create_graphs_for_paper.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXwAAAEACAYAAACwB81wAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAIABJREFUeJzt3Xt4VeWV+PHvCiFyq0gREQnCICJo8QKKCAIHA0HwgjrY\nCip4qeCt0591rFptpTOdGZ2Zdjqt09ZWRaxU2nqXCiQCARGlIMhFEaGKICq0AlJBDIT398c6R2LI\n5SRnn/Puffb6PE8eTk529l7ksvKetd93veKcwxhjTP4r8B2AMcaY3LCEb4wxMWEJ3xhjYsISvjHG\nxIQlfGOMiQlL+MYYExOBJHwReUhEtorIqjo+PlREdorI8uTb3UFc1xhjTPoKAzrPVODnwKP1HLPQ\nOXdhQNczxhjTSIGM8J1zi4AdDRwmQVzLGGNM0+Syhn+WiLwuIn8SkRNzeF1jjDEEV9JpyGvAsc65\nPSIyCngG6JmjaxtjjCFHCd8592m1x7NE5Bci8lXn3Paax4qINfcxxphGcs41WDYPsqQj1FGnF5GO\n1R73B6S2ZJ/inAvV2z333OM9Bospf2IKa1wWU3RjSlcgI3wR+R2QANqLyCbgHqBIc7f7NTBWRG4A\n9gGfAd8I4rrGGGPSF0jCd86Nb+Dj/wf8XxDXMuG0fz/s3u07CmNMfWylbRoSiYTvEA4Rtpgefhju\nvz/BY4/5juTLwvZ1SgljXBZTesIYU7qkMfWfXBARF7aYTMMuughOPBGeegoGDID774c2bXxHZUw8\niAguxzdtTUxVVsL8+XDLLbBsGYjA6afDypW+IzPGVGcJ32Ts5ZfhhBOgQwcd1U+dCnffDcOHwy9+\nAfaCzZhwsIRvMjZ7Nowa9eXnrrhC/xA8+CCMHQs7Gmq8YYzJOkv4JmOzZ8O55x76fM+e8MorUFwM\np52mj40x/thNW5ORLVvg5JNh2zZo1qzu4559FiZN0jr/d78LBTbUMCYwdtPW5MScOTBiRP3JHmDM\nGL2hO3OmvhrYujU38RljDrKEbzIya1bt5ZzadOkCFRVw5pla4ikvz2poxpgarKRjmmz/fp2Z8+ab\n0KlT4z537lyYMAEmToQf/hCaN89OjMbEgZV0TNYtWQLdujU+2QOUlMCKFbB8OSQS8N57QUdnjKnJ\nEr5pssaUc2pz1FHwwgu6SveMM3SVrjEmeyzhmyarazpmYxQUwG23wXPPwa23ws03w969wcRnjPky\nS/imSbZtgw0bYODAYM43YICWeLZu1cfr1gVzXmPMQZbwTZPMmQPnnBPszdYjjoA//AFuuAHOPhum\nTQvu3MYYS/imiYIo59RGBCZPhnnz4L77dCbP3/8e/HWMiSNL+KbRqqqgrCw7CT+lTx9YuhSKirTz\n5ooV2buWMXERyoRv0/DD7bXXdIbNscdm9zqtW2vztSlToLQUfv5z+9kwJhOhTPgvv+w7AlOfbJVz\n6jJunDZemzYNLr4Ytm/P3bWNySehTPi/+Y3vCEx9amuHnG09euhAoHt3bcuwaFFur29MPghla4W2\nbR3vvgvt2vmOxtS0fbuurt22DVq08BPDzJnwzW/Ct74Fd9zRcOM2Y/JdpFsrnHsuTJ/uOwpTm/Jy\nGDLEX7IHOP987bxZVgYjR8KHH/qLxZgoCWXCv+46LeuE7MWHwU85pzbFxdqA7eyzoW9fjcsYU79Q\nlnSqqhzHHw+PPw79+/uOyKQcOACdO2v9/LjjfEdzUEWFbql4+eXwox9Z500TP5Eu6RQU6Cj/17/2\nHYmpbtUq3aQ8TMketNvmihWwZg0MHgzvvus7ImPCKZQJH+Cqq+DJJ22VZZiEpZxTmw4d4Pnn4etf\n1w1WnnjCd0TGhE9oE/7RR8OwYVrWMeGQaTvkbCsogO98B/70J7j9du3J89lnvqMyJjxCm/BBN722\nsk44fPLJwc1Kwu6MMzTWHTvgrLNsFo8xKYEkfBF5SES2isiqeo75mYisF5HXReTUdM47YgT89a/W\nRyUM5s3TVsitWvmOJD1t2+qrw0svtbq+MSlBjfCnAiPr+qCIjAKOc84dD0wGfpXOSZs1g2uvtZW3\nYRD2ck5tROCuu+CWW3TtwJtv+o7IGL8CSfjOuUXAjnoOGQM8mjx2CdBWRDqmc+5rroEZM2D37szj\nNE3jXLhv2DbkppvgP/5D99Fdtsx3NMb4k6safmdgc7X3tySfa1BxMQwapBtjGD/WrtUboiec4DuS\nprviCnjgARg9WuftGxNHob5pm5JaeWv8SJVzpMFlHeF24YXw+9/r1M3nn/cdjTG5V5ij62wBulR7\nvzj5XK2mTJnyxeNEIsHo0QluuEEX1nzta1mL0dRh9mzdXDwfDBumzdcuvBB+/GNdnWtM1FRUVFDR\nhJeqgbVWEJFuwPPOuT61fGw0cJNz7jwRGQD81Dk3oI7zuNpi+v73Ydcu+N//DSRck6bdu3VNxAcf\nwFe+4jua4Lzxhr5qufNOuPFG39EYk5l0WysEMsIXkd8BCaC9iGwC7gGKAOec+7Vz7gURGS0iG4Dd\nwNWNvca11+pWd/feCy1bBhG1Scf8+fp1z6dkD3DSSbBggU79/eQTbbMc9ZKVMQ0JJOE758ancUxG\nRYFu3aBfP223cMUVmZzJNEaUZ+c0pHt3bQQ3YoQu0rrvPkv6Jr9F4qZtyqRJdvM213K9nWGudeqk\nI/0FC2DyZN2g3Zh8FamEf8EFsG6dvpnsW79ee9H0OeSuTH5p3x5efBH+8hcYPx4qK31HZEx2RCrh\nFxXBxIk2ys+V1Og+DmWOr3xFm659/jlcdBHs2eM7ImOCF6mED7qX6aOP6i+mya58L+fU1KKFtlXu\n0EG3TvzkE98RGROsyCX844/XufjPPus7kvy2dy+89BIMH+47ktwqLISpU+G003TO/rZtviMyJjiR\nS/hgu2HlwsKFcPLJ0K6d70hyr6BA13ucf7522ty0yXdExgQjkgn/4oth5Uq9yWayI27lnJpE4F/+\nBa6/XpP+22/7jsiYzEUy4bdoAVdeCQ895DuS/BXFdsjZcMstcM89uvHL66/7jsaYzATWWiEodbVW\nqGntWjjnHH253bx5DgKLkY0boX9/+OgjLW8YXfB3443w1FPavdWYMEm3tUJkf51794YePbQRlgnW\nnDk6S8WS/UH/+I/w29/qlM3Zs31HY0zTRPpX2tomZ4eVc2pXWqqzwyZOhD/+0Xc0xjReZEs6oItj\nunTRDau7ds1yYDFRWanz0Dds0H/NoVau1P5C//qv2tTPGN/yvqQDuqH2+PHw8MO+I8kfixfrzlaW\n7Ot2yinae+dHP9Ke+sZERaQTPmhZ5+GHYf9+35HkByvnpOf443Wtwm9+A3ffrfv+GhN2kU/4J58M\nxxxjN9KCks/tkIPWpYuuRp41C771LThwwHdExtQv8gkfrG1yUD74ADZvhjPO8B1JdHToAPPmwapV\nejN33z7fERlTt7xI+N/4ho60ttS5S65Jx+zZuhlIYa52Os4Tbdvq1277dhg7VvsQGRNGeZHw27SB\nSy/Vplem6ayc03StWsHTT+u/o0fD3//uOyJjDpUXCR+0rPPQQ1ZHbar9+3UTkJEjfUcSXUVF8Nhj\n0LMnlJTAxx/7jsgEbdky+O//9h1F0+VNwu/XTzs7vvii70iiackSOPZY3fLPNF2zZvDLX2rbjyFD\nrMyYb6ZP16m4UZ2VlTcJH6xtciasnBMcEbj3Xm3wN3iwdXXNJ3Pm6MY4q1f7jqRp8irhjx8Pc+fC\n1q2+I4meuLdDzoY77oDvfheGDoU1a3xHYzK1ebNuiDNhApSV+Y6mafIq4bdtq73yp03zHUm0bNum\nG5YPHOg7kvxz/fXwX/+lO4ctWeI7GpOJsjL9Po4apSP9KMqrhA8HG6pFtcbmQ1mZ1pytzXR2jBun\nEwouuEBfgZpoSnWRHTYMXn01mhvd513CHzBAN0ipqPAdSXRYOSf7zjtPO2yOGwfPPOM7GtNYVVX6\nx3rECDj8cN3zeOFC31E1Xt4lfBFrm9wYBw7oyMUSfvYNHaptGL75TbuRGzXLlukMtuJifb+0NJp1\n/LxL+ABXXAEvvGDzoNPx2mtw1FE6JdNkX79+cNVVNpssalLlnJSRI6NZx8/LhP/Vr2q99NFHfUcS\nflbOyb1Jk+CRR+Dzz31HYtJVVqaj+pS+fXU24Pvv+4upKfIy4YPdvE2XtUPOvZ494Wtf01YMJvw+\n+UQ3vRky5OBzzZrpjJ3ycn9xNUUgCV9EzhWRt0TkbRG5vZaPDxWRnSKyPPl2dxDXrc/gwVqfXrw4\n21eKru3bdX744MG+I4mfyZPhgQd8R2HSMXeuTllu2fLLz5eWRq+sk3HCF5EC4H5gJHASME5EetVy\n6ELnXN/k248yvW7DcdnK24a8+KKOWlq08B1J/Fx0EaxdC2+95TsS05Ca5ZyU0lL9Haqqyn1MTRXE\nCL8/sN45955zbh8wAxhTy3EN7rcYtAkTdNPpnTtzfeVosHKOP0VFcM01NiAJO+cOvWGbUlwMHTvq\nntpREUTC7wxsrvb++8nnajpLRF4XkT+JyIkBXLdBHTroN2r69FxcLVqcs/45vl13nU4s+Owz35GY\numzYAJWVcNJJtX985MhoTc/M1VYXrwHHOuf2iMgo4BmgZ10HT5ky5YvHiUSCRCLR5AtPmgTf+Q7c\neKOWeYxatUr3ETjuON+RxNc//IPuLvbEE9pozYTPnDlauqkrd5SWwr//O9x1V27jqqiooKIJq0vF\nZTiNRUQGAFOcc+cm378DcM65++r5nHeBfs657bV8zGUaU3UHDuiG048/Dv37B3bayLv3Xm3d+/Of\n+44k3p55RnvtvPyy70hMbS64AC6/HC67rPaP79