stolenimpl.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "c2a2753c",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import keras\n",
    "import keras.backend as tf\n",
    "from keras.datasets import mnist, cifar10\n",
    "from keras.models import Sequential, Model\n",
    "from keras.layers import Dense, Flatten, Input, Dropout, Activation, Conv2D, MaxPooling2D, ZeroPadding2D\n",
    "import os\n",
    "import numpy as np\n",
    "import hickle\n",
    "from tensorflow import config\n",
    "os.environ['CUDA_VISIBLE_DEVICES']='1'\n",
    "# gpus = config.experimental.list_physical_devices(device_type='GPU')\n",
    "# for gpu in gpus:\n",
    "#    config.experimental.set_memory_growth(gpu, True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "53a6e73d",
   "metadata": {},
   "outputs": [],
   "source": [
    "def build_cifar10model():\n",
    "    model = Sequential()\n",
    "    model.add(Conv2D(32, (3, 3), padding='same',\n",
    "                     input_shape=(32,32,3),activation='relu'))\n",
    "  #  model.add(Activation('relu'))\n",
    "    model.add(Conv2D(32, (3, 3),activation='relu'))\n",
    "   # model.add(Activation('relu'))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "    # model.add(Dropout(0.25))\n",
    "\n",
    "    model.add(Conv2D(64, (3, 3), padding='same',activation='relu'))\n",
    "   # model.add(Activation('relu'))\n",
    "    model.add(Conv2D(64, (3, 3),activation='relu'))\n",
    "  #  model.add(Activation('relu'))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "    # model.add(Dropout(0.25))\n",
    "\n",
    "    model.add(Flatten())\n",
    "    model.add(Dense(512,activation='relu'))\n",
    "    #model.add(Activation('relu'))\n",
    "    # model.add(Dropout(0.5))\n",
    "    model.add(Dense(10, activation='softmax'))\n",
    "   # model.add(Activation('softmax'))\n",
    "\n",
    "\n",
    "#     json_string = model.to_json()\n",
    "#     with open('cifar10.json', 'w') as outfile:\n",
    "#         json.dump(json_string, outfile)\n",
    "    return model\n",
    "#build_cifar10model()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "f0d70aec",
   "metadata": {},
   "outputs": [],
   "source": [
    "import hickle\n",
    "from keras.utils.np_utils import to_categorical\n",
    "def combineandlabeldata(wbs1,zs1,wbs0,zs0):\n",
    "    layernums = len(wbs1)\n",
    "    bs = wbs1[0].shape[0] + wbs0[0].shape[0]\n",
    "    print(bs)\n",
    "    inputx = [None for _ in range(2*layernums)]\n",
    "    inputy = [1 for _ in range(wbs1[0].shape[0])] + [0 for _ in range(wbs0[0].shape[0])]\n",
    "    for i in range(layernums):\n",
    "        inputx[i] = np.concatenate([wbs1[i],wbs0[i]],axis=0)\n",
    "        inputx[i+layernums] = np.concatenate([zs1[i],zs0[i]],axis=0)\n",
    "        \n",
    "\n",
    "  #  inputy = to_categorical(inputy,2)\n",
    "    return inputx, inputy\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "5256870f",
   "metadata": {},
   "outputs": [],
   "source": [
    "def loadtraindata(indexs):\n",
    "    xxx = None\n",
    "    yyy = None\n",
    "    for i in range(indexs):\n",
    "        w0 = hickle.load('1www0_{}.hkl'.format(i))\n",
    "        z0 = hickle.load('1zzz0_{}.hkl'.format(i))\n",
    "        w1 = hickle.load('1www1_{}.hkl'.format(i))\n",
    "        z1 = hickle.load('1zzz1_{}.hkl'.format(i))\n",
    "        xx,yy = combineandlabeldata(w1,z1,w0,z0)\n",
    "      #  print(yy.shape)\n",
    "        if xxx is None:\n",
    "            xxx = xx\n",
    "            yyy = yy\n",
    "        else:\n",
    "            for j in range(len(xxx)):\n",
    "                \n",
    "                xxx[j] = np.concatenate([xxx[j],xx[j]],axis=0)\n",
    "            yyy = np.concatenate([yyy,yy],axis=0)\n",
    "    return xxx,yyy\n",
    "\n",
    "def loadtestdata(indexs):\n",
    "    xxx = None\n",
    "    yyy = None\n",
    "    for i in range(indexs-2,indexs):\n",
    "        w0 = hickle.load('1www0_{}.hkl'.format(i))\n",
    "        z0 = hickle.load('1zzz0_{}.hkl'.format(i))\n",
    "        w1 = hickle.load('1www1_{}.hkl'.format(i))\n",
    "        z1 = hickle.load('1zzz1_{}.hkl'.format(i))\n",
    "        xx,yy = combineandlabeldata(w1,z1,w0,z0)\n",
    "      #  print(yy.shape)\n",
    "        if xxx is None:\n",
    "            xxx = xx\n",
    "            yyy = yy\n",
    "        else:\n",
    "            for j in range(len(xxx)):\n",
    "                \n",
    "                xxx[j] = np.concatenate([xxx[j],xx[j]],axis=0)\n",
    "            yyy = np.concatenate([yyy,yy],axis=0)\n",
    "    return xxx,yyy\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "6366901a",
