system_demo.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "e5b3cc38",
   "metadata": {},
   "source": [
    "# Differential Coverage-Directed DL Framework Testing Demo\n",
    "\n",
    "## FYP21-22 SC3\n",
    "\n",
    "\n",
    "## System Overview\n",
    "\n",
    "1. Coverage GA\n",
    "\n",
    "1. Detection GA\n",
    "\n",
    "1. Localization\n",
    "    1. Inconsistency Localization\n",
    "    1. NaN Inconsistency Localization"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "933b4dab",
   "metadata": {},
   "source": [
    "## 1. Coverage GA"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "3d6b81e8",
   "metadata": {},
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "import os\n",
    "os.environ['KERAS_BACKEND'] = 'tensorflow'\n",
    "os.environ['CUDA_VISIBLE_DEVICES'] = ''\n",
    "import keras\n",
    "from coverage_GA.COV_GA import Cov_GA, cov_ga_main\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "db22c8d8",
   "metadata": {},
   "outputs": [],
   "source": [
    "a = ['Conv2D', 'Flatten', 'Dense']\n",
    "sl = [[0,0,0,0], [], [0]]\n",
    "\n",
    "(x_train, y_train), (x_test, y_test) = keras.datasets.cifar10.load_data()\n",
    "x = x_test[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "24ac16fb",
   "metadata": {},
   "outputs": [],
   "source": [
    "ga = Cov_GA(a, sl, x)\n",
    "ga.initPopulation(5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "575a2f17",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Continuing from the previous populations...\n",
      "Average fitness value: 3900.0\n",
      "Time taken: 96.04044389724731\n",
      "\n",
      "Average fitness value: 4149.4\n",
      "Time taken: 97.49125742912292\n",
      "\n",
      "Average fitness value: 4427.8\n",
      "Time taken: 101.14002990722656\n",
      "\n",
      "WARNING:tensorflow:Large dropout rate: 0.6 (>0.5). In TensorFlow 2.x, dropout() uses dropout rate instead of keep_prob. Please ensure that this is intended.\n",
      "WARNING:tensorflow:Large dropout rate: 0.6 (>0.5). In TensorFlow 2.x, dropout() uses dropout rate instead of keep_prob. Please ensure that this is intended.\n",
      "Average fitness value: 3857.8\n",
      "Time taken: 79.34137034416199\n",
      "\n",
      "WARNING:tensorflow:Large dropout rate: 0.6 (>0.5). In TensorFlow 2.x, dropout() uses dropout rate instead of keep_prob. Please ensure that this is intended.\n",
      "WARNING:tensorflow:Large dropout rate: 0.6 (>0.5). In TensorFlow 2.x, dropout() uses dropout rate instead of keep_prob. Please ensure that this is intended.\n",
      "WARNING:tensorflow:Large dropout rate: 0.6 (>0.5). In TensorFlow 2.x, dropout() uses dropout rate instead of keep_prob. Please ensure that this is intended.\n",
      "Average fitness value: 5229.2\n",
      "Time taken: 104.91035866737366\n",
      "\n",
      "Average fitness value: 4570.4\n",
      "Time taken: 85.846999168396\n",
      "\n",
      "Average fitness value: 5763.2\n",
      "Time taken: 113.80778861045837\n",
      "\n",
      "Average fitness value: 5882.8\n",
      "Time taken: 106.62946724891663\n",
      "\n",
      "Average fitness value: 4787.6\n",
      "Time taken: 108.51917600631714\n",
      "\n",
      "Average fitness value: 3583.2\n",
      "Time taken: 85.07545328140259\n",
      "\n",
      "\n",
      "Total time taken: 978.8074171543121\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<coverage_GA.COV_GA.Cov_GA at 0x7f2e901d55c0>"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "cov_ga_main(a, sl, x, m=2, r=0.3, n=5, maxIter=10, ga=ga)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "36323ebf",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[[3958, 3867, 3888, 3888, 3899],\n",
       " [3886, 3958, 4259, 4151, 4493],\n",
       " [4246, 4429, 4377, 4528, 4559],\n",
       " [4515, 4546, 4496, 5732, 0],\n",
       " [4546, 5616, 5674, 5610, 4700],\n",
       " [5606, 5651, 5859, 0, 5736],\n",
       " [5723, 5846, 5807, 5603, 5837],\n",
       " [5837, 5846, 5884, 5864, 5983],\n",
       " [5884, 5983, 0, 6047, 6024],\n",
       " [6024, 6034, 0, 5858, 0]]"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ga.fit_hist"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "f5d40bb0",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<matplotlib.lines.Line2D at 0x7f2c191fd978>]"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(np.max(np.nan_to_num(ga.fit_hist, -1), axis=1))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ac934dd8",