mjZZ0tW3QFri8ignOuwSFtECWdpUAPEekqIkXA\nZcBzNYLpWO1xf/QPzSHJPhsKCnRlo628/TIr54TD+efDe+9Ft91uPqushAULdPplXVq10nYu8+fn\nLq5MZJzwnXNVwM1AGfAGMMM5t1ZEJovIpORhY0VkjYisAH4KfCPT6zbG1Vfry2bbdk7t2qUrbIcO\n9R2JKSyEa6+1KZphtHgxnHACHHlk/cdFqY6fcUknaEGXdFIuuURnpEya1PCx+e7pp3VXpqj8kOa7\nzZvh1FNh0yZo3dp3NCblzjt1gdWPGphEvmqV5pcNG3ITV21yWdKJBGuodpCVc8KlSxcYNAhmzPAd\niamurvn3NfXpA7t3R6MhXmwSfmmpbvSxYoXvSPxKTce0+ffhcv31VtYJk23bNIGfdVbDx4pEp3tm\nbBJ+s2ZaK437KH/tWv23V21roY03I0dqM67XXvMdiQFdQZtIpL8pUFTq+LFJ+KArG2fM0JdfcZUq\n59iahHBp1kzvL9koPxxS8+/TNXy4ztTZty97MQUhVgm/uFhrpX/4g+9I/LFyTnhdc43uirVrl+9I\n4s05Ha3X1k6hLkcdpYsYw75vcawSPsT75u3u3fDKK7p/rQmfTp2gpMRagfi2erXOlmrsKvQodM+M\nXcIfPVoXuqxZ4zuS3KuogNNP97si0NQv1TY5ZLOlY6Wx5ZyUKNTxY5fwCwv1pXMcR/lWzgm/khL4\n9NPwlwbyWV3dMRsycKC2uw7z1qqxS/igs3WmT4e9e31HklvWDjn8Cgrs5q1Pe/bAq6/CsGGN/9yi\nIt1fYu7c4OMKSiwTfrduupn0k0/6jiR3NmzQH+aTT/YdiWnI1VfraugdO3xHEj8LFuh+tU0te4a9\njh/LhA86iorT5hOpco5Nxwy/Dh106uxvf+s7kvhp7OycmlJ1/LDeg4ltwr/gAli3Tt/iwMo50XL9\n9fCrX4U3ceSrpt6wTTn+eF1TkVrgGDaxTfhFRTBxIjz4oO9Ism/vXnjpJRgxwnckJl1DhmiyX7TI\ndyTxsXmztlTo27fp5xAJ92yd2CZ80D7506bB55/7jiS7XnpJGzy1a+c7EpMuEZ2i+atf+Y4kPsrK\ndFDUrFlm5wlzHT/WCf/44+FrX9ONzvOZlXOiacIEeOEF+NvffEcSD5mWc1JKSnQHszDOAox1wod4\nrLy1dsjR9NWvwoUXwiOP+I4k/1VVacO0IBL+EUfoQDKM5bjYJ/yLL4bXX49GL+umeO89HSFmUpc0\n/lx/vc4mO3DAdyT5bdky6NxZ34IQ1jp+7BN+ixZw5ZXw0EO+I8mO2bN11FIQ++90NA0YoD+jUdkz\nNaqCKuekhLWOb2kALetMnRr+1qZNYeWcaBOxzVFyIdP59zWdcYbO+vnww+DOGQRL+EDv3tCjB8yc\n6TuSYFVW6sgwyJGLyb3LL4fycvjoI9+R5KdPPoGVK2Hw4ODOWVioXWnLy4M7ZxAs4Sfl483bxYt1\nJlKHDr4jMZlo2xbGjtVXoSZ4c+dq47OWLYM9bxjr+Jbwk8aO1Q6Fmzb5jiQ4Vs7JH5Mn283bbAm6\nnJNSWqoj/DB9zyzhJ7VqBePH59fNW2uHnD9OPx3atw/fiDHqnAv+hm1K16662HHlyuDP3VSW8Ku5\n7jp4+GGdkxt1H3ygr1b69/cdiQmKrbwN3vr1OlnjpJOyc/6wzdaxhF/NySfDMcfoyDjq5szRZeKF\nhb4jMUEZNw4WLoT33/cdSf4oK9OknK0usmGr41vCryFf2iZbOSf/tGkDl12WX2VH37JVzkkZOhSW\nLtVdzMJAXMj6r4qI8xnTp5/CscfqnrfHHOMtjIzs3w9HHRXt/4Op3cqVcN55sHGjvXrLVGUlHHkk\nvPOO/pstw4bBP/+zft+yRURwzjX4OsVG+DW0aQOXXhrtKXB//rP+0bJkn39OOQW6dNGmaiYzixfD\nCSdkN9mDlnXCUscPJOGLyLki8paIvC0it9dxzM9EZL2IvC4ipwZx3WyZNEn75IdpOlVjWDknv6U2\nRzGZaepm5Y1VWhqeOn7GCV9ECoD7gZHAScA4EelV45hRwHHOueOByUCof1z79dPpVC++6DuSprF2\nyPnt61/XV3EbN/qOJNqyNf++plNPhe3btZGhb0GM8PsD651z7znn9gEzgDE1jhkDPArgnFsCtBWR\njgFcO2uiuvJ22zadajZwoO9ITLa0bAlXXBHNn8+w2LZNO+QOGJD9axUU6Iy5MIzyg0j4nYHN1d5/\nP/lcfcdsqeWYUBk/Xkf4W7f6jqRxysv1JlFRke9ITDZNnqxrRvKx4V8ulJdDIgHNm+fmemGp44fy\nPv+UKVO+eJxIJEgkEjmPoW1b7ZU/bRp897s5v3yTWTknHnr3hp49dbe2sWN9RxM9uSrnpIwYAf/v\n/+kMuiBmV1VUVFBRUdHoz8t4WqaIDACmOOfOTb5/B+Ccc/dVO+ZXwHzn3O+T778FDHXOHTJ+9j0t\ns7pXXtFt5t5+O3sLM4J04AAcfbTO++3a1Xc0Jtsef1xH+WHryBh2zukMtkWL4LjjcnfdU07Rm+1n\nnRX8uXM5LXMp0ENEuopIEXAZ8FyNY54DJiQDGwDsrC3Zh82AAVovvfNO+Owz39E0bPlynWJmyT4e\nLrlE5+Vv2OA7kmhZvRpat85tsodwzNbJOOE756qAm4Ey4A1ghnNurYhMFpFJyWNeAN4VkQ3AA8CN\nmV43F0R0vvM772ivjeee09FBWFk5J14OOwyuuio/VobnUq6mY9YUhjq+rbRNU3k5fOtbOir42c9y\nPzpIx6BBMGWK1gtNPGzYoDOyNm/WPwCmYcOH6+/ymJpzCbNs715dAb9pk250HiRbaRuwESNg1SoY\nMgTOPBN+8APYs8d3VAft2KEvVYPctceEX48eWht+6infkUTDnj3w6qs6ky3XWrTQQdncubm/dool\n/EYoKoLbb4fXX4d167TM8+yz4SjzlJdrsm/RwnckJtesbXL6FiyAvn3h8MP9XN93Hd8SfhMUF8Pv\nf68LX+64A84/3/+NM9vdKr7GjNGZZGvX+o4k/HzV71NSdXxfg0RL+BkYPlxnSQwdqjN6fJV5nLP+\nOXHWvDlccw088IDvSMIv1/Pva+rdW+fir1/v5/qW8DNUVKQLs3yWeVat0i0ae/TI3TVNuFx3HTz2\nWDSmD/uyebO2VDjtNH8xiPidrWMJPyCpMs+DD+q8/fPOy12Zx8o5pls33c7yD3/wHUl4lZXp5Itm\nzfzG4bOObwk/YCUlOtofNkzLPN//fvbLPFbOMaBtk62sUzff9fuU4cP15nFlZe6vbQk/C4qK4Lbb\nNPGvXw8nngjPPJOdMs+uXbBsmTaCMvE2erSWLVat8h1J+FRVaTPEMKxRad8eevXSDVhyzRJ+FhUX\nw4wZ2u/ke9/TX8igb9bMm6e9OVq3Dva8JnoKC+Gb37RRfm2WLYPOnfUtDHzV8S3h58A55+hov6RE\nk/PddwdX5rFyjqnu2mu1qVpYNs0Oi7CUc1J81fEt4edIUZFuZLxypW68cOKJ8PTTmZV5nLP+OebL\niot1NfiMGb4jCZc5czTJhsWAAZoH/vrX3F7XEn6Ode6sI7CpU3WkP2pU08s8b72l//buHVx8Jvps\n5e2X7dyp9zXC1HakeXO975br1taW8D0ZNkzLPCNGaJnnrrtg9+7GnSNVzolCr36TO6Wl8PHHWrc2\nep9r4EBtdR4mI0fmvqxjCd+j5s3h1lt19PHuu1rmeeqp9Ms8Vs4xtWnWTBdi2c1b5Xt1bV1Sdfxc\nLtK09sghUlEBN90EXbpoC+aePes+dvdu3d1qyxZ/jaBMeH30kZb6Nm7U7Trjyjno3h1mztRV8GHT\no4fey+vTJ7PzWHvkCEoktMxTWqovQesr8yxYAP36WbI3tTv6aC0XTp/uOxK/1q/Xjd5PPNF3JLUr\nLc3t9ExL+CHTvDl85zta5tm4UX9Qn3zy0Jd9s2ZZOwVTv9TN25i+YAa0ZFJaGt77XLmu41vCD6lj\njtHR2bRpcM89Wqt/++2DH7f596Yhw4bpLkuvvuo7En/CNv++pmHD4JVXctdl1xJ+yCUSsGKFJveB\nA3XF7sqVWuo5+WTf0ZkwKyiASZPiO0WzslJLn8OH+46kbocfrt07X3opN9ezhB8BzZvDLbfoFoab\nNukWizYd06Tjqqvguedg+3bfkeTe4sXas6Z9e9+R1C+XdXxL+BHSqZP2PJ8/X2/oGtOQI4/UHk6P\nPuo7ktwLezknJZd1fJuWaUyee+klLe28+Wa8XhX26wc//Wm4VtjWpqoKOnbUGXrFxU07h03LNMYA\ncPbZmugXLvQdSe5s26a9agYM8B1Jw5o10/sMuWizYAnfmDwnEr/NUcrLdcJD8+a+I0lPrur4lvCN\niYErr4QXXsh9d0ZfwtpOoS6lpbpBS1VVdq9jCd+YGGjXDi6+GB55xHck2edc9BJ+cbHW8Zcvz+51\nLOEbExOTJ2tZ58AB35Fk16pVugNc9+6+I2mcXMzWsYRvTEyceSa0aaPtgvNZ1Eb3Kbmo42eU8EWk\nnYiUicg6EZkjIrX25RORjSKyUkRWiMifM7mmMaZpROKxOUpU5t/XNGSIrqrftSt718h0hH8H8KJz\n7gRgHnBnHccdABLOudOcc/0zvKYxpokuvxzmzoUPP/QdSXbs2QNLlugMnahp1Uqnkc6fn71rZJrw\nxwDTko+nARfVcZwEcC1jTIYOPxwuvRQefth3JNmxYAH07RvdtuHZruNnmoSPcs5tBXDOfQQcVcdx\nDigXkaUicl2G1zTGZGDyZPjNb7I/BdCHqJZzUrJdxy9s6AARKQc6Vn8KTeB313J4XT0RBjnnPhSR\nDmjiX+ucW1TXNadMmfLF40QiQSKKr8+MCal+/aBDB00so0f7jiZYZWXR7hvUp492wv3LX+C44+o+\nrqKigoqKikafP6NeOiKyFq3NbxWRo4H5zrneDXzOPcDfnXM/qePj1kvHmCx76CHtovnss74jCc7m\nzVrO2bpVW0NH1cSJWsu/4Yb0PydXvXSeA65KPp4IHPLjIyKtRKRN8nFroBRYk+F1jTEZuOwyWLRI\nk2S+mDNHe9JEOdnDwc3NsyHTL819wAgRWQeUAPcCiEgnEZmZPKYjsEhEVgCvAs8753K4qZcxpqbW\nrWHcOB3p54uozr+vacQInamzb1/w57b2yMbE1OrVui/yxo1Q2ODdvHCrqtL7EmvW6PagUde3L/zs\nZ9rpNB3WHtkYU68+faBrV5g5s+Fjw27ZMujcOT+SPegrlWzM1rGEb0yM5Uvb5KhPx6wpW3V8S/jG\nxNjYsbB0Kbz7ru9IMpNvCX/gQFi7Fj7+ONjzWsI3JsZatoQJE3QhVlTt3KkdMtOtd0fBYYdpb525\nc4M9ryV8Y2Ju0iRttVBZ6TuSppk3DwYN0j9e+SQbdXxL+MbEXK9ecMopcOedunlI1ORbOSclVccP\n8ntiCd8Yw+9+p43Hvv3taCV95zThl5b6jiR4PXvqBudr1wZ3Tkv4xhjat9c9VZcu1Zk7UdkVa/16\n2L8fTjzRdyTBEwl+to4lfGMMAEccocll7Vq49tpodNNMje6lwSVH0RR0Hd8SvjHmC1/5CsyaBZs2\nwZVX6ug5zPKlnUJdzjlHex7t3RvM+SzhG2O+pHVrXX27Y4c2WQvr7J3KSr3vMHy470iyp107XRG9\nqM5m8o1jCd8Yc4iWLeGZZ7SB19ix8PnnviM61OLFOsOofXvfkWRXkHV8S/jGmFoddhj88Y/675gx\n8NlnviP6snydjllTkHV8S/jGmDoVFcHjj+so+rzzdDemsIhLwj/jDL2nEsTG85bwjTH1KizUbQO7\ndoVzz4Vdu3xHBNu2wTvvwJln+o4k+woLoaQEysszP5clfGNMg5o1081STjpJa8o7d/qNp7wchg2D\n5s39xpErQdXxLeEbY9JSUAC//KXut1pSEnwnx8aISzknJZXwM10QZwnfGJM2Efif/9GpkMOGaWkl\n15zT5JeP7RTq0q2bTtFcuTKz81jCN8Y0igjcey9cfDEkEsHcTGyMVat0gVj37rm9rm9BzNaxhG+M\naTQR+OEP4fLLYehQ2Lw5d9eO2+g+JYg6viV8Y0yT3XUXTJ6sST9Xu2bFrX6fkkhoc7tPP236OSzh\nG2Mycuut+pZIaPfKbNq9G5Ys0fsHcdOmDZx+uraTaCpL+MaYjN10E9x9tybiIPu317RwIfTtqzX8\nOMq0jl8Vrj+KAAAJB0lEQVQYXCjGmDi77jpdmVtSokmpT5/grxHXck5KaSmMH9/0z7cRvjEmMBMn\nwk9+AiNGwPLlwZ8/7gn/1FNh+3Z4772mfb4lfGNMoC67DH7xCxg1SuvtQdm0Cf72NzjttODOGTUF\nBfrHtKmzdSzhG2MCd8kl2orhgguC6+VeVqbJriDmWSuTOn7Mv3TGmGw5/3x47DFdoDV/fubni+v8\n+5pGjIB585q2G1lGCV9ExorIGhGpEpG+9Rx3roi8JSJvi8jtmVzTGBMdpaXaU//rX89sdklVFcyd\nawkfoFMn6NJF5+Q3VqYj/NXAxUCdM0NFpAC4HxgJnASME5FeGV7XGBMRiYTunnXllbp1YlMsXQqd\nO8MxxwQaWmQ1ddVtRgnfObfOObceqG/P+P7Aeufce865fcAMYEwm1zXGRMugQZrsr70Wnnqq8Z9v\n5ZwvGznSQ8JPU2egeqeN95PPGWNipH9/mDULbrwRZsxo3OfGfTpmTWefDatXN35fggYXXolIOdCx\n+lOAA+5yzj3fuMsZY+Ksb1/dvGTkSN0YfeLEhj9n507tkDl4cPbji4oWLfRV07x5OiMqXQ0mfOfc\niEwCA7YAx1Z7vzj5XJ2mTJnyxeNEIkEikcgwBGNMWPTpo4lq+HCorNQVuvWZN0+TW4sWuYkvCioq\nKjhwoIJ/+zf9Y5gucc5lfHERmQ/8s3PutVo+1gxYB5QAHwJ/BsY552rtuCEiLoiYjDHhtmGDtmG4\n7Ta4+ea6j5s8GXr1gltuyV1sUfDmmzB6tHYpLSgQnHP13UsFMp+WeZGIbAYGADNFZFby+U4iMhPA\nOVcF3AyUAW8AM+pK9saY+OjRQzs//uQn8OMf136Mc1a/r0vv3joXvzEdSjNqnuacewZ4ppbnPwTO\nr/b+bOCETK5ljMk/3bpp0i8pgb17tb9+devXa1Lr3dtLeKEm0vjZOrbS1hjjVZcumvSnT4cf/EBH\n9Smp0b00WKyIp9LSxi1os4RvjPGuUyeoqNAFWnfccTDp2/z7+g0f3rgNUQK5aRsku2lrTHx9/LEm\n+LPPhv/8TzjqKHjnHWjf3ndk4dW/PyxdmoObtsYYE6T27bVnzpIlmvh79bJk35DGvAKyhG+MCZUj\njtBSjgiMsSYsDWrMDCYr6RhjQsk5fYt7//uG7NsHRUXplXRsT1tjTCiJ2OycdDRvnv6x9rfTGGNi\nwhK+McbEhCV8Y4yJCUv4xhgTE5bwjTEmJizhG2NMTFjCN8aYmLCEb4wxMWEJ3xhjYsISvjHGxIQl\nfGOMiQlL+MYYExOW8I0xJiYs4RtjTExYwjfGmJiwhG+MMTFhCd8YY2LCEr4xxsSEJXxjjIkJS/jG\nGBMTlvCNMSYmMkr4IjJWRNaISJWI9K3nuI0islJEVojInzO5pjHGmKbJdIS/GrgYWNDAcQeAhHPu\nNOdc/wyvmXMVFRW+QziExZSeMMYE4YzLYkpPGGNKV0YJ3zm3zjm3HpAGDpVMr+VTGL/BFlN6whgT\nhDMuiyk9YYwpXblKwg4oF5GlInJdjq5pjDGmmsKGDhCRcqBj9afQBH6Xc+75NK8zyDn3oYh0QBP/\nWufcosaHa4wxpqnEOZf5SUTmA7c655ancew9wN+dcz+p4+OZB2SMMTHjnGuotN7wCL8Rar2YiLQC\nCpxzn4pIa6AU+GFdJ0knaGOMMY2X6bTMi0RkMzAAmCkis5LPdxKRmcnDOgKLRGQF8CrwvHOuLJPr\nGmOMabxASjrGGGPCLzRTJUXkXBF5S0TeFpHbfccDICIPichWEVnlO5YUESkWkXki8oaIrBaRfwpB\nTIeJyJLkwro3ROTffceUIiIFIrJcRJ7zHQuEcxGiiLQVkT+KyNrk9+/MEMTUM/k1Wp7895OQ/Kzf\nmfwarRKR6SJSFIKYvp3MBQ3nA+ec9zf0D88GoCvQHHgd6BWCuM4GTgVW+Y6lWkxHA6cmH7cB1oXk\na9Uq+W8ztHQ3yHdMyXhuAR4DnvMdSzKed4B2vuOoEdMjwNXJx4XA4b5jqhFfAfAB0MVzHF2T37+i\n5Pu/ByZ4jukkYBVwWPJ3rwzoXtfxYRnh9wfWO+fec87tA2YAYzzHhNOpozt8x1Gdc+4j59zrycef\nAmuBzn6jAufcnuTDw9BfUO9fNxEpBkYDD/qOpZpQLUIUkcOBwc65qQDOuf3OuV2ew6ppOPAX59xm\nz3HsAiqB1iJSCLRC/xD51BtY4pz73DlXBSwELqnr4LD84HUGqn8z3ycESSzsRKQb+gpkid9Iviid\nrAA+Aiqcc2/6jgn4H+A2dN1IWIRtEeI/AH8TkanJ8smvRaSl76Bq+AbwuO8gnHM7gB8Dm4AtwE7n\n3It+o2INMFhE2iVnRI4GutR1cFgSvmkkEWkDPAF8OznS98o5d8A5dxpQDAwRkaE+4xGR84CtyVdD\nQsPtP3JlkHOuL/qLeZOInO05nkKgL/B/ybj2AHf4DekgEWkOXAj8MQSxdEdLhF2BY4A2IjLeZ0zO\nubeA+4By4AVgBVBV1/FhSfhbgGOrvV+cfM7UIvly8gngt865Z33HU12yHPAn4HTPoQwCLhSRd9DR\n4TARedRzTDjnPkz++1fgabSc6dP7wGbn3LLk+0+gfwDCYhTwWvLr5dvpwMvOue3J8slTwEDPMeGc\nm+qcO905lwB2Am/XdWxYEv5SoIeIdE3e9b4MCMWsCsI1Okx5GHjTOfe/vgMBEJEjRaRt8nFLYAR6\n490b59z3nHPHOue6oz9P85xzE3zGJCKtkq/MqLYIcY3PmJxzW4HNItIz+VQJEIZyXMo4QlDOSVoH\nDBCRFiIi6NdqreeYSLasQUSORbsX/66uY4NcadtkzrkqEbkZvcNcADzknAvDF/J3QAJoLyKbgHtS\nN7c8xjQIuBxYnayZO+B7zrnZHsPqBExL/hIUoK885nqMJ6w6Ak8n24cUAtNdOBYh/hMwPVk+eQe4\n2nM8wBer9IcDk3zHAuCcW5l8lfgaWjZZAfzab1QAPCkiXwX2ATfWd9PdFl4ZY0xMhKWkY4wxJsss\n4RtjTExYwjfGmJiwhG+MMTFhCd8YY2LCEr4xxsSEJXxjjIkJS/jGGBMT/x+xAZNC/YKCEwAAAABJ\nRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f460a613950>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "import numpy as np\n",
    "#from ggplot import *\n",
    "import numpy as np\n",
    "from matplotlib import pyplot as plt\n",
    "from matplotlib import style\n",
    "\n",
    "# use ggplot style sheet\n",
    "#style.use('ggplot')\n",
    "\n",
    "plt.plot(np.random.randn(10))\n",
