def mains(t): (X_train, y_train, X_test, Y_test) = (t[0], t[1], t[2], t[3]) X_train= np.reshape(t[0], (-1, 1)) X_test =np.reshape(t[2], (-1, 1)) data_dim = X_train.shape[1] idx_test = 0 tot_iter = 1 plt.plot(list(range(0, len(Y_test.flatten()))), Y_test.flatten(), color='Black', linewidth=2, label='Actual') y_predTotal = np.array([]) for time in range(tot_iter): idx_test = idx_test + 1 X_train_plot = np.mean(X_train, axis=1).flatten() X_test_plot = np.mean(X_test, axis=1).flatten() linear_svr = build_SVR('linear', 1000) print(X_train) print(X_test) print(y_train) print(Y_test) linear_svr.fit(X_train, y_train) y_pred = linear_svr.predict(X_test) y_pred.reshape(1, X_test.shape[0]) # logging.info(" %f linear predicted result: %s" %(idx_test, y_pred)) if y_predTotal.shape[0] < 1: y_predTotal = y_pred else: y_predTotal = np.append(y_predTotal, y_pred, axis=0) # plt.scatter(y_train, y_train, color='blue') # plt.show() # plt.plot(X_train, y_train, color='red', linewidth=2) # plt.plot(X_test, y_predTotal.flatten(), color='blue', linewidth=2,label='linear') plt.plot(list(range(0, len(y_predTotal.flatten()))), y_predTotal.flatten(), color='blue', linewidth=2, label='linear') plt.legend() # plt.savefig('images/linear.png') y_predTotal = np.reshape(y_predTotal, (tot_iter, X_test.shape[0])) # logging.info(" linears result: %s" % (y_predTotal)) logging.info(' linears predicted mean result: %s' % y_predTotal.mean(axis=0)) # score= mean_squared_error(y_pred, y_test) # mae_score = mean_absolute_error(y_pred, y_test) # logging.info(" linear svr result: %f_%f" %(score, mae_score)) y_predTotal = np.array([]) idx_test = 0 for time in range(tot_iter): idx_test = idx_test + 1 NUM_ESTIMATOR = 50 NUM_PREEPOCH = 150 NUM_BPEPOCH = 175 BATH_SIZE = 50 rf = build_RF(NUM_ESTIMATOR) rf.fit(X_train, y_train) y_pred = rf.predict(X_test) y_pred.reshape(1, X_test.shape[0]) # logging.info(" r fores predict result: %s" %(y_pred)) if y_predTotal.shape[0] < 1: y_predTotal = y_pred else: y_predTotal = np.append(y_predTotal, y_pred, axis=0) y_predTotal = np.reshape(y_predTotal, (tot_iter, X_test.shape[0])) # plt.scatter(y_train, y_train, color='blue') # plt.show() # plt.clf() # plt.plot(X_train, y_train, color='red', linewidth=2) # plt.plot(X_test, y_predTotal.flatten(), color='blue', linewidth=2) plt.plot(list(range(0, len(y_predTotal.flatten()))), y_predTotal.flatten(), color='red', linewidth=2, label='random forest') plt.legend() t1=list(range(0, len(y_predTotal.flatten()))) t2=y_predTotal.flatten() t=[] for a in t1: for b in t2: t.append((a,b)) # plt.savefig('images/rforest.png') logging.info(' r fore predicted mean result: %s' % y_predTotal.mean(axis=0)) savetxt('dataRF.csv', t, delimiter=',') # score= mean_squared_error(y_pred, y_test) # mae_score = mean_absolute_error(y_pred, y_test) # logging.info(" randomforest result: %f_%f" %(score, mae_score)) # neural network y_predTotal = np.array([]) idx_test = 0 for time in range(tot_iter): idx_test = idx_test + 1 nn_model = build_NN(data_dim) sc = StandardScaler() nn_model.fit(X_train, y_train, epochs=50, batch_size=BATH_SIZE) y_pred = nn_model.predict(X_test) y_predact = y_pred y_pred.reshape(1, X_test.shape[0]) # logging.info(" neu predict result: %s" %(y_pred)) if y_predTotal.shape[0] < 1: y_predTotal = y_predact else: y_predTotal = np.append(y_predTotal, y_predact, axis=0) y_predTotal = np.reshape(y_predTotal, (tot_iter, X_test.shape[0])) # plt.clf() # plt.plot(X_train, y_train, color='red', linewidth=2) # plt.plot(X_test, y_predTotal.flatten(), color='green', linewidth=2) plt.plot(list(range(0, len(y_predTotal.flatten()))), y_predTotal.flatten(), color='green', linewidth=2, label='Neural') plt.legend() # plt.savefig('images/neural.png') logging.info(' neu predicted mean result: %s' % y_predTotal.mean(axis=0)) idx_test = 0 y_predTotal = np.array([]) for time in range(tot_iter): idx_test = idx_test + 1 normal_AE = build_pre_normalAE(data_dim, X_train, epoch_pretrain=NUM_PREEPOCH, hidDim=[140, 280]) normal_AE.fit(X_train, y_train, epochs=NUM_BPEPOCH, batch_size=BATH_SIZE) y_pred = normal_AE.predict(X_test) y_pred.reshape(1, X_test.shape[0]) # logging.info(" ae predict result: %s" %(y_pred)) if y_predTotal.shape[0] < 1: y_predTotal = y_pred else: y_predTotal = np.append(y_predTotal, y_pred, axis=0) y_predTotal = np.reshape(y_predTotal, (tot_iter, X_test.shape[0])) # plt.clf() # plt.plot(X_train, y_train, color='red', linewidth=2) # plt.plot(X_test, y_predTotal.flatten(), color='red', label='AE') plt.plot(list(range(0, len(y_predTotal.flatten()))), y_predTotal.flatten(), color='red', linewidth=2, label='auto encoder') plt.legend() # plt.savefig('images/ae.png') logging.info(' ae predicted mean result: %s' % y_predTotal.mean(axis=0)) # score = mean_squared_error(y_pred, y_test) # mae_score = mean_absolute_error(y_pred, y_test) # logging.info(" normal ae result: %f_%f" %(score, mae_score)) # denoise AE idx_test = 0 y_predTotal = np.array([]) for time in range(tot_iter): idx_test = idx_test + 1 denois_AE = build_pre_denoiseAE(data_dim, X_train, epoch_pretrain=NUM_PREEPOCH, hidDim=[140, 280]) denois_AE.fit(X_train, y_train, epochs=NUM_BPEPOCH, batch_size=BATH_SIZE) y_pred = denois_AE.predict(X_test) y_pred.reshape(1, X_test.shape[0]) # logging.info(" denoiseae predict result: %s" %(y_pred)) if y_predTotal.shape[0] < 1: y_predTotal = y_pred else: y_predTotal = np.append(y_predTotal, y_pred, axis=0) y_predTotal = np.reshape(y_predTotal, (tot_iter, X_test.shape[0])) # plt.clf() # plt.plot(X_train, y_train, color='red', linewidth=2) # plt.plot(X_test, y_predTotal.flatten(), color='cyan', linewidth=2) plt.plot(list(range(0, len(y_predTotal.flatten()))), y_predTotal.flatten(), color='cyan', linewidth=2, label='Denoise AE') plt.legend() # plt.savefig('images/denoiseAE.png') logging.info(' denoiseae predicted mean result: %s' % y_predTotal.mean(axis=0)) # score = mean_squared_error(y_pred, y_test) # mae_score = mean_absolute_error(y_pred, y_test) # logging.info(" denoise ae result: %f_%f" %(score, mae_score)) # rbf using sigmoid function, feature should be scaled to -1 and 1 idx_test = 0 for time in range(tot_iter): idx_test = idx_test + 1 scaler = MinMaxScaler() X_train_rbm = scaler.fit_transform(X_train) rbm = build_RBM(NUM_BPEPOCH, NUM_PREEPOCH, batch_size=BATH_SIZE) rbm.fit(X_train_rbm, y_train) X_test_rbm = scaler.transform(X_test) y_pred = rbm.predict(X_test_rbm) y_predTotal = np.reshape(y_pred, (tot_iter, X_test.shape[0])) logging.info(' dbn predict result: %s' % y_pred) plt.plot(list(range(0, len(y_predTotal.flatten()))), y_predTotal.flatten(), color='#808000', linewidth=2, label='dbn') plt.legend() # plt.plot(X_test, y_predTotal.flatten(), color='magenta', linewidth=2) # plt.savefig('images/biLSTM.png') # score = mean_squared_error(y_pred, y_test) # mae_score = mean_absolute_error(y_pred, y_test) # logging.info(" dbn result: %f_%f" %(score, mae_score)) # Bi-directional LSTM # t = load_data(False) X_train_init = X_train X_test_init = X_test # X_train = np.reshape(X_train, ( 1,X_train.shape[1],X_train.shape[0])) data_dim = X_train.shape[1] X_train = np.reshape(X_train, (X_train.shape[0], 1, X_train.shape[1])) X_test = np.reshape(X_test, (X_test.shape[0], 1, X_test.shape[1])) timesteps = 1 #data_dim = X_train.shape[2] #timesteps = X_train.shape[0] biLSTM = build_BILSTM(timesteps, data_dim) biLSTM.fit(X_train, y_train, epochs=NUM_BPEPOCH, batch_size=BATH_SIZE) y_pred = biLSTM.predict(X_test) y_predTotal = np.reshape(y_pred, (tot_iter, X_test.shape[0])) logging.info('predicted BILSTM mean result: %s' % y_predTotal.mean(axis=0)) # plt.clf() # plt.plot(X_train, y_train, color='red', linewidth=2) plt.plot(list(range(0, len(y_predTotal.flatten()))), y_predTotal.flatten(), color='yellow', linewidth=2, label='BILSTM') plt.legend() # plt.plot(X_test, y_predTotal.flatten(), color='magenta', linewidth=2) # plt.savefig('images/biLSTM.png') # score = mean_squared_error(y_pred, y_test) # mae_score = mean_absolute_error(y_pred, y_test) # logging.info(" bi-lstm result: %f_%f" %(score, mae_score)) # LSTM LSTM = build_LSTM(timesteps, data_dim) LSTM.fit(X_train, y_train, epochs=NUM_BPEPOCH, batch_size=BATH_SIZE) y_pred = LSTM.predict(X_test) # score = mean_squared_error(y_pred, y_test) # mae_score = mean_absolute_error(y_pred, y_test) y_predTotal = np.reshape(y_pred, (tot_iter, X_test.shape[0])) logging.info('predicted mean result: %s' % y_predTotal.mean(axis=0)) # plt.clf() # plt.plot(X_train_plot, y_train, color='red', linewidth=2) # plt.plot(X_test_plot, y_predTotal.flatten(), color='blue', linewidth=2) plt.plot(list(range(0, len(y_predTotal.flatten()))), y_predTotal.flatten(), color='blue', linewidth=2, label='LSTM') plt.legend() # plt.savefig('images/LSTM.png') # CNN X_train = X_train_init X_test = X_test_init X_train = X_train.reshape(50, 1, 1) X_test = X_test.reshape(25, 1, 1) CNN = build_CNN(1, 1) CNN.fit(X_train, y_train, epochs=NUM_BPEPOCH, batch_size=BATH_SIZE) y_pred = CNN.predict(X_test) savetxt('dataLSTM.csv', y_predTotal, delimiter=',') #plt.clf() #plt.plot(X_train_plot, y_train, color='red', linewidth=2) #plt.plot(X_test_plot, y_pred.flatten(), color='blue', linewidth=2) y_predTotal = np.reshape(y_pred, (tot_iter, X_test.shape[0])) logging.info('predicted mean result: %s' % y_predTotal.mean(axis=0)) # plt.clf() y_predTotal = preprocessing.normalize(y_predTotal) plt.plot(list(range(0, len(y_predTotal.flatten()))), y_predTotal.flatten(), color='green', linewidth=2, label='CNN') plt.legend() plt.savefig('images/combo.png') savetxt('dataCNN.csv', y_predTotal, delimiter=',')
denois_AE = build_pre_denoiseAE(data_dim, X_train, epoch_pretrain=NUM_PREEPOCH, hidDim=[140, 280]) denois_AE.fit(X_train, y_train, epochs=NUM_BPEPOCH, batch_size=BATH_SIZE) y_pred = denois_AE.predict(X_test) score = mean_squared_error(y_pred, y_test) mae_score = mean_absolute_error(y_pred, y_test) logging.info(" denoise ae result: %f_%f" % (score, mae_score)) # rbf using sigmoid function, feature should be scaled to -1 and 1 scaler = MinMaxScaler() X_train_rbm = scaler.fit_transform(X_train) rbm = build_RBM(NUM_BPEPOCH, NUM_PREEPOCH, batch_size=BATH_SIZE) rbm.fit(X_train_rbm, y_train) X_test_rbm = scaler.transform(X_test) y_pred = rbm.predict(X_test_rbm) score = mean_squared_error(y_pred, y_test) mae_score = mean_absolute_error(y_pred, y_test) logging.info(" dbn result: %f_%f" % (score, mae_score)) #Bi-directional LSTM #t = load_data(False) """ t=LoadIt() X_train, y_train, X_test, y_test = t[0], t[1], t[2], t[3] data_dim = X_train.shape[2] timesteps = X_train.shape[1] biLSTM = build_BILSTM(timesteps, data_dim)