   "metadata": {},
   "outputs": [],
   "source": [
    "%%capture\n",
    "xxx,yyy = loadtraindata(5)\n",
    "#xxx2,yyy2 = loadtestdata(5)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "57078a47",
   "metadata": {},
   "outputs": [],
   "source": [
    "bsize= xxx[0].shape[0]\n",
    "\n",
    "\n",
    "    \n",
    "shuffleindex = np.random.choice(bsize, bsize, replace=False) \n",
    "\n",
    "for index,i in enumerate(xxx):\n",
    "\n",
    "    xxx[index] = i[shuffleindex]\n",
    "\n",
    "yyy = yyy[shuffleindex]\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "5af4a60d",
   "metadata": {},
   "outputs": [],
   "source": [
    "from keras.layers import Conv1D,Reshape,Lambda,Multiply,Add,Concatenate,Permute\n",
    "from keras.models import Model\n",
    "\n",
    "def models(inputswb,inputsz):\n",
    "\n",
    "    \n",
    "    confidence = []\n",
    "    inputwbs = []\n",
    "    inputzs = []\n",
    "#     inputwbs = Input(shape=(tf.int_shape(inputswb)))\n",
    "#     inputzs = Input(shape=(tf.int_shape(inputsz)))\n",
    "    \n",
    "    for i in range(len(inputswb)):\n",
    "        \n",
    "#         inputwb = Lambda(lambda x:x[:,i,:])(inputwbs)\n",
    "#         inputz = Lambda(lambda x:x[:,i,:])(inputzs)\n",
    "        inputwb = Input(shape=(tf.int_shape(inputswb[i])))\n",
    "        inputz = Input(shape=(tf.int_shape(inputsz[i])))\n",
    "        inputwbs.append(inputwb)\n",
    "        inputzs.append(inputz)\n",
    "        \n",
    "        wb = inputwb\n",
    "        z = inputz\n",
    "      #  print(wb.shape,z.shape)\n",
    "     #   z = tf.squeeze(inputz,axis=1)\n",
    "     #   print(wb.shape,z.shape)\n",
    "        t1_1 = Conv1D(16, kernel_size=2, strides=2)(wb)\n",
    "        t1_1_2 = Conv1D(1, kernel_size=1, strides=1)(t1_1)\n",
    "        t1_2 = Conv1D(1, kernel_size=1, strides=1)(t1_1_2)\n",
    "    #    print(t1_2.shape,z.shape)\n",
    "     #   dwb = Reshape((z.shape[1]+1,10))(t1_2)  #tf.reshape(t1_2,shape=(tf.shape(wb)[0],z.shape[1]+1,10))\n",
    "        dw = Lambda(lambda x: x[:,:-1])(t1_2)\n",
    "        db = Lambda(lambda x: x[:,-1])(t1_2)\n",
    "        #print(dw.shape)\n",
    "        dw_2 = Reshape(z.shape[1:])(dw)\n",
    "        #dw = dwb[:,:-1]\n",
    "        #db = dwb[:,-1]\n",
    "       # print(z.shape,dw_2.shape)\n",
    "      #  print(dw.shape,z.shape,db.shape)\n",
    "        part1 = Multiply()([z,dw_2])\n",
    "     #   print(part1.shape)\n",
    "        part1_1 = Lambda(lambda x: tf.sum(tf.batch_flatten(x),axis=1))(part1)\n",
    "        \n",
    "       # print(part1_1.shape,db.shape)\n",
    "        \n",
    "        part2 = Add()([part1_1,db])\n",
    "     #   print(part2.shape)\n",
    "        \n",
    "        confidence.append(part2)\n",
    "    confidence_0 = Concatenate(axis=1)(confidence)\n",
    "  #  confidence_0 = Lambda(lambda x:tf.variable(x))(confidence)\n",
    "    #print(confidence_0.shape)\n",
    "   # confidence_1 = Permute((1,0,2))(confidence_0)#Lambda(lambda x:tf.permute_dimensions(x,(1,0,2)))(confidence_0)\n",
    "    \n",
    "   # confidence_2 = Reshape(shape=(tf.shape(wb)[0],len(inputswb)*10))(condidence_1)\n",
    "    #print(confidence_2.shape)\n",
    "    \n",
    "    t3 = Dense(16,activation='relu')(confidence_0)\n",
    "  #  t4 = Dense(8,activation='relu')(t3)\n",
    "  #  t5 = Dense(8,activation='relu')(t4)\n",
    "    output = Dense(1,activation='sigmoid')(t3)\n",
    "    \n",
    "    return Model(inputwbs+inputzs,output)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "539079d8",
   "metadata": {},
   "outputs": [],
   "source": [
    "attackmodel = models([i[0] for i in xxx[:len(xxx)//2]],[i[0] for i in xxx[len(xxx)//2:]])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "4094d1be",
   "metadata": {},
   "outputs": [],
   "source": [
    "#bce = keras.losses.BinaryCrossentropy(from_logits=True)\n",
    "\n",
    "\n",
    "attackmodel.compile(loss=keras.losses.binary_crossentropy,\n",
    "                      optimizer=keras.optimizers.Adam(),\n",
    "                      metrics=['accuracy'])\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "7a9b1d1f",
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "%%capture uu\n",
    "attackmodel.summary()\n",
    "attackmodel.fit(xxx, yyy,\n",
    "                  batch_size=64,#steps_per_epoch=200,# \n",
    "                  epochs=32,\n",
    "                  verbose=2,\n",
    "                 # validation_data=(xxx2,yyy2),\n",
    "                \n",