   "metadata": {},
   "source": [
    "## 2. Detection GA"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "2678b628",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import os\n",
    "import cv2\n",
    "import redis\n",
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "from detection_GA.GA_keras import GA, ga_main"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "acd332ba",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Configured\n"
     ]
    }
   ],
   "source": [
    "import configRedis\n",
    "configRedis.config()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "2fffdabc",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/data/yylaiai/anaconda3/envs/audee_test/lib/python3.6/site-packages/keras/engine/saving.py:341: UserWarning: No training configuration found in save file: the model was *not* compiled. Compile it manually.\n",
      "  warnings.warn('No training configuration found in save file: '\n",
      "/data/yylaiai/anaconda3/envs/audee_test/lib/python3.6/site-packages/ipykernel_launcher.py:5: UserWarning: Update your `Model` call to the Keras 2 API: `Model(inputs=Tensor(\"co..., outputs=Tensor(\"so...)`\n",
      "  \"\"\"\n"
     ]
    }
   ],
   "source": [
    "# load model\n",
    "\n",
    "model = keras.models.load_model('external_models/alexnet-cifar10_origin.h5')\n",
    "new_output = keras.layers.Softmax()(keras.layers.Dense(10, name=\"dense_3\")(model.layers[-2].output)) # without specifying activation \n",
    "new_model = keras.models.Model(input=model.input, output=new_output)\n",
    "new_model.layers[-2].set_weights(model.layers[-1].get_weights())\n",
    "model = new_model\n",
    "\n",
    "# model = keras.applications.ResNet50()\n",
    "# new_output = keras.layers.Softmax()(keras.layers.Dense(1000)(model.layers[-2].output)) # without specifying activation \n",
    "# new_model = keras.models.Model(input=model.input, output=new_output)\n",
    "# new_model.layers[-2].set_weights(model.layers[-1].get_weights())\n",
    "# model = new_model\n",
    "\n",
    "# model = keras.applications.MobileNetV2()\n",
    "# new_output = keras.layers.Softmax()(keras.layers.Dense(1000)(model.layers[-2].output)) # without specifying activation \n",
    "# new_model = keras.models.Model(input=model.input, output=new_output)\n",
    "# new_model.layers[-2].set_weights(model.layers[-1].get_weights())\n",
    "# model = new_model\n",
    "\n",
    "# model = keras.applications.VGG16()\n",
    "# new_output = keras.layers.Softmax()(keras.layers.Dense(1000)(model.layers[-2].output)) # without specifying activation \n",
    "# new_model = keras.models.Model(input=model.input, output=new_output)\n",
    "# new_model.layers[-2].set_weights(model.layers[-1].get_weights())\n",
    "# model = new_model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "080f12eb",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(10000, 224, 224, 3)\n"
     ]
    }
   ],
   "source": [
    "# load inputs\n",
    "(x_train, y_train), (x_test, y_test) = tf.keras.datasets.cifar10.load_data()\n",
    "\n",
    "# expand cifar10 images from 32x32 to 224x224\n",
    "x_test_224 = np.array([cv2.resize(i, (224, 224)) for i in x_test])\n",
    "print(x_test_224.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "ef6a7fdd",
   "metadata": {},
   "outputs": [],
   "source": [
    "mut_level = 'i+w'\n",
    "model = model\n",
    "x = x_test[0]/255\n",
    "input_scale = 255 # image scale\n",
    "init_noise = 0 # scale of Cauchy noise added in initialization\n",
    "r1 = 0.5 # crossover rate\n",
    "r2 = 0.3 # mutation rate\n",
    "r3 = 0.1 # mutation scale (larger scale can trigger NaN inconsistency easier, smaller scale can increase layer distance easier)\n",
    "m = 3 # top m chromosomes selected\n",
    "n = 6 # population size\n",
    "layer_idx = -2 # index of layer used in calculation of fitness values\n",
    "db_flag = 2\n",
    "maxIter = 5\n",
    "dynamicWeightMutDecay = np.inf\n",
    "\n",
    "r = redis.Redis(db=db_flag)\n",
    "r.flushdb()\n",
    "\n",
    "fit = ['inc', ['tensorflow', 'theano']]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "e1aaa4d2",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[[3072, 3072, 0.1],\n",
       " [3072, 5760, 0.1],\n",
       " [5760, 5760, 0.1],\n",