    "import csv"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYEAAAFVCAYAAAAJ9lMdAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f45dc45fdd0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "str1='''No\t0.389\t2.93\tNo non-linearity\n",
    "$\\\\tanh{}$\t0.401\t2.78\t\n",
    "$1.73\\\\tanh{\\\\frac{2x}{3}}$\t0.423\t2.66\tAs recommended in Efficient BackProp, LeCun98\n",
    "OPLU\t0.435\t2.57\n",
    "VLReLU\t0.469\t2.40\ty=max(x,x/3)\n",
    "ReLU\t0.470\t2.36\tWith LRN layers\n",
    "ReLU, noLRN\t0.471\t2.36\tNo LRN, as in rest\n",
    "APL2\t0.471\t2.38\n",
    "RReLU\t0.478\t2.32\t\n",
    "MaxS\t0.482\t2.30\tsqrt(2) narrower layers, 2 pieces. Same complexity, as for ReLU\n",
    "PReLU\t0.485\t2.29\t\n",
    "CSoftplus\t0.486\t2.29\tShifted BNLL aka softplus, y = log(1 + exp(x)) - log(2). Same as ELU, as expected\n",
    "ELU\t0.488\t2.28\talpha=1, as in paper\n",
    "ConvReLU,FCMax\t0.490\t2.26\tReLU in convolution, Maxout (sqrt(2) narrower) 2 pieces in FC.\n",
    "ConvELU,FCMax\t0.499\t2.22\tELU in convolution, Maxout (sqrt(2) narrower) 2 pieces in FC.\n",
    "MaxW\t0.517\t2.12\tsame width layers, 2 pieces'''\n",
    "names=[]\n",
    "acces=[]\n",
    "losses=[]\n",
    "for name_acc_loss_comm in str1.splitlines():\n",
    "    try:\n",
    "        name,acc,loss,comm = name_acc_loss_comm.strip().split(\"\\t\")\n",
    "    except:\n",
    "        name,acc,loss = name_acc_loss_comm.strip().split(\"\\t\")    \n",
    "    names.append(name)\n",
    "    acces.append(float(acc))\n",
    "    losses.append(float(loss))\n",
    "#Non-linearities\n",
    "accn=100.*(np.asarray(acces[:]) - acces[6])\n",
    "\n",
    "plt.rc('text', usetex=True)\n",
    "fig, ax = plt.subplots()\n",
    "rects1=ax.bar(2*np.asarray(range(len(acces[0:5])))+1, accn[0:5], 1, color='r')\n",
    "rects1=ax.bar(2*np.asarray(range(len(acces[5:])))+11, accn[5:], 1, color='g')\n",
    "rects1=ax.bar(2*np.asarray(range(1))+(len(names)*2-1), accn[-1], 1, color='b')\n",
    "\n",
    "#rects1=ax.bar(2*5+1, accn[5], 1, color='r')\n",
    "\n",
    "# add some text for labels, title and axes ticks\n",
    "ax.set_ylabel('Top-1 acc [\\%]')\n",
    "ax.set_xticks( 2*np.asarray(range(len(names)))+1.5)\n",
    "ax.set_xticklabels( names[:], rotation=60 ) ;\n",
    "ax.set_xlim([0,2*len(names)+1])\n",
    "ax.spines['right'].set_visible(False)\n",
    "ax.spines['top'].set_visible(False)\n",
    "ax.xaxis.set_ticks_position('bottom')\n",
    "ax.yaxis.set_ticks_position('left')\n",
    "plt.suptitle('Non-linearities, gain over ReLU')\n",
    "plt.savefig('../graphs/non-lin.pdf', format='pdf', dpi=1000,bbox_inches='tight')\n",
    "#ax.legend(names)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXwAAAFFCAYAAADvklavAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAH0xJREFUeJzt3Xu8nFV97/HPF1BBuV+aqFRIuHipFxAQUGs3JCDWHrUE\nQltFPZUEOEqpBYRga1JfCgE8ba1UA8GXFm+ESzi2Hi0kAY5itRAE6qVaCKDFqi25gCIKyu/8sdYk\nk82efZ2ZtedZ3/frtV975pmZ5/mtPXt+s5611rOWIgIzM2u+bUoHYGZm/eGEb2ZWCSd8M7NKOOGb\nmVXCCd/MrBJO+GZmlXDCNzOrhBN+ZSTtIulJSbdLWivpHkkLJrGfeZKWSjpe0tm9iHWickxFYpE0\nJ/9dDxq2faOkj01wX119j4Ztu1rSjXnfB3V67VjPbe1X0rL8nHskzZtojNZf25UOwIpYFxGHQUou\nwEZg+ST2ExGxsquRTUFEXFc4hHXAqcDpAJIOBh6a7L669R61buQvjXURsSjHdjFw7EgvGu25+f5t\nkuaQ/geOzTHeD5R+D2wUruHXSW239yAnhWE1ujmbn9xhe36sVdOfk5+3LD9vwbDX35AfW5u3zcr7\nvEHSCkk7jxjoCK/N21uvvUHS8eONpVOZ8u2D8mNzJK3It1v7uD0nutZxluVa7b5tu70TOLTt/qnA\ntWPEvEDSsraYju72e9RmFXBZ2/43jhLDiM/NTspf9OuAiwAi4mFgfYfj2jThGn6dZku6nfRBDuBE\nSecAt0XEh9pqa7t32j5sf61a5KyImJ+fdwewPJ/6t14/B2gloxOAO3INck7e5yPtO+30WkmzgGUR\nsTIn4aVA60yjYyxt+x2pTKcAfwDcRUrUF+QvioiIw9r2s3/ezSERccAIf9vVko6OiJtIyf89+e+7\n77CYLwJWRsRySXNzWddHxE35WN1+j4iIB3L5lwELgEPy9qfE0Pa32uq57X/jtv3NBq7O74NNY074\nddrcXNCSP9hXQ6qtSWolzv06bB/J6hGeNxu4Km9fI22uuF4OLJJ0A3AfcO4I++v02g3AsZKOYeua\n8FixtDylTDkRXwScR/qyuEvSacCsXNsfXstdPcIxA1gBnKYU7Kq2+DYOi7k9pvOAe4Hd2rb16j0i\nIk7LyX01W77ARorhKc/NX1ar2mJ6D3Ai8I6IuHu041p5btKp00hJ8l5yLS7X2FrWddjevp/hv9tv\n3wcck18/t+3x+cBVEfHa/JyFI8S0rsNrFwFrI+J04JoRXtcplpZOZV2Tk+qKfP8O4BsRcVJEzG/b\n3lFE3EWq2S9k6yaR4TG3x3QZ6QvvilFiHi3u0d6jLTtMzV2t5q1NbJ3ct4phlOee1DoDkHQCMDci\nDnOyHwyu4dfpKTXA3BxwtaQbSR/uE/L2S0ba3raf4T/Dj3EhcE1O2He2Pb4WuFbSpvzcsyRtiIj2\npoilHV67Argo15bvJzV/tI8i6RTLqGUlJeLWtlZTR/vzLhz+d+tgBbAwIh6QtF8+/lXAxW0xz8ox\nHwPcmGP6WG7bXzM85tHiHuM9atd6L07N+z8BNjdxDY/hAtL7s9Vzh8U1FzhE0j3ks5YOzVw2TcjT\nI1svactIjpta7e25Vj/Sc0+JiCsm81ozG5sTvvVU7kRsNbsEcGqrs2+E57Y6Oyf8WjMbmxO+mVkl\n3GlrZlYJJ3wzs0o44ZuZVcIJ38ysEk74ZmaVcMI3M6uEE76ZWSWKTa3QNk/HfhFxXqk4zMxqUaSG\nny+ZXxURy0nzoBw91mvMzGxqSjXpzCZNvARppsSOM/yZmVl3FGnSyTX7lpeT5zw3M7PeKdppm2dA\nvCPPIW5mZj1Uej78ORGxaKQHJMXixYs33x8aGmJoaKhfcZmZDapOq8CVmy1T0oJW046kORGxZtjj\n4Zk8zcwmrGPCLzlKZ6mkeyWtZ4TVfczMrLum7Xz4ruGbmU3K9Krhm5lZ/znhm5lVwgnfzKwSTvhm\nZpVwwjczq4QTvplZJZzwzcwq4YRvZlYJJ3wzs0o44ZuZVcIJ38ysEk74ZmaVcMI3M6uEE76ZWSWc\n8M3MKuGEb2ZWCSd8M7NKOOGbmVXCCd/MrBJO+GZmlXDCNzOrhBO+mVVt5t4zkdTzn5l7zyxdVBQR\n5Q4uLY2I8zo8FiVjM7M6SIIlfTjQEuhTTlOnB4rV8CUtAOaVOr6ZWW2KJfyIWA7cV+r4Zma1cRu+\nmVklnPDNzCrhhG9mVontCh+/Y28ywJIlSzbfHhoaYmhoqMfhmJk1V7FhmZLmAZcD50bEFSM87