    "               )\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "b2477fa7",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'Model: \"model_1\"\\n__________________________________________________________________________________________________\\nLayer (type)                    Output Shape         Param #     Connected to                     \\n==================================================================================================\\ninput_1 (InputLayer)            (None, 65538, 1)     0                                            \\n__________________________________________________________________________________________________\\ninput_3 (InputLayer)            (None, 57602, 1)     0                                            \\n__________________________________________________________________________________________________\\ninput_5 (InputLayer)            (None, 28802, 1)     0                                            \\n__________________________________________________________________________________________________\\ninput_7 (InputLayer)            (None, 21634, 1)     0                                            \\n__________________________________________________________________________________________________\\ninput_9 (InputLayer)            (None, 1026, 1)      0                                            \\n__________________________________________________________________________________________________\\ninput_11 (InputLayer)           (None, 22, 1)        0                                            \\n__________________________________________________________________________________________________\\nconv1d_1 (Conv1D)               (None, 32769, 16)    48          input_1[0][0]                    \\n__________________________________________________________________________________________________\\nconv1d_4 (Conv1D)               (None, 28801, 16)    48          input_3[0][0]                    \\n__________________________________________________________________________________________________\\nconv1d_7 (Conv1D)               (None, 14401, 16)    48          input_5[0][0]                    \\n__________________________________________________________________________________________________\\nconv1d_10 (Conv1D)              (None, 10817, 16)    48          input_7[0][0]                    \\n__________________________________________________________________________________________________\\nconv1d_13 (Conv1D)              (None, 513, 16)      48          input_9[0][0]                    \\n__________________________________________________________________________________________________\\nconv1d_16 (Conv1D)              (None, 11, 16)       48          input_11[0][0]                   \\n__________________________________________________________________________________________________\\nconv1d_2 (Conv1D)               (None, 32769, 1)     17          conv1d_1[0][0]                   \\n__________________________________________________________________________________________________\\nconv1d_5 (Conv1D)               (None, 28801, 1)     17          conv1d_4[0][0]                   \\n__________________________________________________________________________________________________\\nconv1d_8 (Conv1D)               (None, 14401, 1)     17          conv1d_7[0][0]                   \\n__________________________________________________________________________________________________\\nconv1d_11 (Conv1D)              (None, 10817, 1)     17          conv1d_10[0][0]                  \\n__________________________________________________________________________________________________\\nconv1d_14 (Conv1D)              (None, 513, 1)       17          conv1d_13[0][0]                  \\n__________________________________________________________________________________________________\\nconv1d_17 (Conv1D)              (None, 11, 1)        17          conv1d_16[0][0]                  \\n__________________________________________________________________________________________________\\nconv1d_3 (Conv1D)               (None, 32769, 1)     2           conv1d_2[0][0]                   \\n__________________________________________________________________________________________________\\nconv1d_6 (Conv1D)               (None, 28801, 1)     2           conv1d_5[0][0]                   \\n__________________________________________________________________________________________________\\nconv1d_9 (Conv1D)               (None, 14401, 1)     2           conv1d_8[0][0]                   \\n__________________________________________________________________________________________________\\nconv1d_12 (Conv1D)              (None, 10817, 1)     2           conv1d_11[0][0]                  \\n__________________________________________________________________________________________________\\nconv1d_15 (Conv1D)              (None, 513, 1)       2           conv1d_14[0][0]                  \\n__________________________________________________________________________________________________\\nconv1d_18 (Conv1D)              (None, 11, 1)        2           conv1d_17[0][0]                  \\n__________________________________________________________________________________________________\\nlambda_1 (Lambda)               (None, 32768, 1)     0           conv1d_3[0][0]                   \\n__________________________________________________________________________________________________\\nlambda_4 (Lambda)               (None, 28800, 1)     0           conv1d_6[0][0]                   \\n__________________________________________________________________________________________________\\nlambda_7 (Lambda)               (None, 14400, 1)     0           conv1d_9[0][0]                   \\n__________________________________________________________________________________________________\\nlambda_10 (Lambda)              (None, 10816, 1)     0           conv1d_12[0][0]                  \\n__________________________________________________________________________________________________\\nlambda_13 (Lambda)              (None, 512, 1)       0           conv1d_15[0][0]                  \\n__________________________________________________________________________________________________\\nlambda_16 (Lambda)              (None, 10, 1)        0           conv1d_18[0][0]                  \\n__________________________________________________________________________________________________\\ninput_2 (InputLayer)            (None, 32, 32, 32)   0                                            \\n__________________________________________________________________________________________________\\nreshape_1 (Reshape)             (None, Dimension(32) 0           lambda_1[0][0]                   \\n__________________________________________________________________________________________________\\ninput_4 (InputLayer)            (None, 30, 30, 32)   0                                            \\n__________________________________________________________________________________________________\\nreshape_2 (Reshape)             (None, Dimension(30) 0           lambda_4[0][0]                   \\n__________________________________________________________________________________________________\\ninput_6 (InputLayer)            (None, 15, 15, 64)   0                                            \\n__________________________________________________________________________________________________\\nreshape_3 (Reshape)             (None, Dimension(15) 0           lambda_7[0][0]                   \\n__________________________________________________________________________________________________\\ninput_8 (InputLayer)            (None, 13, 13, 64)   0                                            \\n__________________________________________________________________________________________________\\nreshape_4 (Reshape)             (None, Dimension(13) 0           lambda_10[0][0]                  \\n__________________________________________________________________________________________________\\ninput_10 (InputLayer)           (None, 512)          0                                            \\n__________________________________________________________________________________________________\\nreshape_5 (Reshape)             (None, Dimension(512 0           lambda_13[0][0]                  \\n__________________________________________________________________________________________________\\ninput_12 (InputLayer)           (None, 10)           0                                            \\n__________________________________________________________________________________________________\\nreshape_6 (Reshape)             (None, Dimension(10) 0           lambda_16[0][0]                  \\n__________________________________________________________________________________________________\\nmultiply_1 (Multiply)           (None, 32, 32, 32)   0           input_2[0][0]                    \\n                                                                 reshape_1[0][0]                  \\n__________________________________________________________________________________________________\\nmultiply_2 (Multiply)           (None, 30, 30, 32)   0           input_4[0][0]                    \\n                                                                 