       " [5760, 6144, 0],\n",
       " [6144, 620800, 0.1],\n",
       " [620800, 620800, 0.1],\n",
       " [620800, 621824, 0],\n",
       " [621824, 1506944, 0.1],\n",
       " [1506944, 2834432, 0.1],\n",
       " [2834432, 3719424, 0.1],\n",
       " [3719424, 3719424, 0.1],\n",
       " [3719424, 3720448, 0],\n",
       " [3720448, 3720448, 0.1],\n",
       " [3720448, 4773120, 0.1],\n",
       " [4773120, 4773120, 0.1],\n",
       " [4773120, 21554432, 0.1],\n",
       " [21554432, 21554432, 0.1],\n",
       " [21554432, 21595402, 0.1],\n",
       " [21595402, 21595402, 0.1]]"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ga_inc = GA(fit, mut_level, model, x, input_scale, db_flag)\n",
    "ga_inc.initPopulation(init_noise, n)\n",
    "ga_inc.initDynamicWeightMutation(r3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "f384e4dd",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Continuing from the previous populations...\n",
      "\n",
      "Running at iteration 1:\n",
      "Average fitness value: 170.12336425781248\n",
      "Time taken: 102.31689882278442\n",
      "\n",
      "Running at iteration 2:\n",
      "Average fitness value: 3.63286621339577e+19\n",
      "Time taken: 101.36361455917358\n",
      "\n",
      "Running at iteration 3:\n",
      "Average fitness value: 2.2774815910001333e+21\n",
      "Time taken: 98.87646913528442\n",
      "\n",
      "Running at iteration 4:\n",
      "Average fitness value: 1.976032871780307e+22\n",
      "Time taken: 98.28308248519897\n",
      "\n",
      "Running at iteration 5:\n",
      "Average fitness value: 3.76303026569268e+23\n",
      "Time taken: 101.69381880760193\n",
      "\n",
      "Running at iteration 6:\n",
      "Average fitness value: 1.4042025912103168e+24\n",
      "Time taken: 102.27084541320801\n",
      "\n",
      "Running at iteration 7:\n",
      "Average fitness value: 1.5678057863383035e+25\n",
      "Time taken: 99.83339619636536\n",
      "\n",
      "Running at iteration 8:\n",
      "Average fitness value: 4.926622299376932e+25\n",
      "Time taken: 102.26488828659058\n",
      "\n",
      "Running at iteration 9:\n",
      "Average fitness value: 8.180348670274019e+25\n",
      "Time taken: 104.95323753356934\n",
      "\n",
      "Running at iteration 10:\n",
      "Average fitness value: 3.408564540622528e+26\n",
      "Time taken: 92.85399222373962\n",
      "\n",
      "\n",
      "Total time taken: 1004.7156031131744\n"
     ]
    }
   ],
   "source": [
    "ga_inc = ga_main(fit, mut_level, model, x, input_scale, init_noise, r1, r2, r3, m, n, layer_idx, db_flag, maxIter, dynamicWeightMutDecay, ga_inc)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "398a95dd",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<matplotlib.lines.Line2D at 0x7fe23c333710>]"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(np.max(np.nan_to_num(ga_inc.fit_hist, -1), axis=1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "312b89fb",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([9.74862389e+25, 9.98086914e+25, 1.45713606e+26, 3.65732910e+26,\n",
       "       2.64055283e+26, 1.07234199e+27])"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ga_inc.fit_hist[-1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "96aefd65",
   "metadata": {},
   "outputs": [],
   "source": [
    "mut_level = 'i+w'\n",
    "model = model\n",
    "x = x_test[0]/255\n",
    "input_scale = 255 # image scale\n",
    "init_noise = 0 # scale of Cauchy noise added in initialization\n",
    "r1 = 0.5 # crossover rate\n",
    "r2 = 0.3 # mutation rate\n",
    "r3 = 20 # mutation scale (larger scale can trigger NaN inconsistency easier, smaller scale can increase layer distance easier)\n",
    "m = 3 # top m chromosomes selected\n",
    "n = 6 # population size\n",
    "layer_idx = -2 # index of layer used in calculation of fitness values\n",
    "db_flag = 2\n",
    "maxIter = 10\n",
    "dynamicWeightMutDecay = np.inf\n",
    "\n",
    "r = redis.Redis(db=db_flag)\n",
    "r.flushdb()\n",
    "\n",
    "fit = ['inc', ['tensorflow', 'theano']]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "4d66496a",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[[3072, 3072, 20],\n",
       " [3072, 5760, 20],\n",
       " [5760, 5760, 20],\n",
       " [5760, 6144, 0],\n",
       " [6144, 620800, 20],\n",
       " [620800, 620800, 20],\n",
       " [620800, 621824, 0],\n",