def mains(t): X_train, y_train, X_test = t[0], t[1], t[2] data_dim = X_train.shape[1] logging.info("X_Train: %s" %(X_train)) logging.info("Y_Train: %s" %(y_train)) logging.info("X_test: %s" %(X_test)) idx_test=0 tot_iter = 1 y_predTotal=np.array([]) for time in range(tot_iter): idx_test = idx_test+1 #t = load_data(True) """ linear_regression=build_LN() linear_regression.fit(X_train,y_train) y_pred = linear_regression.predict(X_test) logging.info(" linear predicted result: %s" %(y_pred)) """ X_train_plot = X_train.flatten() X_test_plot = X_test.flatten() linear_svr = build_SVR('linear',1000) linear_svr.fit(X_train, y_train) y_pred = linear_svr.predict(X_test) y_pred.reshape(1,X_test.shape[0]) #logging.info(" %f linear predicted result: %s" %(idx_test, y_pred)) if (y_predTotal.shape[0] < 1): y_predTotal = y_pred else: y_predTotal = np.append(y_predTotal,y_pred,axis=0) #plt.scatter(y_train, y_train, color='blue') #plt.show() print(X_train.shape) print(X_test.shape) print(y_train.shape) print(y_predTotal.flatten()) plt.plot( X_train_plot, y_train, color='red', linewidth=2) plt.plot( X_test, y_predTotal.flatten(), color='blue', linewidth=2) plt.savefig('linear.png') y_predTotal= np.reshape(y_predTotal,(tot_iter,X_test.shape[0])) logging.info(" linears result: %s" %(y_predTotal)) logging.info(" linears predicted mean result: %s" %(y_predTotal.mean(axis=0))) #score= mean_squared_error(y_pred, y_test) #mae_score = mean_absolute_error(y_pred, y_test) #logging.info(" linear svr result: %f_%f" %(score, mae_score)) y_predTotal=np.array([]) idx_test = 0 for time in range(tot_iter): idx_test = idx_test+1 NUM_ESTIMATOR = 50 NUM_PREEPOCH = 150 NUM_BPEPOCH = 175 BATH_SIZE = 50 rf = build_RF(NUM_ESTIMATOR) rf.fit(X_train, y_train) y_pred = rf.predict(X_test) y_pred.reshape(1,X_test.shape[0]) #logging.info(" r fores predict result: %s" %(y_pred)) if (y_predTotal.shape[0] < 1): y_predTotal = y_pred else: y_predTotal = np.append(y_predTotal,y_pred,axis=0) y_predTotal= np.reshape(y_predTotal,(tot_iter,X_test.shape[0])) #plt.scatter(y_train, y_train, color='blue') #plt.show() plt.clf() plt.plot( X_train_plot, y_train, color='red', linewidth=2) plt.plot( X_test, y_predTotal.flatten(), color='blue', linewidth=2) plt.savefig('rforest.png') logging.info(" r fore predicted mean result: %s" %(y_predTotal.mean(axis=0))) #score= mean_squared_error(y_pred, y_test) #mae_score = mean_absolute_error(y_pred, y_test) #logging.info(" randomforest result: %f_%f" %(score, mae_score)) #neural network y_predTotal=np.array([]) idx_test = 0 for time in range(tot_iter): idx_test = idx_test+1 nn_model = build_NN(data_dim) nn_model.fit(X_train, y_train, epochs=NUM_BPEPOCH, batch_size=BATH_SIZE) y_pred = nn_model.predict(X_test) y_pred.reshape(1,X_test.shape[0]) #logging.info(" neu predict result: %s" %(y_pred)) if (y_predTotal.shape[0] < 1): y_predTotal = y_pred else: y_predTotal = np.append(y_predTotal,y_pred,axis=0) y_predTotal= np.reshape(y_predTotal,(tot_iter,X_test.shape[0])) plt.clf() plt.plot( X_train_plot, y_train, color='red', linewidth=2) plt.plot( X_test, y_predTotal.flatten(), color='blue', linewidth=2) plt.savefig('neural.png') logging.info(" neu predicted mean result: %s" %(y_predTotal.mean(axis=0))) idx_test = 0 y_predTotal=np.array([]) for time in range(tot_iter): idx_test = idx_test+1 normal_AE = build_pre_normalAE(data_dim, X_train, epoch_pretrain=NUM_PREEPOCH, hidDim=[140,280]) normal_AE.fit(X_train, y_train, epochs=NUM_BPEPOCH, batch_size=BATH_SIZE) y_pred = normal_AE.predict(X_test) y_pred.reshape(1,X_test.shape[0]) #logging.info(" ae predict result: %s" %(y_pred)) if (y_predTotal.shape[0] < 1): y_predTotal = y_pred else: y_predTotal = np.append(y_predTotal,y_pred,axis=0) y_predTotal= np.reshape(y_predTotal,(tot_iter,X_test.shape[0])) plt.clf() plt.plot( X_train_plot, y_train, color='red', linewidth=2) plt.plot( X_test, y_predTotal.flatten(), color='blue', linewidth=2) plt.savefig('ae.png') logging.info(" ae predicted mean result: %s" %(y_predTotal.mean(axis=0))) #score = mean_squared_error(y_pred, y_test) #mae_score = mean_absolute_error(y_pred, y_test) #logging.info(" normal ae result: %f_%f" %(score, mae_score)) #denoise AE idx_test = 0 y_predTotal=np.array([]) for time in range(tot_iter): idx_test = idx_test+1 denois_AE = build_pre_denoiseAE(data_dim, X_train, epoch_pretrain=NUM_PREEPOCH, hidDim=[140,280]) denois_AE.fit(X_train, y_train, epochs=NUM_BPEPOCH, batch_size=BATH_SIZE) y_pred = denois_AE.predict(X_test) y_pred.reshape(1,X_test.shape[0]) #logging.info(" denoiseae predict result: %s" %(y_pred)) if (y_predTotal.shape[0] < 1): y_predTotal = y_pred else: y_predTotal = np.append(y_predTotal,y_pred,axis=0) y_predTotal= np.reshape(y_predTotal,(tot_iter,X_test.shape[0])) logging.info(" denoiseae predicted mean result: %s" %(y_predTotal.mean(axis=0))) #score = mean_squared_error(y_pred, y_test) #mae_score = mean_absolute_error(y_pred, y_test) #logging.info(" denoise ae result: %f_%f" %(score, mae_score)) # rbf using sigmoid function, feature should be scaled to -1 and 1 idx_test = 0 for time in range(tot_iter): idx_test = idx_test+1 scaler = MinMaxScaler() X_train_rbm = scaler.fit_transform(X_train) rbm = build_RBM(NUM_BPEPOCH, NUM_PREEPOCH, batch_size=BATH_SIZE) rbm.fit(X_train_rbm, y_train) X_test_rbm = scaler.transform(X_test) y_pred = rbm.predict(X_test_rbm) logging.info(" dbn predict result: %s" %(y_pred)) #score = mean_squared_error(y_pred, y_test) #mae_score = mean_absolute_error(y_pred, y_test) #logging.info(" dbn result: %f_%f" %(score, mae_score)) #Bi-directional LSTM #t = load_data(False) X_train = np.reshape(X_train, (X_train.shape[0],1,X_train.shape[1])) X_test = np.reshape(X_test, (X_test.shape[0],1,X_test.shape[1])) data_dim = X_train.shape[2] timesteps = X_train.shape[1] #biLSTM = build_BILSTM(timesteps, data_dim) #biLSTM.fit(X_train, y_train, epochs=NUM_BPEPOCH, batch_size=BATH_SIZE) #y_pred = biLSTM.predict(X_test) #score = mean_squared_error(y_pred, y_test) #mae_score = mean_absolute_error(y_pred, y_test) #logging.info(" bi-lstm result: %f_%f" %(score, mae_score)) #LSTM LSTM = build_LSTM(timesteps, data_dim) LSTM.fit(X_train, y_train, epochs=NUM_BPEPOCH, batch_size=BATH_SIZE) y_pred = LSTM.predict(X_test) score = mean_squared_error(y_pred, y_test) mae_score = mean_absolute_error(y_pred, y_test) logging.info(" lstm result: %f_%f" %(score, mae_score)) #CNN CNN = build_CNN(timesteps, data_dim) CNN.fit(X_train, y_train, epochs=NUM_BPEPOCH, batch_size=BATH_SIZE) y_pred = CNN.predict(X_test) score = mean_squared_error(y_pred, y_test) mae_score = mean_absolute_error(y_pred, y_test) logging.info(" cnn result: %f_%f" %(score, mae_score))
