)\n",
    "    else:\n",
    "        X_val_val[pos_val_val] = X_val[i][D/2-d/2:D/2+d/2, D/2-d/2:D/2+d/2, :]\n",
    "        pos_val_val += 1\n",
    "        y_val_val.append(c)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 106,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "np.save('../data/X_val_train.npy', X_val_train)\n",
    "np.save('../data/y_val_train.npy', np.asarray(y_val_train))\n",
    "np.save('../data/X_val_val.npy', X_val_val)\n",
    "np.save('../data/y_val_val.npy', np.asarray(y_val_val))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### [!!!] cyclicly add `X_val_train` to all blocks"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 107,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(240, 1024, 1024, 3)\n"
     ]
    }
   ],
   "source": [
    "X_val_train = np.load('../data/X_val_train.npy')\n",
    "y_val_train = np.load('../data/y_val_train.npy')\n",
    "print X_val_train.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "N_BLOCKS = [21, 16, 16, 17, 12, 19, 31, 16, 31, 23]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 127,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "for c in xrange(10):\n",
    "    print c\n",
    "    c_ind = np.arange(len(y_val_train))[y_val_train == 0]\n",
    "    for block_num, i in enumerate(c_ind):\n",
    "        block_num = block_num % N_BLOCKS[c]\n",
    "        X_block = np.load('X_{0}_{1}.npy'.format(c, block_num))\n",
    "        X_block = np.concatenate((X_block, X_val_train[i].reshape((1, 1024, 1024, 3))))\n",
    "        N_IMAGES_PER_CLASS[c] -= len(X_block)\n",
    "        np.save('X_{0}_{1}.npy'.format(c, block_num), X_block)\n",
    "        \n",
    "print N_IMAGES_PER_CLASS"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### check count"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "N_IMAGES_PER_CLASS = [1014, 746, 767, 807, 598, 918, 1492, 790, 1478, 1081]\n",
    "for i in xrange(10):\n",
    "    N_IMAGES_PER_CLASS[i] += 24"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "for c in xrange(10):\n",
    "    for b in xrange(N_BLOCKS[c]):\n",
    "        N_IMAGES_PER_CLASS[c] -= len(np.load('../data/X_{0}_{1}.npy'.format(c, b)))\n",
    "    print c, N_IMAGES_PER_CLASS[c]\n",
    "print N_IMAGES_PER_CLASS"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### generate `N_IMAGES_PER_BLOCK`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0\n",
      "1\n",
      "2\n",
      "3\n",
      "4\n",
      "5\n",
      "6\n",
      "7\n",
      "8\n",
      "9\n"
     ]
    }
   ],
   "source": [
    "N_BLOCKS = [21, 14, 16, 16, 12, 18, 31, 16, 18, 22]\n",
    "N_IMAGES_PER_BLOCK = []\n",
    "for c in xrange(10):\n",
    "    print c\n",
    "    N_IMAGES_PER_BLOCK.append([])\n",
    "    for b in xrange(N_BLOCKS[c]):\n",
    "        N_IMAGES_PER_BLOCK[c].append(len(np.load('../data/X_{0}_{1}.npy'.format(c, b))))"
   ]
  }
 ],
 "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": "2.7.12"
  },
  "latex_envs": {
   "LaTeX_envs_menu_present": true,
   "autocomplete": true,
   "bibliofile": "biblio.bib",
   "cite_by": "apalike",
   "current_citInitial": 1,
   "eqLabelWithNumbers": true,
   "eqNumInitial": 1,
   "hotkeys": {
    "equation": "Ctrl-E",
    "itemize": "Ctrl-I"
   },
   "labels_anchors": false,
   "latex_user_defs": false,
   "report_style_numbering": false,
   "user_envs_cfg": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}