       " [621824, 1506944, 20],\n",
       " [1506944, 2834432, 20],\n",
       " [2834432, 3719424, 20],\n",
       " [3719424, 3719424, 20],\n",
       " [3719424, 3720448, 0],\n",
       " [3720448, 3720448, 20],\n",
       " [3720448, 4773120, 20],\n",
       " [4773120, 4773120, 20],\n",
       " [4773120, 21554432, 20],\n",
       " [21554432, 21554432, 20],\n",
       " [21554432, 21595402, 20],\n",
       " [21595402, 21595402, 20]]"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ga_nan = GA(fit, mut_level, model, x, input_scale, db_flag)\n",
    "ga_nan.initPopulation(init_noise, n)\n",
    "ga_nan.initDynamicWeightMutation(r3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "id": "de1520e4",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Continuing from the previous populations...\n",
      "\n",
      "Running at iteration 1:\n",
      "Average fitness value: 170.12336425781248\n",
      "Time taken: 104.15072274208069\n",
      "\n",
      "Running at iteration 2:\n",
      "Average fitness value: nan\n",
      "Time taken: 98.91877341270447\n",
      "\n",
      "Running at iteration 3:\n",
      "Average fitness value: nan\n",
      "Time taken: 104.75501346588135\n",
      "\n",
      "Running at iteration 4:\n",
      "Average fitness value: nan\n",
      "Time taken: 104.70004725456238\n",
      "\n",
      "Running at iteration 5:\n",
      "Average fitness value: nan\n",
      "Time taken: 105.17480039596558\n",
      "\n",
      "Running at iteration 6:\n",
      "Average fitness value: nan\n",
      "Time taken: 99.77638792991638\n",
      "\n",
      "Running at iteration 7:\n",
      "Average fitness value: nan\n",
      "Time taken: 97.55729293823242\n",
      "\n",
      "Running at iteration 8:\n",
      "Average fitness value: nan\n",
      "Time taken: 105.62516832351685\n",
      "\n",
      "Running at iteration 9:\n",
      "Average fitness value: nan\n",
      "Time taken: 106.11227130889893\n",
      "\n",
      "Running at iteration 10:\n",
      "Average fitness value: nan\n",
      "Time taken: 95.303701877594\n",
      "\n",
      "\n",
      "Total time taken: 1022.0787642002106\n"
     ]
    }
   ],
   "source": [
    "ga_nan = ga_main(fit, mut_level, model, x, input_scale, init_noise, r1, r2, r3, m, n, layer_idx, db_flag, maxIter, dynamicWeightMutDecay, ga_nan)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "id": "704687c4",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[[[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [17, 18]]],\n",
       "  array([ 6.19607843e-01,  4.39215686e-01,  1.92156863e-01, ...,\n",
       "         -3.41296792e-02, -5.36657404e+01, -1.82542689e-02])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [17, 18]]],\n",
       "  array([ 0.61960784,  0.43921569,  1.        , ..., -0.03412968,\n",
       "         -0.02668719, -0.01825427])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [17, 18]]],\n",
       "  array([ 6.19607843e-01,  4.39215686e-01,  1.00000000e+00, ...,\n",
       "         -3.41296792e-02,  1.83795551e+01, -1.82542689e-02])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [17, 18]]],\n",
       "  array([ 6.19607843e-01,  4.39215686e-01,  1.92156863e-01, ...,\n",
       "         -3.41296792e-02, -5.36657404e+01, -1.82542689e-02])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [17, 18]]],\n",
       "  array([ 6.19607843e-01,  4.39215686e-01,  1.00000000e+00, ...,\n",
       "         -3.41296792e-02,  1.83795551e+01, -1.82542689e-02])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [17, 18]]],\n",
       "  array([ 6.19607843e-01,  4.39215686e-01,  0.00000000e+00, ...,\n",
       "         -3.41296792e-02, -5.36657404e+01, -1.82542689e-02])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [17, 18]]],\n",
       "  array([ 6.19607843e-01,  1.00000000e+00,  0.00000000e+00, ...,\n",
       "          1.51228273e+00,  1.83795551e+01, -1.82542689e-02])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [17, 18]]],\n",
       "  array([ 6.19607843e-01,  0.00000000e+00,  1.92156863e-01, ...,\n",
       "         -3.41296792e-02,  3.73097303e+01,  2.94679653e+02])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [17, 18]]],\n",
       "  array([ 6.19607843e-01,  4.39215686e-01,  0.00000000e+00, ...,\n",
       "         -3.41296792e-02, -5.36657404e+01, -1.82542689e-02])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [17, 18]]],\n",