reshape_2[0][0]                  \\n__________________________________________________________________________________________________\\nmultiply_3 (Multiply)           (None, 15, 15, 64)   0           input_6[0][0]                    \\n                                                                 reshape_3[0][0]                  \\n__________________________________________________________________________________________________\\nmultiply_4 (Multiply)           (None, 13, 13, 64)   0           input_8[0][0]                    \\n                                                                 reshape_4[0][0]                  \\n__________________________________________________________________________________________________\\nmultiply_5 (Multiply)           (None, 512)          0           input_10[0][0]                   \\n                                                                 reshape_5[0][0]                  \\n__________________________________________________________________________________________________\\nmultiply_6 (Multiply)           (None, 10)           0           input_12[0][0]                   \\n                                                                 reshape_6[0][0]                  \\n__________________________________________________________________________________________________\\nlambda_3 (Lambda)               (None,)              0           multiply_1[0][0]                 \\n__________________________________________________________________________________________________\\nlambda_2 (Lambda)               (None, 1)            0           conv1d_3[0][0]                   \\n__________________________________________________________________________________________________\\nlambda_6 (Lambda)               (None,)              0           multiply_2[0][0]                 \\n__________________________________________________________________________________________________\\nlambda_5 (Lambda)               (None, 1)            0           conv1d_6[0][0]                   \\n__________________________________________________________________________________________________\\nlambda_9 (Lambda)               (None,)              0           multiply_3[0][0]                 \\n__________________________________________________________________________________________________\\nlambda_8 (Lambda)               (None, 1)            0           conv1d_9[0][0]                   \\n__________________________________________________________________________________________________\\nlambda_12 (Lambda)              (None,)              0           multiply_4[0][0]                 \\n__________________________________________________________________________________________________\\nlambda_11 (Lambda)              (None, 1)            0           conv1d_12[0][0]                  \\n__________________________________________________________________________________________________\\nlambda_15 (Lambda)              (None,)              0           multiply_5[0][0]                 \\n__________________________________________________________________________________________________\\nlambda_14 (Lambda)              (None, 1)            0           conv1d_15[0][0]                  \\n__________________________________________________________________________________________________\\nlambda_18 (Lambda)              (None,)              0           multiply_6[0][0]                 \\n__________________________________________________________________________________________________\\nlambda_17 (Lambda)              (None, 1)            0           conv1d_18[0][0]                  \\n__________________________________________________________________________________________________\\nadd_1 (Add)                     (None, 1)            0           lambda_3[0][0]                   \\n                                                                 lambda_2[0][0]                   \\n__________________________________________________________________________________________________\\nadd_2 (Add)                     (None, 1)            0           lambda_6[0][0]                   \\n                                                                 lambda_5[0][0]                   \\n__________________________________________________________________________________________________\\nadd_3 (Add)                     (None, 1)            0           lambda_9[0][0]                   \\n                                                                 lambda_8[0][0]                   \\n__________________________________________________________________________________________________\\nadd_4 (Add)                     (None, 1)            0           lambda_12[0][0]                  \\n                                                                 lambda_11[0][0]                  \\n__________________________________________________________________________________________________\\nadd_5 (Add)                     (None, 1)            0           lambda_15[0][0]                  \\n                                                                 lambda_14[0][0]                  \\n__________________________________________________________________________________________________\\nadd_6 (Add)                     (None, 1)            0           lambda_18[0][0]                  \\n                                                                 lambda_17[0][0]                  \\n__________________________________________________________________________________________________\\nconcatenate_1 (Concatenate)     (None, 6)            0           add_1[0][0]                      \\n                                                                 add_2[0][0]                      \\n                                                                 add_3[0][0]                      \\n                                                                 add_4[0][0]                      \\n                                                                 add_5[0][0]                      \\n                                                                 add_6[0][0]                      \\n__________________________________________________________________________________________________\\ndense_1 (Dense)                 (None, 16)           112         concatenate_1[0][0]              \\n__________________________________________________________________________________________________\\ndense_2 (Dense)                 (None, 1)            17          dense_1[0][0]                    \\n==================================================================================================\\nTotal params: 531\\nTrainable params: 531\\nNon-trainable params: 0\\n__________________________________________________________________________________________________\\nEpoch 1/32\\n - 87s - loss: 1.2776 - accuracy: 0.6472\\nEpoch 2/32\\n - 84s - loss: 0.4633 - accuracy: 0.7710\\nEpoch 3/32\\n - 84s - loss: 0.4883 - accuracy: 0.7858\\nEpoch 4/32\\n - 85s - loss: 0.4365 - accuracy: 0.7952\\nEpoch 5/32\\n - 86s - loss: 0.4143 - accuracy: 0.8103\\nEpoch 6/32\\n - 85s - loss: 0.4338 - accuracy: 0.8043\\nEpoch 7/32\\n - 85s - loss: 0.4866 - accuracy: 0.8098\\nEpoch 8/32\\n - 85s - loss: 0.4893 - accuracy: 0.7991\\nEpoch 9/32\\n - 85s - loss: 0.4099 - accuracy: 0.8192\\nEpoch 10/32\\n - 85s - loss: 0.4312 - accuracy: 0.8105\\nEpoch 11/32\\n - 84s - loss: 0.4836 - accuracy: 0.8077\\nEpoch 12/32\\n - 84s - loss: 0.4426 - accuracy: 0.8031\\nEpoch 13/32\\n - 84s - loss: 0.4067 - accuracy: 0.8203\\nEpoch 14/32\\n - 82s - loss: 0.4122 - accuracy: 0.8165\\nEpoch 15/32\\n - 84s - loss: 0.4223 - accuracy: 0.8133\\nEpoch 16/32\\n - 87s - loss: 0.4215 - accuracy: 0.8145\\nEpoch 17/32\\n - 86s - loss: 0.4090 - accuracy: 0.8185\\nEpoch 18/32\\n - 86s - loss: 0.4291 - accuracy: 0.8092\\nEpoch 19/32\\n - 85s - loss: 0.4618 - accuracy: 0.8087\\nEpoch 20/32\\n - 84s - loss: 0.4267 - accuracy: 0.8085\\nEpoch 21/32\\n - 85s - loss: 0.4125 - accuracy: 0.8175\\nEpoch 22/32\\n - 84s - loss: 0.4155 - accuracy: 0.8155\\nEpoch 23/32\\n - 85s - loss: 0.4279 - accuracy: 0.8085\\nEpoch 24/32\\n - 85s - loss: 0.4091 - accuracy: 0.8184\\nEpoch 25/32\\n - 84s - loss: 0.4246 - accuracy: 0.8096\\nEpoch 26/32\\n - 83s - loss: 0.4487 - accuracy: 0.8137\\nEpoch 27/32\\n - 82s - loss: 0.4189 - accuracy: 0.8152\\nEpoch 28/32\\n - 83s - loss: 0.4074 - accuracy: 0.8197\\nEpoch 29/32\\n - 83s - loss: 0.4245 - accuracy: 0.8107\\nEpoch 30/32\\n - 83s - loss: 0.4088 - accuracy: 0.8174\\nEpoch 31/32\\n - 82s - loss: 0.4108 - accuracy: 0.8181\\nEpoch 32/32\\n - 82s - loss: 0.4110 - accuracy: 0.8174\\n'"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "uu.stdout"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b999d5b3",
   "metadata": {},
   "source": [
    "import membership_attackload_mnist22\n",
    "targetmodel = build_cifar10model()\n",
    "\n",
    "targetmodel.load_weights('shadow_999.h5')\n",
    "\n",
    "proxymodel = build_cifar10model()\n",
    "proxymodel.load_weights('proxy_999.h5')\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "85d4196a",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0.77613354, 0.7722682, 0.7892767, 0.7622627, 0.8001671, 0.76460016, 0.8101902, 0.78200316, 0.7836705, 0.7705675]\n"