mGZ\nZtZzNQ3LLFbDj4jrgOtKHd/MrDZuwzczq4QTvplZJZzwzcwq4YRvZlYJJ3wzs0o44ZuZVcIJ38ys\nEk74ZmaVcMI3M6uEE76ZWSWc8M3MKuGEb2ZWCSd8M7NKOOGbmVXCCd/MrBJO+GZmlXDCNzOrhBO+\nmVklnPDNzCrhhG9mVgknfDOzSjjhm5lVwgnfzKwS25U6sKR5wCZgdkQsLxWHmVktitTwJR0MRESs\nyfcPKhGHmVlNSjXpnESq3QPcB8wtFIeZTcLMvWciqec/M/eeWbqojVKqSWdXYEPb/T0KxWFmk/CT\nH/4ElvThOEt+0vuDVMSdtmZmlSiV8DcCu+fbuwLrC8VhZlaPiOj7D3AwcEq+fQ5w0AjPicWLF2/+\nufnmm6MX9pkxI4Ce/+wzY0ZP4m9qWZpWnsWLF/elLIsXL+55WSIiZjy3P+/NjOf2/r1pUlmyjrlX\nkZJr30k6BbgfmBURV4zwePQjNkn04y8goNflaVJZoHnlMesTdXxguv6jO+FP4hgNKgs0rzxmfdIx\n4bvT1sysEk74ZmaVcMI3M6uEE76ZWSWc8M3MKuGEb2ZWCSd8M7NKOOGbmVXCCd/MrBJO+GZmlXDC\nNzOrhBO+mVklnPDNzCrhhG9mVolR17SVtPNYO4iIR7oXjpmZ9cpYi5g/AKzjqfMrR942Cy9AbmY2\nEMZK+JdHxHmdHpS0tMvxmJlZj0xoxavcxLMQuDYiHuhVUPlYXvFqosdoUFmgeeUx65OurXi1CLgT\nuGhK4ZiZWd+NmvAlnT1s024RsYZRvkHMzGx6GquGv0bS1ZKOz/cvk3QjcFWP4zIzsy4bVxu+pHnA\nfODCiLir51HhNvxJHaNBZYHmlcesTybXhi9pZ0nLgLmkztpjJH1sPOPzzcxsehmrSWc5cBlwOWmI\n5iWkjttF3Ti4h3WamfXPWAlfwEZgU75NRGyKiCknfEkLgHlT3Y+ZmY3PqG34knYhDcEM4LyIeLir\nB5duiIjXdnjMbfgTPUaDygLNK49Zn3Rswx/rSts5EXFax71Kx0fEykmHZWZmfTNWwr9Y0uwOj4nU\nkeuEb2Y2AMZK+OeO8fho8+wsgKeckW+YyBnBkiVLNt8eGhpiaGhovC81M7NhJjSXTtcPLt0YEcd2\neMxt+BM9RoPKAs0rj1mfdG0una7JF3MdIumUUjGYmdVkvFfa7gvMjYgrJF0IXObZMid4HFzDn/Bx\nGlYesz6Zcg3/GmBtvr0m3zczswEy3oS/sTWHTkSsJl2MZWZmA2SsUTotm/JUyauBY0hX3pqZ2QAZ\nVw0/IuaT2oXOB3bP983MbICMK+HnTtuNrcQvaZ9eBmVmZt03mU7b1cC1vQnHzMx6xZ22ZmaVmEyn\n7VzcaWtmNnAm02m7pzttzcwGz6Tm0pF0UK/XtvWVtpM4RoPKAs0rj1mfTHo+/PTqNO/NwrZNs4ED\nphiUmZn10Xg7bReSpkK+H7gYuK5nEZmZWU+Me7bMiLgz/Yo1wKzehWRmZr0w3oQvScfn36cAu/Uw\nJjMz64Fxd9pKmgVsABYBK3KNv3eBudN24sdoUFmgeeUx65OOnbZFV7wajRP+JI7RoLJA88pj1ifT\nb8UrMzPrLyd8M7NKOOGbmVVi1Auv8vw5I4qID3U/HDMz65WxaviPABeRJkt7eNiPmZkNkDFH6Ui6\nMSKO7VM87cf1KJ2JHqNBZYHmlcesTzwsc5TjNCapNKks0LzymPVJd4ZlSrpw6rGYmVkJ410ApWW/\nbh1Y0oLWPiPivG7t18zMRjbRYZkrunFQSXOAVRGxHJgt6ehu7NfMzDqbaMK/o0vHnU1aKhHgvnzf\nzMx6aFwJX9I8SfcCl0u6R9LvT+WgEbE8Iq7Id18OrJ3K/szMbGzjbcM/LyL2b92RdDtw/VQPLulg\n4I5OyyUuWbJk8+2hoSGGhoamekgzs2qNa1jm8LH44xmbnztlh+98Q0SsbHvO2Z2u2PWwzEkco0Fl\ngeaVx6xPpjYOX9Iy0ipXq4Bj8uZrANqaZiYWkbQgd9oiaU5eSav9cSf8iR6jQWWB5pXHrE+mPA5/\nHbA672h1/tmNSa58lUfpLJV0r6T1PPVMwMzMusxX2jaoFtmkskDzymPWJ1Or4UuaI2mDpPX556ju\nxWZmZv0w3iadpcCsiNgDOAy4uHchmZlZL4w34W+MiIcBIuI+YGPvQjIzs14Y7zj8TXkxlG8Ah5Dm\nxzczswEyag1f0gqAiJhP6giY33bfzMwGyKijdCTdHhGH9TGe9mN7lM5Ej9GgskDzymPWJ5O78ErS\nBuCykR6LiEVTj6szJ/xJHKNBZYHmlcesTzom/LHa8DeQZrM0M7MBN1bC39Sa/sDMzAbbWMMyPW2x\nmVlDeGqFBrUTN6ks0LzymPVJdxYxNzOzweWEb2ZWCSd8M7NKOOGbmVXCCd/MrBJO+GZmlXDCNzOr\nhBO+mVklnPDNzCrhhG9mVgknfDOzSjjhm5lVoljClzRP0hxJy0rFYGZWkyIJX9IcYG5ErAFmSzqo\nRBxmZjUpPj1yp3VzPT3yJI7RoLJA88pj1ieTXuKwZyTtAiwELiwVg5lZTaZDDf9GYGFEPDBseyxe\nvHjz/aGhIYaGhnpx/MbUIptUFmheecz6pGMNv2cJX9IC2OrzKmB9RKyUdDAQEXGXpKXAQxHxoWGv\nd5PORI/RoLJA88pj1if9b9IZY/HzucAd+fauwG29isPMzJIiTTqSdgbmk76JZkfEohGe4xr+RI/R\noLJA88pj1if9b9KZKif8SRyjQWWB5pXHrE+8iLmZWe2c8M3MKuGEb2ZWCSd8M7NKOOGbmVXCCd+m\nrX1mzEDQ8599ZszoW5nMSvKwzAYN/WtSWcxs0jws08ysdk74ZmaVcMI3M6uEE76ZWSWc8M3MKuGE\nb2ZWCSd8M7NKOOE3iC9UMrPR+MIrX6xkZs3iC6/MzGrnhG9mVgknfDOzSjjhm5lVwgnfzKwSTvhm\nZpUonvAlnVM6BjOzGhRN+JLmAHNLxmBmVoviNXwzM+uPYglf0sERsYZRrgozM7PuKVnD363gsc3M\nqrNdr3YsaQFsNU2NgPURsTLX7m/K2ztOMLNkyZLNt4eGhhgaGupBpGZmdSgyeZqkeaREvwdwKnBK\nRNw17DmePM3MbOKm1+RpEXFdRKzMd3cpEYOZWW08PbJr+GbWLNOrhm9mZv3nhG9mVonqE76XBTSz\nWlTfhm9m1jBuwzczq50TvplZJZzwzcwq4YRvZlYJJ3wzs0o44ZuZVcIJ38ysEk74ZmaVcMI3M6uE\nE76ZWSWc8M3MKuGEb2ZWCSd8M7NKOOGbmVXCCd/MrBJO+GZmlXDCNzOrhBO+mVklnPDNzCrhhG9m\nVoliCV/S0vx7QakYJuuWW24pHULXNKks0KzyNKks0KzyDGpZStbwF0q6B1hXMIZJGdQ3eyRNKgs0\nqzxNKgs0qzyDWpbtCh77lIhYWfD4ZmZVKVnD313SHEnnFIzBzKwaioiyAaS2/Bsj4qZh28sGZmY2\noCJCI23vWZNO7oxtT9oC1kfEyvzY+tyksx6YDWyV8DsFbGZmk9OzhB8Ry0d5eB2wNt/eA1jVqzjM\nzCwp0mkbETdJmicJ4KGIuKtEHGZmNSnehm9mZv3hK237SNJekp5XOo6aSDpU0sGl45gKSTtK2rt0\nHDY2SdM6p07r4JpAud1K0puAvwNWS/qIpN8oG9kWkrZpxdkEknbNv18MfBg4X9JXJR1QNrJJeyvw\nF6WDmKxWEpS0k6QjSsfTSxHxJICkSyUdWDqe4Zzweyy2tJn9MXBORBwIPAR8VtLO5SLb8mUEnAF8\nR9KhJePphlym90g6E/gT4NyIOBE4G/i0pA8WDXCc2ioKTwOuzJt2kPSyfNZS8qLJcZOkiHhS0guA\na4FLJd0r6Q2lY+u2ti+21wN7A5+R9ElJu5SNbAsn/D6Q9BrS33pHgIj4S+BxYM+CMSkiQtIs4AXA\nZcD7JF0naa9ScXXB04EvAk8C+wNvkrRLRHwtIg4HLika3Ti1VRTeD1wNvAy4HTgHeENE/KpUbBPR\nVo4zgb+PiEOBDwIfkvTccpF1X/5i25FUuTgjIg4D7gbulnT6dGjuKR5A0+Wa2r7Ao8ACSadK+jNg\nQ0TcV6oppe2DeB7w3Yj4m4h4A/AgcJukPx3EZp6I+GVE3Ap8mpRYfgX8iaQT8uObSsY3EZKeCfwD\n6UzlNcB9wFnA4pJxjZek3XJz4fbATOB+gIj4BPCvpC/nxsifl11IZ/CP5c0fY8sX9psKhbaZE36P\nRXIlqUlnHelN/13gc63H+x3TsES+Enh6W/PST4DTSWcfO/Y7tqloO6V+J/A24BnAx0k14yNy4hkI\n+T16FnBHRNwbEb8Evgm8tMT/zETlJqezgD8EtiX9v79a0u9IeivwSETcXzLGHngeW77Y3iZpiNSH\ntA64HHh1udASJ/weaGt/PUjSX0t6EDg5Ij4CnEu6qvhISceWiK+VMCT9LfBM4HBSu/dHgLkR8U/A\nkcCvS8Q3WfmU+jnAm4FbI+KLwG8BOwEfjIhfFA1wHNq+jM8inaF8SdLv5W2rgC8XCWzidgC+DxxI\nas7Zk5QQjyHVds8rF1rPvBQ4HvhbYHdSxe4J0kWm7wb+qlxoicfh95Cka4AVwD3AZ4AHIuL3JO1J\n+mdYGRE/KxTb04FFwFeBrwN/AHwPeC4pSf57RHyqRGxTIekM4AUR8c58/w3AkRGxqGxk45dHd3wY\neAtwAanW+BvA2yPieyVjmyhJ+wLHAT8nNeGsi4ibS8bUK7nv6yPA9yPiXEnbkjpvXwnsmSt8RTnh\n90ge8XIh8Nq2oVrXA6dHxI+LBpe1DVv8h4j4cN52BKkp4eZW3INE0m7AEmA5qT9iEbAqIlaXjGs8\nJD0rIh6VtIh05rUKOIx0Rng26Sxx2r8nuePyMFLz4ItI/SgLSBWJR4CDI2Kgzh7Hkgc//DoifiDp\nfODOiPhS6biGG4ihXQPqh6RTud+VtJZ0GrtTyWQvaZvc7HEy8EtSR/IJwAckzYyIH0fE10vFN1WS\nnhYRGyXdCXyAlHC2HZBkPx84UdKlwIeAlwBvBz4LHAu8f0CS/eHA14Cfka47mQl8B7gDuIpUw29a\nsn8l8D7gfkmHkMr7Pkn7R8SDZaPbmmv4XSRp24j4taTFwAzgU8ApwCbSB3hxRHytlXgLxLd9RPxC\n0p+Skv2RwG8C+wE3RMTp/Y5pqtq+xF5PGslyJPBHpL/546Ra17RPMJI+BcwFvgL8J2nM+g7AUcCM\niHhHwfDGLfehvIXUdv9fwPsGZQjpRChd+TwX2Bm4htRc9Z/AbqQ+sZ0i4rPlIhyZE36XSdoJOB84\ngDRCYXdS5/hjpYYE5rbEO4DVpAR/KfAL4C7SP+pzSP0LT7TG55eIc7IkPYPUmflO4PXAe4F3RcTl\nRQOboNwUsC1pVNFxpGadNaS+nidKxjYRuX/oZaQOzANJ783HS/VX9YKkFcAs4AvAPFKy34Y0Qucr\nwDUR8fh0+zx5lE6XSLpI0oyI+ClwC2ks7l8CrwJ+GhGbCo5rfx6p5vEDUm1kKenqze+TLlLaqZVQ\nptM/5wS8A/hmRKzNF7W9itR2PGhuAeYAvwNcx5Zx6wOT7AEi4vGIuJ00ymg5cBDpy7hJPgPsSqo0\nHUpK/peRLvjbPiIeh+n3eXINvwvy+O+/Jp3ifTgiLs+1nDeQhqGtjIgbCsd4BnAaKRmeTRoHfbGk\n+RFxdcnYJkPSdq2mAkm/SRr6dyXwbeBdwKMRcUXBEMdFaTK9nSPiW/n+TqQa42pSP5AGoe2+k1zJ\neTbpQsNpPyx2IiT9Nukz/jPgGRFxfuGQxuSE30WSriRdXLENcFZEXJeT0YPT4Zs+X216HCnBzB/2\n2LQ69RyLpNcBLwQ2kk6rF5Caqx4ljQZ5/XRPlJJeQRoS+y22nH3NJtUY9wLeGBE/KhehjaTtTH07\n0oiws4BXR8TaUv1z4+VROlPUNifNW4GHI2J2Hvv9d5L+kDSUbrok0n8hdSLvlc9AnmjFNo1iHFP+\nwN1C6pw9Bvh5RFygNEHXL0jLZ07bD12bB4E/J0298TzSBWJfICX7B4FpMXzXttb2WXkCeK+kXwEL\nJT0QEQ8VDG1MruF3QU5AnwO+mKdRaK3p+5qIOLlocCOQ9Deky/Q/MSCJcStt49VfR1oi88WkOUz+\nBfhWRKwddQfTSFsH51GkseufBz43CCOLaiXpGRHxyzwi6emkSsZlwJkR8UDR4MbgGn4X5Br+5cCc\n/E/wLNJokb+ALUMHS8Y4zCeAw6dZTOMi6TTgOElfiYj/3bb9TaSLrN5YLLhJyJ17t0v6LmlCsTeT\nrna+qGhgNqI84m1/SY+RmnOuiIgvS3pXRPxH2ejG5hp+F+Qa/m6kUTl7Aj8ljcw5q2hgDaM0ne5V\npPH225M6A08mfeh+WDK2bmhyB2dT5KuIX0kacr0TaSbT/x6UkVRO+F0maVZE3N92EdZAdYZOZ5Ku\nI13I821J5wCvyA/9FmmO+HvLRWe1yKPybiI1If4Y+MdB+d/zOPxJ0JZpeLeX9MJ84Q8AsWXK1yfz\nfSf7LlCaG/4J4PmSXkWa7G1JpNWsvkCq8Zv1jKSX5ZtHAldFxLnAPYOS7MEJf1La2r6vJs1muF7S\nWcOe40TfRRHxc1Kt6nLS/D9/nmv6B5A+gN8uGZ81W54McanScoXfiYhlABHxhfz4QCwW5CadCWob\nhvk20rS7p+XJk/4eeHfrH8B6J49smUG60O11wIURcWfZqKyp2j7z7we+RLpC/dmkfrp/LxvdxLiG\nP0H5jRdwCGm9SiLin0nzYL+8ZGwVeZJ00dXhpFNrJ3vrpbdIupE0fHYNaYnJ95BmnB0oruFPkqSX\nkC5i+kdS2/K5wPkRcVfRwCrhznDrh1y5O5w0N9YDpNWsro+IVSXjmiyPwx+nttO67YFnR8Q3Jd1K\nmhHw58BNEXGXE1F/+G9sfbI7cG/rClpJNwCvkbRmIK9j8edmYiRdRprvZDtgYUTcUzgkM+uitjUW\njiJd57EP6QroS0kXxe0XEbcMYuXObfjj0OqBl3QkqXJ5DGmiq+slXS9pB0k+WzJrgLaa+5mkVbv+\nD2na6q+Trve4JT9voJI9OOGPS9sb+0fAhrzto6SO29sj4rFo4Ko+ZrWRtGv+/SrSYjQbSCvCnUma\nf2pgxtyPxE0646S0OPY84E3ArcA/tXfQDuLpnZltkafuWAF8NP9+EXA0aXK0HwCHRcT7y0U4dU74\nE5DHfx9Imq/8SNIl1R53b9YAkt4MLCMtUfgD4Kv5562k+XPeHREDfYGfE/4kSNqDNH/Lv0ahdWrN\nrPsknUianO/zwG+TJkP8MfCxiHho0M/k3YY/DsMvm46I9RHxZSd7s8b5Ommt2oWkRXbWkhakeRgG\ns6O2nWv449Ra9KB0HGbWXZJmkpL8f0TEo3nbScD9EXGbpD1z7X66rWsxYU74o8gTcx0CrG3NiCdp\nGfCBiHiwaHBmNmV5FtafAf9GWg/5k8AzSVNv7w/8VUR8uliAXeYmnWEkPa3t7idJvfQ3S7owD9n6\nCTAk6UUl4jOzrnoSOB24mfTZ3g9YTRqCuY40UVpjuIY/TD6VezHwK9Ii3xdImgFcQlp39EWkqRR2\njIiHy0VqZt2Q17fYn9RJexTwXeCjEbGhaGA94ITfJnfOvoQ0K95xpHk0/ldrURNJz4+I7xUM0cx6\nJA+7filpvYUXAidHxCNlo+ouJ/wR5AnSDiJdTv0M4EHgao/KMWs+STsDz4mI7w76MMzhnPCzttkw\nX0VK9D8E/i/pm34IuHVQp0Q1MwN32m7W9i1+FinZfwrYATgCWOZkb2aDzgm/jaRXAM+MiI9HxOPA\nY8DBpOFaZmYDzQm/TUTcBnxD0nGSdgdeTfoCaFTHjZnVqfo2/OGdMpIOAxaRpkWdCbw3Iu5uwlV2\nZla3qhN+W0ftvqS57k8EzomI1bmGv7G1aHmTeurNrE5VJ/wWSZ8G/hn4EWmFm5uB/xkRjzvZm1lT\nVN+GL+lg0pjbj0bE9RHxHOBpwE4w+LPjmZm1VJnw26c7jog7gdslzZG0S27D3zsi1peL0Mys+6ps\n0pF0IPBGYCPwZdJUCq8F/ps0S97FEXGDO2rNrElqTfgC3glcAFwKLCHNh/1s4L8i4kflojMz643t\nSgdQ0Ofz722AK4EvAT+PiLvLhWRm1jtV1fAlHUpaumyviPj9tu1HAH9GWuHm3FLxmZn1UjUJX9K2\npKGXbwe+R1qncj7wnYhYk5+zTUQ86aGYZtZENY3SeQvw2Yj4N9LUxytJix58StL/AGh10DrZm1kT\nVZHwcyft84FnSnoh8D7gqog4E/hj0rq1ZmaNVlOTziuBK0hrVF4ZEZ/L2/8fcElEfMFNOWbWZNUk\n/BZJO5LWqz0ROAB4QUTMLxuVmVnvVdGkM8yjwCzgcNKMmGfA1lffmpk1UXU1/BZJ20bEr/NtN+WY\nWeNVm/DNzGpTY5OOmVmVnPDNzCrhhG9mVgknfDOzSjjhm5lVwgnfzKwSTvhmZpX4/yUYEqlIrH8H\nAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f45dc328ed0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#Pooling\n",
    "\n",
    "str1='''Stoch, no drop\t0.429\t2.96\tStoch pool does not prevent overfitting without dropout :(. Good start,bad finish\n",
    "Avg\t0.435\t2.56\t\n",
    "Stoch\t0.438\t2.54\tUnderfitting, may be try without Dropout\n",
    "Max\t0.471\t2.36\t290K iters stopped\n",
    "No\t0.472\t2.35\tStrided conv2,conv3,conv4\n",
    "Max+Avg\t0.483\t2.29\tElement-wise sum'''\n",
    "names=[]\n",
    "acces=[]\n",
    "losses=[]\n",
    "for name_acc_loss_comm in str1.splitlines():\n",
    "    try:\n",
    "        name,acc,loss,comm = name_acc_loss_comm.strip().split(\"\\t\")\n",
    "    except:\n",
    "        name,acc,loss = name_acc_loss_comm.strip().split(\"\\t\")    \n",
    "    names.append(name)\n",
    "    acces.append(float(acc))\n",
    "    losses.append(float(loss))\n",
    "#Non-linearities\n",
    "accn=100.*(np.asarray(acces[:]) - acces[3])\n",
    "\n",
    "plt.rc('text', usetex=True)\n",
    "fig, ax = plt.subplots()\n",
    "rects1=ax.bar(2*np.asarray(range(len(acces[0:3])))+1, accn[0:3], 1, color='r')\n",
    "rects1=ax.bar(2*np.asarray(range(len(acces[3:])))+7, accn[3:], 1, color='g')\n",