       "  array([ 6.19607843e-01,  0.00000000e+00,  1.92156863e-01, ...,\n",
       "         -3.41296792e-02,  3.73097303e+01,  2.94679653e+02])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [15, 16, 17, 18]]],\n",
       "  array([ 6.19607843e-01,  4.39215686e-01,  0.00000000e+00, ...,\n",
       "         -3.41296792e-02,  3.73097303e+01,  2.94679653e+02])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [15, 16, 17, 18]]],\n",
       "  array([ 6.19607843e-01,  0.00000000e+00,  1.00000000e+00, ...,\n",
       "         -3.41296792e-02,  3.73097303e+01, -1.82542689e-02])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [15, 16, 17, 18]]],\n",
       "  array([ 6.19607843e-01,  4.39215686e-01,  1.92156863e-01, ...,\n",
       "         -3.41296792e-02,  3.73097303e+01, -1.82542689e-02])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [15, 16, 17, 18]]],\n",
       "  array([ 6.19607843e-01,  4.39215686e-01,  0.00000000e+00, ...,\n",
       "         -3.41296792e-02,  3.73097303e+01,  2.94679653e+02])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [15, 16, 17, 18]]],\n",
       "  array([ 6.19607843e-01,  4.39215686e-01,  1.92156863e-01, ...,\n",
       "         -3.41296792e-02,  3.73097303e+01, -1.82542689e-02])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [17, 18]]],\n",
       "  array([ 0.61960784,  0.43921569,  0.19215686, ..., -0.03412968,\n",
       "         -0.02668719, -0.01825427])],\n",
       " [[['nan', 'tensorflow', [15, 16, 17, 18]],\n",
       "   ['nan', 'theano', [15, 16, 17, 18]]],\n",
       "  array([ 6.19607843e-01,  0.00000000e+00,  1.92156863e-01, ...,\n",
       "         -3.41296792e-02,  3.73097303e+01, -1.82542689e-02])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [15, 16, 17, 18]]],\n",
       "  array([ 0.61960784,  0.43921569,  1.        , ..., -0.03412968,\n",
       "         -0.02668719, -0.01825427])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [17, 18]]],\n",
       "  array([ 0.61960784,  0.43921569,  0.19215686, ..., -0.03412968,\n",
       "         -0.02668719, -0.01825427])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [15, 16, 17, 18]]],\n",
       "  array([ 0.61960784,  0.43921569,  1.        , ..., -0.03412968,\n",
       "         -0.02668719, -0.01825427])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [15, 16, 17, 18]]],\n",
       "  array([ 0.61960784,  0.43921569,  0.19215686, ..., -0.03412968,\n",
       "         -0.02668719, -0.01825427])],\n",
       " [[['nan', 'tensorflow', [15, 16, 17, 18]],\n",
       "   ['nan', 'theano', [15, 16, 17, 18]]],\n",
       "  array([ 0.        ,  0.43921569,  1.        , ..., -0.03412968,\n",
       "         -0.02668719, -0.01825427])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [17, 18]]],\n",
       "  array([ 0.        ,  0.43921569,  0.        , ..., -0.03412968,\n",
       "         -0.02668719, -0.01825427])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [15, 16, 17, 18]]],\n",
       "  array([ 0.61960784,  0.43921569,  0.19215686, ..., -0.03412968,\n",
       "         -0.02668719, -0.01825427])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [17, 18]]],\n",
       "  array([ 0.        ,  0.43921569,  0.        , ..., -0.03412968,\n",
       "         -0.02668719, -0.01825427])],\n",
       " [[['nan', 'tensorflow', [15, 16, 17, 18]],\n",
       "   ['nan', 'theano', [15, 16, 17, 18]]],\n",
       "  array([ 0.61960784,  0.43921569,  0.        , ..., -0.03412968,\n",
       "         -0.02668719, -0.01825427])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [15, 16, 17, 18]]],\n",
       "  array([ 0.        ,  0.43921569,  0.        , ..., -0.03412968,\n",
       "         -0.02668719, -0.01825427])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [15, 16, 17, 18]]],\n",
       "  array([ 0.61960784,  1.        ,  0.19215686, ..., -0.03412968,\n",
       "         -0.02668719, -0.01825427])],\n",
       " [[['nan', 'tensorflow', [15, 16, 17, 18]],\n",
       "   ['nan', 'theano', [15, 16, 17, 18]]],\n",
       "  array([ 0.61960784,  0.43921569,  0.        , ..., -0.03412968,\n",
       "         -0.02668719, -0.01825427])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [15, 16, 17, 18]]],\n",
       "  array([ 0.61960784,  1.        ,  0.19215686, ..., -0.03412968,\n",