def mains(t): X_train, y_train, X_test, Y_test = t[0], t[1], t[2],t[3] data_dim = X_train.shape[1] #logging.info("X_Train: %s" % (X_train)) #logging.info("Y_Train: %s" % (y_train)) #logging.info("X_test: %s" % (X_test)) idx_test = 0 tot_iter = 1 plt.plot(list(range(0,len(Y_test.flatten()))), Y_test.flatten(), color='Black', linewidth=2,label='Actual') y_predTotal = np.array([]) for time in range(tot_iter): idx_test = idx_test+1 # t = load_data(True) """ linear_regression=build_LN() linear_regression.fit(X_train,y_train) y_pred = linear_regression.predict(X_test) logging.info(" linear predicted result: %s" %(y_pred)) """ X_train_plot = np.mean(X_train, axis = 1).flatten() X_test_plot =np.mean(X_test, axis = 1).flatten() print("plot") print(X_train_plot) print(X_test_plot) print(y_train) linear_svr = build_SVR('linear', 1000) linear_svr.fit(X_train, y_train) y_pred = linear_svr.predict(X_test) y_pred.reshape(1, X_test.shape[0]) # logging.info(" %f linear predicted result: %s" %(idx_test, y_pred)) if (y_predTotal.shape[0] < 1): y_predTotal = y_pred else: y_predTotal = np.append(y_predTotal, y_pred, axis=0) # plt.scatter(y_train, y_train, color='blue') # plt.show() print(X_train.shape) print(X_test.shape) print(y_train.shape) print(y_predTotal.flatten()) #plt.plot(X_train, y_train, color='red', linewidth=2) #plt.plot(X_test, y_predTotal.flatten(), color='blue', linewidth=2,label='linear') plt.plot(list(range(0,len(y_predTotal.flatten()))), y_predTotal.flatten(), color='blue', linewidth=2,label='linear') plt.legend() #plt.savefig('images/linear.png') y_predTotal = np.reshape(y_predTotal, (tot_iter, X_test.shape[0])) #logging.info(" linears result: %s" % (y_predTotal)) logging.info(" linears predicted mean result: %s" % (y_predTotal.mean(axis=0))) # score= mean_squared_error(y_pred, y_test) # mae_score = mean_absolute_error(y_pred, y_test) # logging.info(" linear svr result: %f_%f" %(score, mae_score)) y_predTotal = np.array([]) idx_test = 0 for time in range(tot_iter): idx_test = idx_test+1 NUM_ESTIMATOR = 50 NUM_PREEPOCH = 150 NUM_BPEPOCH = 175 BATH_SIZE = 50 rf = build_RF(NUM_ESTIMATOR) rf.fit(X_train, y_train) y_pred = rf.predict(X_test) y_pred.reshape(1, X_test.shape[0]) # logging.info(" r fores predict result: %s" %(y_pred)) if (y_predTotal.shape[0] < 1): y_predTotal = y_pred else: y_predTotal = np.append(y_predTotal, y_pred, axis=0) y_predTotal = np.reshape(y_predTotal, (tot_iter, X_test.shape[0])) # plt.scatter(y_train, y_train, color='blue') # plt.show() #plt.clf() #plt.plot(X_train, y_train, color='red', linewidth=2) #plt.plot(X_test, y_predTotal.flatten(), color='blue', linewidth=2) plt.plot(list(range(0,len(y_predTotal.flatten()))), y_predTotal.flatten(), color='red', linewidth=2,label='random forest') plt.legend() #plt.savefig('images/rforest.png') logging.info(" r fore predicted mean result: %s" % (y_predTotal.mean(axis=0))) # score= mean_squared_error(y_pred, y_test) # mae_score = mean_absolute_error(y_pred, y_test) # logging.info(" randomforest result: %f_%f" %(score, mae_score)) # neural network y_predTotal = np.array([]) idx_test = 0 for time in range(tot_iter): idx_test = idx_test+1 nn_model = build_NN(data_dim) sc = StandardScaler() #train_sc = sc.fit_transform(X_train) #test_sc = sc.transform(X_test) #logging.info("X_Train: %s" %(np.percentile(X_train))) nn_model.fit(X_train, y_train, epochs=50,batch_size=BATH_SIZE) y_pred = nn_model.predict(X_test) #y_predact = sc.inverse_transform(y_pred) y_predact = y_pred y_pred.reshape(1, X_test.shape[0]) # logging.info(" neu predict result: %s" %(y_pred)) if (y_predTotal.shape[0] < 1): y_predTotal = y_predact else: y_predTotal = np.append(y_predTotal, y_predact, axis=0) y_predTotal = np.reshape(y_predTotal, (tot_iter, X_test.shape[0])) #plt.clf() #plt.plot(X_train, y_train, color='red', linewidth=2) #plt.plot(X_test, y_predTotal.flatten(), color='green', linewidth=2) plt.plot(list(range(0,len(y_predTotal.flatten()))), y_predTotal.flatten(), color='green', linewidth=2,label='Neural') plt.legend() #plt.savefig('images/neural.png') logging.info(" neu predicted mean result: %s" % (y_predTotal.mean(axis=0))) idx_test = 0 y_predTotal = np.array([]) for time in range(tot_iter): idx_test = idx_test+1 normal_AE = build_pre_normalAE( data_dim, X_train, epoch_pretrain=NUM_PREEPOCH, hidDim=[140, 280]) normal_AE.fit(X_train, y_train, epochs=NUM_BPEPOCH, batch_size=BATH_SIZE) y_pred = normal_AE.predict(X_test) y_pred.reshape(1, X_test.shape[0]) # logging.info(" ae predict result: %s" %(y_pred)) if (y_predTotal.shape[0] < 1): y_predTotal = y_pred else: y_predTotal = np.append(y_predTotal, y_pred, axis=0) y_predTotal = np.reshape(y_predTotal, (tot_iter, X_test.shape[0])) #plt.clf() #plt.plot(X_train, y_train, color='red', linewidth=2) #plt.plot(X_test, y_predTotal.flatten(), color='red', label='AE') plt.plot(list(range(0,len(y_predTotal.flatten()))), y_predTotal.flatten(), color='yellow', linewidth=2,label='auto encoder') plt.legend() #plt.savefig('images/ae.png') logging.info(" ae predicted mean result: %s" % (y_predTotal.mean(axis=0))) # score = mean_squared_error(y_pred, y_test) # mae_score = mean_absolute_error(y_pred, y_test) # logging.info(" normal ae result: %f_%f" %(score, mae_score)) # denoise AE idx_test = 0 y_predTotal = np.array([]) for time in range(tot_iter): idx_test = idx_test+1 denois_AE = build_pre_denoiseAE( data_dim, X_train, epoch_pretrain=NUM_PREEPOCH, hidDim=[140, 280]) denois_AE.fit(X_train, y_train, epochs=NUM_BPEPOCH, batch_size=BATH_SIZE) y_pred = denois_AE.predict(X_test) y_pred.reshape(1, X_test.shape[0]) # logging.info(" denoiseae predict result: %s" %(y_pred)) if (y_predTotal.shape[0] < 1): y_predTotal = y_pred else: y_predTotal = np.append(y_predTotal, y_pred, axis=0) y_predTotal = np.reshape(y_predTotal, (tot_iter, X_test.shape[0])) #plt.clf() #plt.plot(X_train, y_train, color='red', linewidth=2) #plt.plot(X_test, y_predTotal.flatten(), color='cyan', linewidth=2) plt.plot(list(range(0,len(y_predTotal.flatten()))), y_predTotal.flatten(), color='cyan', linewidth=2,label='Denoise AE') plt.legend() #plt.savefig('images/denoiseAE.png') logging.info(" denoiseae predicted mean result: %s" % (y_predTotal.mean(axis=0))) # score = mean_squared_error(y_pred, y_test) # mae_score = mean_absolute_error(y_pred, y_test) # logging.info(" denoise ae result: %f_%f" %(score, mae_score)) # rbf using sigmoid function, feature should be scaled to -1 and 1 idx_test = 0 for time in range(tot_iter): idx_test = idx_test+1 scaler = MinMaxScaler() X_train_rbm = scaler.fit_transform(X_train) rbm = build_RBM(NUM_BPEPOCH, NUM_PREEPOCH, batch_size=BATH_SIZE) rbm.fit(X_train_rbm, y_train) X_test_rbm = scaler.transform(X_test) y_pred = rbm.predict(X_test_rbm) y_predTotal = np.reshape(y_pred, (tot_iter, X_test.shape[0])) logging.info(" dbn predict result: %s" % (y_pred)) plt.plot(list(range(0,len(y_predTotal.flatten()))), y_predTotal.flatten(), color = '#808000', linewidth=2,label='dbn') plt.legend() plt.savefig('images/Combined') #plt.plot(X_test, y_predTotal.flatten(), color='magenta', linewidth=2) #plt.savefig('images/biLSTM.png') # score = mean_squared_error(y_pred, y_test) # mae_score = mean_absolute_error(y_pred, y_test) # logging.info(" dbn result: %f_%f" %(score, mae_score)) # Bi-directional LSTM # t = load_data(False) X_train_init = X_train