    "#rects1=ax.bar(2*np.asarray(range(1))+(len(names)*2-1), accn[-1], 1, color='b')\n",
    "\n",
    "#rects1=ax.bar(2*5+1, accn[5], 1, color='r')\n",
    "\n",
    "# add some text for labels, title and axes ticks\n",
    "ax.set_ylabel('Top-1 acc [\\%]')\n",
    "ax.set_xticks( 2*np.asarray(range(len(names)))+1.5)\n",
    "ax.set_xticklabels( names[:], rotation=60 ) ;\n",
    "ax.set_xlim([0,2*len(names)+1])\n",
    "ax.spines['right'].set_visible(False)\n",
    "ax.spines['top'].set_visible(False)\n",
    "ax.xaxis.set_ticks_position('bottom')\n",
    "ax.yaxis.set_ticks_position('left')\n",
    "plt.suptitle('Poolings, gain over MaxPool 3x3/2')\n",
    "plt.savefig('../graphs/pool.pdf', format='pdf', dpi=1000,bbox_inches='tight')\n",
    "#ax.legend(names)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXwAAAFbCAYAAADWTjVEAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nRK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f45dcb26c10>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#Pooling\n",
    "\n",
    "str1='''MaxPool 3x3/2\t0.471\t2.36\tdefault alexnet\n",
    "MaxPool 2x2/2\t0.484\t2.29\tLeads to larger feature map, Pool5=4x4 instead of 3x3\n",
    "MaxPool 3x3/2 pad1\t0.488\t2.25\tLeads to even larger feature map, Pool5=5x5 instead of 3x3'''\n",
    "names=[]\n",
    "acces=[]\n",
    "losses=[]\n",
    "for name_acc_loss_comm in str1.splitlines():\n",
    "    try:\n",
    "        name,acc,loss,comm = name_acc_loss_comm.strip().split(\"\\t\")\n",
    "    except:\n",
    "        name,acc,loss = name_acc_loss_comm.strip().split(\"\\t\")    \n",
    "    names.append(name)\n",
    "    acces.append(float(acc))\n",
    "    losses.append(float(loss))\n",
    "#Non-linearities\n",
    "accn=100.*(np.asarray(acces[:]) - acces[0])\n",
    "\n",
    "plt.rc('text', usetex=True)\n",
    "fig, ax = plt.subplots()\n",
    "#rects1=ax.bar(2*np.asarray(range(len(acces[0:3])))+1, accn[0:3], 1, color='r')\n",
    "rects1=ax.bar(2*np.asarray(range(len(acces[:])))+1, accn[:], 1, color='g')\n",
    "#rects1=ax.bar(2*np.asarray(range(1))+(len(names)*2-1), accn[-1], 1, color='b')\n",
    "\n",
    "#rects1=ax.bar(2*5+1, accn[5], 1, color='r')\n",
    "\n",
    "# add some text for labels, title and axes ticks\n",
    "ax.set_ylabel('Top-1 acc [\\%]')\n",
    "ax.set_xticks( 2*np.asarray(range(len(names)))+1.5)\n",
    "ax.set_xticklabels( names[:], rotation=60 ) ;\n",
    "ax.set_xlim([0,2*len(names)+1])\n",
    "ax.spines['right'].set_visible(False)\n",
    "ax.spines['top'].set_visible(False)\n",
    "ax.xaxis.set_ticks_position('bottom')\n",
    "ax.yaxis.set_ticks_position('left')\n",
    "plt.suptitle('Poolings, gain over MaxPool 3x3/2')\n",
    "plt.savefig('../graphs/pool-stride.pdf', format='pdf', dpi=1000,bbox_inches='tight')\n",
    "#ax.legend(names)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXwAAAFaCAYAAAAdEubhAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f45dc13d6d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#Colorspace\n",
    "\n",
    "str1='''OpenCV RGB2Gray\t0.413\t2.70\tRGB->Grayscale Gray = 0.299 R + 0.587 G + 0.114 B\n",
    "Learned RGB2Gray\t0.419\t2.66\tRGB->conv1x1x1\n",
    "Histeq(RGB)\t0.448\t2.48\tRGB -> HiestEq\n",
    "HSV\t0.451\t2.46\tRGB->HSV->BatchNorm(HSV)\n",
    "YCrCb\t0.458\t2.42\tRGB->YCrCb->BatchNorm(YCrCb)\n",
    "CLAHE\t0.467\t2.38\tRGB -> LAB -> CLAHE(L)->RGB->BatchNorm(RGB)\n",
    "BN(RGB)\t0.469\t2.38\tInput = BatchNorm(Img)\n",
    "RGB\t0.471\t2.36\tdefault, no changes. Input = 0.04 * (Img - [104, 117,124])'''\n",
    "names=[]\n",
    "acces=[]\n",
    "losses=[]\n",
    "for name_acc_loss_comm in str1.splitlines():\n",
    "    try:\n",
    "        name,acc,loss,comm = name_acc_loss_comm.strip().split(\"\\t\")\n",
    "    except:\n",
    "        name,acc,loss = name_acc_loss_comm.strip().split(\"\\t\")    \n",
    "    names.append(name)\n",
    "    acces.append(float(acc))\n",
    "    losses.append(float(loss))\n",
    "#Non-linearities\n",
    "accn=100.*(np.asarray(acces[:]) - acces[7])\n",
    "\n",
    "plt.rc('text', usetex=True)\n",
    "fig, ax = plt.subplots()\n",
    "rects1=ax.bar(2*np.asarray(range(len(acces[0:7])))+1, accn[0:7], 1, color='r')\n",
    "#rects1=ax.bar(2*np.asarray(range(len(acces[8:])))+17, accn[8:], 1, color='g')\n",
    "#rects1=ax.bar(2*np.asarray(range(1))+(len(names)*2-1), accn[-1], 1, color='b')\n",
    "\n",
    "#rects1=ax.bar(2*5+1, accn[5], 1, color='r')\n",
    "\n",
    "# add some text for labels, title and axes ticks\n",
    "ax.set_ylabel('Top-1 acc [\\%]')\n",
    "ax.set_xticks( 2*np.asarray(range(len(names)))+1.5)\n",
    "ax.set_xticklabels( names[:], rotation=60 ) ;\n",
    "ax.set_xlim([0,2*len(names)+1])\n",
    "ax.spines['right'].set_visible(False)\n",
    "ax.spines['top'].set_visible(False)\n",
    "ax.xaxis.set_ticks_position('bottom')\n",
    "ax.yaxis.set_ticks_position('left')\n",
    "plt.suptitle('Pre-processings and colorspace, gain over RGB')\n",