       "         -0.02668719, -0.01825427])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [15, 16, 17, 18]]],\n",
       "  array([ 1.        ,  0.43921569,  0.        , ..., -0.03412968,\n",
       "         -0.02668719, -0.01825427])],\n",
       " [[['nan', 'tensorflow', [18]], ['nan', 'theano', [15, 16, 17, 18]]],\n",
       "  array([ 0.        ,  0.43921569,  0.        , ..., -0.03412968,\n",
       "         -0.02668719, -0.01825427])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [17, 18]]],\n",
       "  array([ 0.61960784,  0.43921569,  0.        , ..., -0.03412968,\n",
       "         -0.02668719, -0.01825427])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [15, 16, 17, 18]]],\n",
       "  array([ 1.        ,  0.43921569,  0.        , ..., -0.03412968,\n",
       "         -0.02668719, -0.01825427])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [17, 18]]],\n",
       "  array([ 0.61960784,  0.43921569,  0.        , ..., -0.03412968,\n",
       "         -0.02668719, -0.01825427])],\n",
       " [[['nan', 'tensorflow', [18]], ['nan', 'theano', [18]]],\n",
       "  array([ 0.        ,  0.        ,  0.19215686, ..., -0.03412968,\n",
       "         -0.02668719, -0.01825427])],\n",
       " [[['nan', 'tensorflow', [17, 18]], ['nan', 'theano', [15, 16, 17, 18]]],\n",
       "  array([ 0.        ,  0.43921569,  0.7210153 , ..., -0.03412968,\n",
       "         -0.02668719, -0.01825427])]]"
      ]
     },
     "execution_count": 35,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ga_nan.F"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "ab722124",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "id": "55cc1ebb",
   "metadata": {},
   "source": [
    "## Localization"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "ba0a8a1d",
   "metadata": {},
   "outputs": [],
   "source": [
    "# extract model and input from detection GA\n",
    "def formatP(ga, idx):\n",
    "    formatted_c = ga.formatPopulations(ga.P)[idx]\n",
    "    if ga.mut_level == 'i':\n",
    "        x = formatted_c\n",
    "    elif ga.mut_level == 'w':\n",
    "        x = ga.input\n",
    "        ga.model.set_weights(formatted_c)\n",
    "    elif ga.mut_level == 'i+w':\n",
    "        x = formatted_c[0]\n",
    "        ga.model.set_weights(formatted_c[1])\n",
    "\n",
    "    return x, ga.model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "6aa621bb",
   "metadata": {},
   "outputs": [],
   "source": [
    "from localization.inc_localization import IncLocalizer\n",
    "from localization.nan_localization import NanLocalizer"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7688cc1d",
   "metadata": {},
   "source": [
    "### Inconsistency Localization"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "5ef8e543",
   "metadata": {},
   "outputs": [],
   "source": [
    "import importlib\n",
    "import localization.inc_localization\n",
    "importlib.reload(localization.inc_localization)\n",
    "\n",
    "from localization.inc_localization import IncLocalizer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "96dd77ae",
   "metadata": {},
   "outputs": [],
   "source": [
    "from localization.model_configs import get_configs\n",
    "\n",
    "inc_x, inc_model = formatP(ga_inc, 5)\n",
    "frameworks = ['tensorflow', 'theano']\n",
    "model_config = get_configs(inc_model)\n",
    "model_name = 'alexnet'\n",
    "\n",
    "incL = IncLocalizer(inc_model, frameworks, inc_x, model_name, model_config, db_flag)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "id": "9b2cbb41",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Distance of layer between TF and TH:\n",
      "\n",
      "0     InputLayer                     0.0\n",
      "1     Conv2D                         0.025303180019060772\n",
      "2     MaxPooling2D                   0.021129305164019268\n",
      "3     BatchNormalization             0.014197592933972677\n",
      "4     Conv2D                         0.004810093902051449\n",
      "5     MaxPooling2D                   0.0055399201810359955\n",
      "6     BatchNormalization             0.005648587312963273\n",
      "7     Conv2D                         0.001654188252157635\n",
      "8     Conv2D                         0.001463281335654082\n",
      "9     Conv2D                         0.0003547866363078356\n",
      "10    MaxPooling2D                   0.0010434419382363558\n",
      "11    BatchNormalization             0.0013569925213232636\n",
