Constrained_Coalition+Nystrom.ipynb

{"nbformat":4,"nbformat_minor":0,"metadata":{"colab":{"name":"FOR_UPLOAD.ipynb","provenance":[],"authorship_tag":"ABX9TyMOK01XBqEguR1j4u7mqU+2"},"kernelspec":{"name":"python3","display_name":"Python 3"},"language_info":{"name":"python"}},"cells":[{"cell_type":"code","metadata":{"id":"OnG8yAbdenW1"},"source":["!pip install pyitlib"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"jEpDFenHflRP"},"source":["from google.colab import drive\n","drive.mount('/content/drive/')"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"-doduIn5fok_"},"source":["import numpy as np\n","from pyitlib import discrete_random_variable as drv\n","import pandas as pd\n","import math"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"oE3KQia_fr8d"},"source":["train = np.load(\"/content/drive/My Drive/VGG16_train_features.npy\")\n","print(train.shape)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"_R0tooPsfuJ-"},"source":["from itertools import chain, combinations\n","\n","def powerset(iterable):\n","    \"powerset([1,2,3]) --> () (1,) (2,) (3,) (1,2) (1,3) (2,3) (1,2,3)\"\n","    s = list(iterable)\n","    return chain.from_iterable(combinations(s, r) for r in range(1, len(s)+1))"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"X_JFZ5BLf0cV"},"source":["b = []\n","for i in range(1152):\n","  a=[]\n","  if(i<8):\n","    for j in range(i):\n","      a.append(j)\n","    for j in range(i+1,8):\n","      a.append(j)\n","    print(a)\n","    b.append(a)\n","  elif (i>=1148):\n","    for j in range(1144,i):\n","      a.append(j)\n","    for j in range(i+1,1152):\n","      a.append(j)\n","    print(a)\n","    b.append(a)\n","  else:\n","    for j in range(i-4,i):\n","      a.append(j)\n","    for j in range(i+1,i+4):\n","      a.append(j)\n","    print(a)\n","    b.append(a)\n","print(b)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"7sVX0GlRf3Q5"},"source":["import pandas as pd\n","dataframe_train = pd.read_csv(\"/content/drive/My Drive/train_feature_csv\")\n","y_train = dataframe_train['label']\n","y_train = y_train.to_numpy()\n","print(y_train)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"jD6GpABJgAxl"},"source":["for i in range(1000,1152):\n","  X_i = train[:,i]\n","  shap_score = 0\n","  classifier_list =list(powerset(b[i]))\n","  for k in range(len(classifier_list)):\n","    classifier_subset = list(classifier_list[k])\n","    m_info_classif = 0\n","    m_info_class = drv.information_mutual(X_i, y_train)\n","    for j in range(len(classifier_subset)):\n","      X_j= train[:,classifier_subset[j]]\n","      temp = drv.information_mutual(X_i, X_j)\n","      m_info_classif = m_info_classif + temp\n","    m_info_classif = m_info_classif/len(classifier_subset)\n","    marginal_info = m_info_class-m_info_classif\n","    shap_score = shap_score + math.factorial(len(classifier_subset))*math.factorial((8 - len(classifier_subset) -1))*marginal_info/math.factorial(8)\n","  print(shap_score)\n","  shapley_value.append(shap_score)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"UzPosEkigGbG"},"source":["minimum = min(shapley_value)\n","sh = shapley_value/minimum\n","sh = [ -x for x in sh]"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"j0bu2rVPgX2U"},"source":["features = []\n","for i in range(len(sh)):\n","  features.append([i,sh[i]])\n","\n","def Sort(sub_li): \n","    sub_li.sort(key = lambda x: x[1],reverse=True) \n","    return sub_li\n","\n","f = Sort(features)\n","selected = f[0:750]"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"YjJHY14ggZ38"},"source":["numbers=[]\n","for i in range(len(selected)):\n","  numbers.append(selected[i][0])"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"SAS9EbWggz7e"},"source":["x_train_selected = []\n","for i in range(len(numbers)):\n","  x_train_selected.append(train[:,numbers[i]])\n","\n","x_train_selected = np.transpose(x_train_selected)\n","x_train_selected.shape"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"eaYuQxkmqCBO"},"source":["a = x_train_selected\n","a.shape"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"DXjD17pmqJWi"},"source":["!pip install cur"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"q9pgRcEyqKL9"},"source":["r = 300\n","from cur import cur_decomposition\n","C, U, k = cur_decomposition(a, r)\n","print(a.shape)\n","print(k.shape)\n","print(C.shape)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"X70FZ9OIqRsb"},"source":["similarity = np.zeros([300,300])\n","for i in range(300):\n","  for j in range(300):\n","    similarity[i][j] = np.dot(k[i],k[j])/(np.linalg.norm(k[i])*np.linalg.norm(k[j]))\n","print(similarity)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"7BEWvYJMqVIb"},"source":["I=np.identity(750)\n","o=np.ones([750,750])/750\n","temp=I-o\n","temp1=np.dot(temp,similarity)\n","similarity=np.dot(temp1,temp)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"umh34oZHrWIS"},"source":["eig_val,eig_vec = np.linalg.eig(similarity)\n","eig_val = np.real(eig_val)\n","eig_vec = np.real(eig_vec)\n","sig = np.diag(eig_val)similarity1 = np.zeros([3975,300])\n","for i in range(3975):\n","  for j in range(300):\n","    similarity1[i][j] = np.dot(a[i],k[j])/(np.linalg.norm(a[i])*np.linalg.norm(k[j]))\n","print(similarity1)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"33N-9tkcrjnH"},"source":["temp = np.dot(similarity1,eig_vec)\n","representation = np.dot(temp,np.linalg.inv(sig))\n","print(representation)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"4m0binsornQD"},"source":["np.save('/content/' +'REPRESENTATION_2500_K'+'.npy', representation)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"cm1dkMaorpZ_"},"source":["representation = a\n","representation_train = representation[0:3086]\n","representation_test = representation[3086:3206]\n","representation_val = representation[3206:3975]\n","print(representation_test.shape)\n","print(representation_train.shape)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"AQAUIEhJr2TW"},"source":["import pandas as pd\n","dataframe_train = pd.read_csv(\"/content/drive/My Drive/train_feature_csv\")\n","y_train = dataframe_train['label']\n","y_train = y_train.to_numpy()\n","print(y_train)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"CqQ7aBoeratl"},"source":["import pandas as pd\n","dataframe_test = pd.read_csv(\"/content/drive/My Drive/test_features_csv\")\n","y_test = dataframe_test['label']\n","y_test = y_test.to_numpy()\n","print(y_test)\n","print(y_test.shape)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"abQYe0hYr8Tg"},"source":["import pandas as pd\n","dataframe_test = pd.read_csv(\"/content/drive/My Drive/val_feature_csv\")\n","y_val = dataframe_test['label']\n","y_val = y_val.to_numpy()\n","print(y_val)\n","print(y_val.shape)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"Bf622Aoyr_u5"},"source":["import keras\n","y_train = keras.utils.to_categorical(y_train, num_classes=3)\n","y_val = keras.utils.to_categorical(y_val, num_classes=3)\n","y_test = keras.utils.to_categorical(y_test)\n","print(y_test)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"CcmAp_7gsEc7"},"source":["from keras.models import Sequential\n","from keras.layers import Dense\n","from keras.layers import Dropout\n","# 32 .4 16 .5 3 (90)\n","model = Sequential()\n","model.add(Dense(32, activation='relu'))\n","model.add(Dropout(0.4))\n","model.add(Dense(16, activation='relu'))\n","model.add(Dropout(0.5))\n","model.add(Dense(3,activation='softmax',name= 'output'))"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"SIBF7MqDsJwc"},"source":["from sklearn.preprocessing import StandardScaler\n","sc = StandardScaler()\n","representation_train = sc.fit_transform(representation_train)\n","representation_test = sc.fit_transform(representation_test)\n","representation_val = sc.fit_transform(representation_val)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"hBZCDzPsuaeX"},"source":["adam_opt=Adam(lr = 0.001, beta_1=0.6, beta_2=0.8,amsgrad=True)\n","\n","model.compile(optimizer=adam_opt, loss='categorical_crossentropy', metrics=['accuracy'])"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"cfjOxRbSumnt"},"source":["model.compile(loss='categorical_crossentropy', optimizer='adam',metrics=['accuracy'])\n","history = model.fit(representation_train,y_train, epochs=500, batch_size=32, validation_data=(representation_test,y_test))"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"MSe9uuI8utGo"},"source":["import pandas as pd\n","dataframe_test = pd.read_csv(\"/content/drive/My Drive/test_feature_csv\")\n","y_test = dataframe_test['label']\n","y_test = y_test.to_numpy()\n","print(y_test)\n","print(y_test.shape)"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"IFjloFugutth"},"source":["model.predict(representation_test)"],"execution_count":null,"outputs":[]}]}