    "plt.savefig('../graphs/colorspace.pdf', format='pdf', dpi=1000,bbox_inches='tight')\n",
    "#ax.legend(names)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYMAAAEfCAYAAACjwKoqAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nHbLmAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f45ddccda10>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#Learned colorspace\n",
    "str1='''A\t0.463\t2.40\tRGB -> conv1x1x10 tanh -> conv1x1x3 tanh\n",
    "RGB\t0.471\t2.36\tdefault, no changes. Input = 0.04 * (Img - [104, 117,124])\n",
    "B\t0.480\t2.32\tRGB -> conv1x1x3 vlrelu -> conv1x1x3 vlrelu\n",
    "C\t0.482\t2.30\tRGB -> conv1x1x10 vlrelu -> conv1x1x3 -> sum(RGB) ->vlrelu\n",
    "D\t0.482\t0.482\tRGB and log (RGB) -> conv1x1x10 vlrelu -> conv1x1x3 vlrelu\n",
    "E\t0.483\t2.30\tRGB -> conv1x1x16 vlrelu -> conv1x1x3 vlrelu\n",
    "F\t0.485\t2.28\tRGB -> conv1x1x10 vlrelu -> conv1x1x3 vlrelu'''\n",
    "basic_idx=1\n",
    "names=[]\n",
    "acces=[]\n",
    "losses=[]\n",
    "for name_acc_loss_comm in str1.splitlines():\n",
    "    try:\n",
    "        name,acc,loss,comm = name_acc_loss_comm.strip().split(\"\\t\")\n",
    "    except:\n",
    "        name,acc,loss = name_acc_loss_comm.strip().split(\"\\t\")    \n",
    "    names.append(name)\n",
    "    acces.append(float(acc))\n",
    "    losses.append(float(loss))\n",
    "#Non-linearities\n",
    "accn=100.*(np.asarray(acces[:]) - acces[basic_idx])\n",
    "\n",
    "plt.rc('text', usetex=True)\n",
    "fig, ax = plt.subplots()\n",
    "rects1=ax.bar(2*np.asarray(range(len(acces[0:basic_idx])))+1, accn[0:basic_idx], 1, color='r')\n",
    "rects1=ax.bar(2*np.asarray(range(len(acces[basic_idx:])))+2*basic_idx+1, accn[basic_idx:], 1, color='g')\n",
    "#rects1=ax.bar(2*np.asarray(range(1))+(len(names)*2-1), accn[-1], 1, color='b')\n",
    "\n",
    "#rects1=ax.bar(2*5+1, accn[5], 1, color='r')\n",
    "\n",
    "# add some text for labels, title and axes ticks\n",
    "ax.set_ylabel('Top-1 acc [\\%]')\n",
    "ax.set_xticks( 2*np.asarray(range(len(names)))+1.5)\n",
    "ax.set_xticklabels( names[:], rotation=60 ) ;\n",
    "ax.set_xlim([0,2*len(names)+1])\n",
    "ax.spines['right'].set_visible(False)\n",
    "ax.spines['top'].set_visible(False)\n",
    "ax.xaxis.set_ticks_position('bottom')\n",
    "ax.yaxis.set_ticks_position('left')\n",
    "plt.suptitle('Learned colorspace transform, gain over RGB')\n",
    "plt.savefig('../graphs/learned-colorspace.pdf', format='pdf', dpi=1000,bbox_inches='tight')\n",
    "#ax.legend(names)\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXsAAAEgCAYAAABVd9M/AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nIOwBIAcIewDIAcIeAHLg/wFsgdyk/tpacwAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f45ddc0e610>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#Network width\n",
    "str1='''4sqrt(2)x wider\t0.565\t1.96\t5.65\n",
    "4x wider\t0.563\t1.92\t4\n",
    "2sqrt(2)x wider\t0.552\t1.94\t2.82\n",
    "2 wider\t0.533\t2.04\t2\n",
    "sqrt(2) wider\t0.506\t2.17\t1.41\n",
    "Default\t0.471\t2.36\t1\n",
    "sqrt(2)x narrower\t0.460\t2.41\t0.707\n",
    "2x narrower\t0.416\t2.68\t0.5\n",
    "4x narrower\t0.256\t3.33\t0.25'''\n",
    "basic_idx=1\n",
    "names=[]\n",
    "acces=[]\n",
    "losses=[]\n",
    "xt=[]\n",
    "for name_acc_loss_comm in str1.splitlines():\n",
    "    try:\n",
    "        name,acc,loss,comm = name_acc_loss_comm.strip().split(\"\\t\")\n",
    "    except:\n",
    "        name,acc,loss = name_acc_loss_comm.strip().split(\"\\t\")    \n",
    "    names.append(name)\n",
    "    acces.append(float(acc))\n",
    "    losses.append(float(loss))\n",
    "    xt.append(float(comm))\n",
    "#Non-linearities\n",
    "accn=100.*np.asarray(acces[:])# - acces[basic_idx])\n",
    "xtarr=np.asarray(xt)\n",
    "plt.rc('text', usetex=True)\n",
    "fig, ax = plt.subplots()\n",
    "rects1=ax.plot(xtarr,accn, 'r.-')\n",
    "#rects1=ax.bar(2*np.asarray(range(len(acces[basic_idx:])))+2*basic_idx+1, accn[basic_idx:], 1, color='g')\n",
    "#rects1=ax.bar(2*np.asarray(range(1))+(len(names)*2-1), accn[-1], 1, color='b')\n",
    "\n",
    "#rects1=ax.bar(2*5+1, accn[5], 1, color='r')\n",
    "\n",
    "# add some text for labels, title and axes ticks\n",
    "ax.set_ylabel('Top-1 acc [\\%]')\n",
    "ax.set_xlabel('Network width, compated to caffenet')\n",
    "\n",
    "#ax.set_xticks( 2*np.asarray(range(len(names)))+1.5)\n",
    "#ax.set_xticklabels( names[:], rotation=60 ) ;\n",
    "#ax.set_xlim([0,2*len(names)+1])\n",
    "ax.set_ylim([0,60])\n",
    "\n",
    "ax.spines['right'].set_visible(False)\n",
    "ax.spines['top'].set_visible(False)\n",
    "ax.xaxis.set_ticks_position('bottom')\n",
    "ax.yaxis.set_ticks_position('left')\n",
    "plt.suptitle('Influence of the network width')\n",
    "plt.savefig('../graphs/width.pdf', format='pdf', dpi=1000,bbox_inches='tight')\n",
    "#ax.legend(names)\n",
    "\n",
    "\n"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}