      "12    Flatten                        0.0013569925213232636\n",
      "13    Dense                          0.0015525349881500006\n",
      "14    Dropout                        0.0015525349881500006\n",
      "15    Dense                          0.00014380089123733342\n",
      "16    Dropout                        0.00014380089123733342\n",
      "17    Dense                          nan\n",
      "18    Softmax                        nan\n",
      "\n",
      "---------------------------------------------------------------\n",
      "\n",
      "Rate of change of inconsistencies between layers:\n",
      "\n",
      "1     Conv2D                         253031.80019060773\n",
      "2     MaxPooling2D                   -0.16495390526039924\n",
      "3     BatchNormalization             -0.43889911018343625\n",
      "4     Conv2D                         -0.809898404537753\n",
      "5     MaxPooling2D                   -0.7810552554110479\n",
      "6     BatchNormalization             -0.7767606686284086\n",
      "7     Conv2D                         -0.934621588548541\n",
      "8     Conv2D                         -0.9421663383343295\n",
      "9     Conv2D                         -0.9859746785380985\n",
      "10    MaxPooling2D                   -0.958758629812014\n",
      "11    BatchNormalization             -0.9463669326545422\n",
      "12    Flatten                        -0.9463669326545422\n",
      "13    Dense                          -0.9386389832867355\n",
      "14    Dropout                        -0.9386389832867355\n",
      "15    Dense                          -0.9943129550347254\n",
      "16    Dropout                        -0.9943129550347254\n",
      "17    Dense                          nan\n",
      "18    Softmax                        nan\n"
     ]
    }
   ],
   "source": [
    "# manually analyze rate of change of inconsistencies between layers\n",
    "\n",
    "layers_dist = incL.compute_all_layers_dist()\n",
    "\n",
    "print('Distance of layer between TF and TH:\\n')\n",
    "for i in range(len(model.layers)):\n",
    "    print('{:<5} {:<30} {}'.format(i, model.layers[i].__class__.__name__, layers_dist[i]))\n",
    "\n",
    "print()\n",
    "print('---------------------------------------------------------------')\n",
    "print()\n",
    "\n",
    "print('Rate of change of inconsistencies between layers:\\n')\n",
    "for i in range(1,len(model.layers)):\n",
    "    print('{:<5} {:<30} {}'.format(i, model.layers[i].__class__.__name__, incL.compute_dists_change(i)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "20112f1f",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Distance of layer between TF and TH:\n",
      "\n",
      "0     InputLayer                     0.0\n",
      "1     Conv2D                         0.004019102081656456\n",
      "2     MaxPooling2D                   0.0029788166284561157\n",
      "3     BatchNormalization             0.003173970617353916\n",
      "4     Conv2D                         0.002450015861541033\n",
      "5     MaxPooling2D                   0.0013138349685404035\n",
      "6     BatchNormalization             0.00153858649233977\n",
      "7     Conv2D                         0.0019705957836574977\n",
      "8     Conv2D                         0.0006761581257537559\n",
      "9     Conv2D                         0.00017674024113350443\n",
      "10    MaxPooling2D                   0.0005101709393784404\n",
      "11    BatchNormalization             7.358996663242579e-05\n",
      "12    Flatten                        7.358996663242579e-05\n",
      "13    Dense                          1.1955818990827538e-05\n",
      "14    Dropout                        1.1955818990827538e-05\n",
      "15    Dense                          1.7807902622735128e-05\n",
      "16    Dropout                        1.7807902622735128e-05\n",
      "17    Dense                          0.018568551540374754\n",
      "18    Softmax                        0.0\n"
     ]
    }
   ],
   "source": [
    "print('Distance of layer between TF and TH:\\n')\n",
    "for i in range(len(model.layers)):\n",
    "    print('{:<5} {:<30} {}'.format(i, model.layers[i].__class__.__name__, layers_dist[i]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "b8f3be04",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Change rate between layers:\n",