     ]
    }
   ],
   "source": [
    "\n",
    "def loadcaldata():\n",
    "    xxx = [None for _ in range(10)]\n",
    "  #  yyy = [None for _ in range(10)]\n",
    "    for i in range(10):\n",
    "        w = hickle.load('calw_{}.hkl'.format(i))\n",
    "        z = hickle.load('calz_{}.hkl'.format(i))\n",
    "        #w1 = hickle.load('1www1_{}.hkl'.format(i))\n",
    "        #z1 = hickle.load('1zzz1_{}.hkl'.format(i))\n",
    "      #  xx,yy = combineandlabeldata2(w,z)\n",
    "      #  print(yy.shape)\n",
    "#         layernum = w.shape[0]\n",
    "#         xt = []\n",
    "#         for j in range(layernum):\n",
    "#             xt.append(w[j])\n",
    "#         for j in range(layernum):\n",
    "#             xt.append(z[j])\n",
    "            \n",
    "        xxx[i] = w+z\n",
    "       # yyy[i] = yy\n",
    "    return xxx\n",
    "\n",
    "\n",
    "def getcalvalue():\n",
    "    values = []\n",
    "    xxx = loadcaldata()\n",
    "    for i in range(10):\n",
    "        calres = attackmodel.predict(xxx[i])\n",
    "        calres = calres.reshape(-1)\n",
    "        calres.sort(axis=0)\n",
    "       # print(calres)\n",
    "        vv = calres[int(0.90*calres.shape[0])]\n",
    "        #print(vv)\n",
    "        values.append(vv)\n",
    "    return values\n",
    "\n",
    "values = getcalvalue()\n",
    "print(values)\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "0b57f817",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "3017\n",
      "2977\n",
      "2993\n",
      "2992\n",
      "3001\n",
      "3046\n",
      "2984\n",
      "2958\n",
      "3027\n",
      "3005\n"
     ]
    }
   ],
   "source": [
    "def loadtestdatabylabel():\n",
    "    xxx = [None for _ in range(10)]\n",
    "    yyy = [None for _ in range(10)]\n",
    "    for i in range(10):\n",
    "        w0 = hickle.load('testw0_{}.hkl'.format(i))\n",
    "        z0 = hickle.load('testz0_{}.hkl'.format(i))\n",
    "        w1 = hickle.load('testw1_{}.hkl'.format(i))\n",
    "        z1 = hickle.load('testz1_{}.hkl'.format(i))\n",
    "        xx,yy = combineandlabeldata(w1,z1,w0,z0)\n",
    "      #  print(yy.shape)\n",
    "        xxx[i] = xx\n",
    "        yyy[i] = yy\n",
    "    return xxx,yyy\n",
    "\n",
    "def loadtestdataall():\n",
    "    xxx = None\n",
    "    yyy = None\n",
    "    for i in range(10):\n",
    "        w0 = hickle.load('testw0_{}.hkl'.format(i))\n",
    "        z0 = hickle.load('testz0_{}.hkl'.format(i))\n",
    "        w1 = hickle.load('testw1_{}.hkl'.format(i))\n",
    "        z1 = hickle.load('testz1_{}.hkl'.format(i))\n",
    "        xx,yy = combineandlabeldata(w1,z1,w0,z0)\n",
    "      #  print(yy.shape)\n",
    "        if xxx is None:\n",
    "            xxx = xx\n",
    "            yyy = yy\n",
    "        else:\n",
    "            for j in range(len(xxx)):\n",
    "                \n",
    "                xxx[j] = np.concatenate([xxx[j],xx[j]],axis=0)\n",
    "            yyy = np.concatenate([yyy,yy],axis=0)\n",
    "    return xxx,yyy\n",
    "\n",
    "testxx, testyy = loadtestdataall()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "2cdeb1a7",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.8190132962022552 0.9999333333333333 0.6935312341055162 0.7790333333333334\n"
     ]
    }
   ],
   "source": [
    "\n",
    "\n",
    "\n",
    "from sklearn.metrics import f1_score, recall_score, precision_score, accuracy_score\n",
    "ttt = attackmodel.predict(testxx)\n",
    "\n",
    "testresa = []\n",
    "\n",
    "for j in ttt:\n",
    "    if j >= 0.5:\n",
    "        testresa.append(1)\n",
    "    else:\n",
    "        testresa.append(0)\n",
    "\n",
    "f1 = f1_score(testyy,testresa)\n",
    "recall = recall_score(testyy,testresa)\n",
    "precision = precision_score(testyy,testresa)\n",
    "accuracy = accuracy_score(testyy,testresa)\n",
    "\n",
    "print(f1,recall,precision,accuracy)\n",
    "\n",
    "\n",
    "        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "77f13cf2",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "3017\n",
      "2977\n",
      "2993\n",
      "2992\n",
      "3001\n",
      "3046\n",
      "2984\n",
      "2958\n",
      "3027\n",
      "3005\n",
      "0 0.4080865068171132 0.2847769028871391 0.7197346600331676 0.5826980444149817\n",
      "1 0.3744427934621099 0.25386165211551376 0.7132075471698113 0.5757473967080954\n",
      "2 0.5738684884713919 0.4438573315719947 0.8115942028985508 0.6665552956899432\n",
      "3 0.7187993680884676 0.6177868295994569 0.8593012275731823 0.7620320855614974\n",