      "\n",
      "1     Conv2D                         40191.02081656456\n",
      "2     MaxPooling2D                   -0.258828850121316\n",
      "3     BatchNormalization             -0.21027344411461665\n",
      "4     Conv2D                         -0.39039744412869803\n",
      "5     MaxPooling2D                   -0.6730856170339948\n",
      "6     BatchNormalization             -0.6171661784904273\n",
      "7     Conv2D                         -0.509679845994381\n",
      "8     Conv2D                         -0.8317431888184517\n",
      "9     Conv2D                         -0.9560011570603529\n",
      "10    MaxPooling2D                   -0.8730417309178592\n",
      "11    BatchNormalization             -0.981665523371232\n",
      "12    Flatten                        -0.981665523371232\n",
      "13    Dense                          -0.9970004446788455\n",
      "14    Dropout                        -0.9970004446788455\n",
      "15    Dense                          -0.9955444134783701\n",
      "16    Dropout                        -0.9955444134783701\n",
      "17    Dense                          3.61998455492488\n",
      "18    Softmax                        -0.9999946145793203\n"
     ]
    }
   ],
   "source": [
    "print('Rate of change of inconsistencies between layers:\\n')\n",
    "for i in range(1,len(model.layers)):\n",
    "    print('{:<5} {:<30} {}'.format(i, model.layers[i].__class__.__name__, incL.compute_dists_change(i)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "4f9c6544",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Localizing layer 17: Dense...\n",
      "\n",
      "Parameter \"units\" is set to be \"300\"\n",
      "Maximizing inconsistency of the input...\n",
      "New rate of change of 17: 7.448445541877155e+21\n",
      "\n",
      "Time taken: 166.50839904800523\n",
      "\n",
      "\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "([],\n",
       " [],\n",
       " [('Parameter \"a\": 10 -> a_prime',\n",
       "   <keras.engine.sequential.Sequential at 0x7fe23c221a90>,\n",
       "   <keras.engine.sequential.Sequential at 0x7fe23c2214a8>)])"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t1 = 3\n",
    "t2 = 0.1\n",
    "\n",
    "\n",
    "'''\n",
    "X: localized inconsistent parameters and layer index\n",
    "Y: parameters and layer index which cause NaN or crash\n",
    "M: inconsistent models for replication\n",
    "'''\n",
    "X, Y, M = incL.main(t1, t2)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6151377c",
   "metadata": {},
   "source": [
    "### NaN Inconsistency Localization"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "id": "674c18f4",
   "metadata": {},
   "outputs": [],
   "source": [
    "nanL = NanLocalizer(ga_nan, 'alexnet', 2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "id": "a4c98e63",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "([],\n",
       " [[10, 'theano', 16],\n",
       "  [10, 'theano', 15],\n",
       "  [11, 'theano', 16],\n",
       "  [11, 'theano', 15],\n",
       "  [12, 'theano', 16],\n",
       "  [12, 'theano', 15],\n",
       "  [13, 'theano', 16],\n",
       "  [13, 'theano', 15],\n",
       "  [14, 'theano', 16],\n",
       "  [14, 'theano', 15],\n",
       "  [17, 'theano', 16],\n",
       "  [17, 'theano', 15],\n",
       "  [19, 'theano', 16],\n",
       "  [19, 'theano', 15],\n",
       "  [20, 'theano', 16],\n",
       "  [20, 'theano', 15],\n",
       "  [23, 'theano', 16],\n",
       "  [23, 'theano', 15],\n",
       "  [26, 'theano', 16],\n",
       "  [26, 'theano', 15],\n",
       "  [27, 'theano', 16],\n",
       "  [27, 'theano', 15],\n",
       "  [29, 'theano', 16],\n",
       "  [29, 'theano', 15],\n",
       "  [30, 'theano', 16],\n",
       "  [30, 'theano', 15],\n",
       "  [31, 'theano', 16],\n",
       "  [31, 'theano', 17],\n",
       "  [31, 'theano', 15],\n",
       "  [33, 'theano', 16],\n",
       "  [33, 'theano', 15],\n",
       "  [36, 'theano', 16],\n",
       "  [36, 'theano', 15]])"
      ]
     },
     "execution_count": 43,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "L, _ = nanL.localiseNan()\n",
    "L"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "e5192735",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "fyp",
   "language": "python",
   "name": "fyp"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}