      "4 0.48584474885844753 0.3486238532110092 0.8012048192771084 0.6247917360879707\n",
      "5 0.6914440572977159 0.5886618325642716 0.8377110694183865 0.7383453709783322\n",
      "6 0.337544214249621 0.2232620320855615 0.6915113871635611 0.560656836461126\n",
      "7 0.4395180722891566 0.3095723014256619 0.7574750830564784 0.6068289384719405\n",
      "8 0.330126582278481 0.2180602006688963 0.6791666666666667 0.5629335976214074\n",
      "9 0.44032334759866854 0.3101138647019424 0.759016393442623 0.6083194675540765\n",
      "0.49240592930444693 0.35986666666666667 0.7794945848375451 0.6290333333333333\n"
     ]
    }
   ],
   "source": [
    "\n",
    "from sklearn.metrics import f1_score, recall_score, precision_score, accuracy_score\n",
    "\n",
    "def caltest():\n",
    "    testxx2,testyy2 = loadtestdatabylabel()\n",
    "    totaltestres = [] \n",
    "    totaltrue = []\n",
    "    for i in range(10):\n",
    "        testres = []\n",
    "        caltest = attackmodel.predict(testxx2[i])\n",
    "        \n",
    "        \n",
    "        for j in caltest:\n",
    "            if j >= values[i]:\n",
    "                testres.append(1)\n",
    "            else:\n",
    "                testres.append(0)\n",
    "                \n",
    "        totaltestres += testres\n",
    "        totaltrue += testyy2[i]\n",
    "        \n",
    "        f1 = f1_score(testyy2[i],testres)\n",
    "        recall = recall_score(testyy2[i],testres)\n",
    "        precision = precision_score(testyy2[i],testres)\n",
    "        accuracy = accuracy_score(testyy2[i],testres)\n",
    "        \n",
    "        print(i,f1,recall,precision,accuracy)\n",
    "        \n",
    "    f1 = f1_score(totaltrue,totaltestres)\n",
    "    recall = recall_score(totaltrue,totaltestres)\n",
    "    precision = precision_score(totaltrue,totaltestres)\n",
    "    accuracy = accuracy_score(totaltrue,totaltestres)\n",
    "\n",
    "    print(f1,recall,precision,accuracy)\n",
    "    \n",
    "caltest()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "bd23e11f",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "0b4aafd8",
   "metadata": {},
   "outputs": [],
   "source": [
    "'''\n",
    "from attribution.methods import InternalInfluence, AumannShapley\n",
    "\n",
    "\n",
    "\n",
    "import membership_attackload_mnist22\n",
    "\n",
    "x_train1, x_test1, y_train1, y_test1 = membership_attackload_mnist22.targetload(database='cifar10')\n",
    "\n",
    "model1 = build_cifar10model() #model_from_json(json_string)\n",
    "model1.load_weights('shadow_999.h5')\n",
    "model1.compile(loss=keras.losses.categorical_crossentropy,\n",
    "                      optimizer=keras.optimizers.SGD(),\n",
    "                      metrics=['accuracy'])\n",
    "\n",
    "from keras import layers, models\n",
    "\n",
    "input_layer = layers.Input(batch_shape=model1.layers[0].input_shape)\n",
    "prev_layer = input_layer\n",
    "for layer in model1.layers:\n",
    "    prev_layer = layer(prev_layer)\n",
    "\n",
    "model2 = models.Model([input_layer], [prev_layer])\n",
    "\n",
    "def _get_inbound_layers(l):\n",
    "\n",
    "    inbound_layers = []\n",
    "    for i_node in l._inbound_nodes:\n",
    "        print(i_node.inbound_layers)\n",
    "        for i_layer in i_node.inbound_layers:\n",
    "            print(i_layer)\n",
    "            inbound_layers.append(i_layer)\n",
    "    return inbound_layers\n",
    "\n",
    "import tensorflow as tf\n",
    "tf.compat.v1.disable_eager_execution()\n",
    "\n",
    "attr = InternalInfluence(model1, model1.layers[2], Q=None,D='linear_interp',multiply_activation=True).compile().get_attributions(x_train1[:1], batch_size=1)\n",
    "\n",
    "import keras.backend as K\n",
    "\n",
    "\n",
    "z = model1.layers[0].output\n",
    "r = K.variable(1)\n",
    "baseline = K.variable(K.zeros(shape=K.int_shape(z)[1:]))\n",
    "\n",
    "b = K.expand_dims(baseline, axis=0)\n",
    "print('rr')\n",
    "        # Allocate the alpha term for the interpolation.\n",
    "\n",
    "a1=1. + K.arange(1.0, dtype='float32')\n",
    "print('rrr',a1)\n",
    "a2=K.cast(r, 'float32')\n",
    "print('rrr2',a2)\n",
    "a3=K.int_shape(z)[0] \n",
    "print('rrr3',a3)\n",
    "a = K.tile(\n",
    "     a1/ a2,\n",
    "    [a3])\n",
    "print('rrrr')\n",
    "for _ in range(K.ndim(z) - 1):\n",
    "    a = K.expand_dims(a, axis=-1)\n",
    "\n",
    "\n",
    "K.shape(z)\n",
    "'''"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python [conda env:tf2]",
   "language": "python",
   "name": "conda-env-tf2-py"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}