X_test_init = X_test #X_train = np.reshape(X_train, ( 1,X_train.shape[1],X_train.shape[0])) X_test = np.reshape(X_test, (1,X_test.shape[1], X_test.shape[0])) print (X_train) print(X_test) data_dim = X_train.shape[2] timesteps = X_train.shape[0] biLSTM = build_BILSTM(timesteps, data_dim) biLSTM.fit(X_train, y_train, epochs=NUM_BPEPOCH, batch_size=BATH_SIZE) y_pred = biLSTM.predict(X_test) y_predTotal = np.reshape(y_pred, (tot_iter, X_test.shape[0])) logging.info("predicted BILSTM mean result: %s" % (y_predTotal.mean(axis=0))) #plt.clf() #plt.plot(X_train, y_train, color='red', linewidth=2) plt.plot(list(range(0,len(y_predTotal.flatten()))), y_predTotal.flatten(), color='yellow', linewidth=2,label='BILSTM') plt.legend() #plt.plot(X_test, y_predTotal.flatten(), color='magenta', linewidth=2) #plt.savefig('images/biLSTM.png') # score = mean_squared_error(y_pred, y_test) # mae_score = mean_absolute_error(y_pred, y_test) # logging.info(" bi-lstm result: %f_%f" %(score, mae_score)) # LSTM LSTM = build_LSTM(timesteps, data_dim) LSTM.fit(X_train, y_train, epochs=NUM_BPEPOCH, batch_size=BATH_SIZE) y_pred = LSTM.predict(X_test) #score = mean_squared_error(y_pred, y_test) #mae_score = mean_absolute_error(y_pred, y_test) y_predTotal = np.reshape(y_pred, (tot_iter, X_test.shape[0])) logging.info("predicted mean result: %s" % (y_predTotal.mean(axis=0))) #plt.clf() #plt.plot(X_train_plot, y_train, color='red', linewidth=2) #plt.plot(X_test_plot, y_predTotal.flatten(), color='blue', linewidth=2) plt.plot(list(range(0,len(y_predTotal.flatten()))), y_predTotal.flatten(), color='blue', linewidth=2,label='LSTM') plt.legend() #plt.savefig('images/LSTM.png') # CNN CNN = build_CNN(timesteps, data_dim) CNN.fit(X_train, y_train, epochs=NUM_BPEPOCH, batch_size=BATH_SIZE) y_pred = CNN.predict(X_test) plt.clf() plt.plot(X_train_plot, y_train, color='red', linewidth=2) plt.plot(X_test_plot, y_pred.flatten(), color='blue', linewidth=2) plt.savefig('CNN.png')
def mains(t): X_train, y_train, X_test = t[0], t[1], t[2] data_dim = X_train.shape[1] logging.info("X_Train: %s" % (t[0])) logging.info("Y_Train: %s" % (t[1])) logging.info("X_test: %s" % (t[2])) idx_test = 0 tot_iter = 10 y_predTotal = np.array([]) for time in range(tot_iter): idx_test = idx_test + 1 #t = load_data(True) """ linear_regression=build_LN() linear_regression.fit(X_train,y_train) y_pred = linear_regression.predict(X_test) logging.info(" linear predicted result: %s" %(y_pred)) """ linear_svr = build_SVR('linear', 1000) linear_svr.fit(X_train, y_train) y_pred = linear_svr.predict(X_test) y_pred.reshape(1, X_test.shape[0]) #logging.info(" %f linear predicted result: %s" %(idx_test, y_pred)) if (y_predTotal.shape[0] < 1): y_predTotal = y_pred else: y_predTotal = np.append(y_predTotal, y_pred, axis=0) #plt.scatter(y_train, y_train, color='blue') #plt.show() y_predTotal = np.reshape(y_predTotal, (tot_iter, X_test.shape[0])) logging.info(" linears predicted mean result: %s" % (y_predTotal.mean(axis=0))) #score= mean_squared_error(y_pred, y_test) #mae_score = mean_absolute_error(y_pred, y_test) #logging.info(" linear svr result: %f_%f" %(score, mae_score)) y_predTotal = np.array([]) idx_test = 0 for time in range(tot_iter): idx_test = idx_test + 1 NUM_ESTIMATOR = 75 NUM_PREEPOCH = 70 NUM_BPEPOCH = 150 BATH_SIZE = 100 rf = build_RF(NUM_ESTIMATOR) rf.fit(X_train, y_train) y_pred = rf.predict(X_test) y_pred.reshape(1, X_test.shape[0]) #logging.info(" r fores predict result: %s" %(y_pred)) if (y_predTotal.shape[0] < 1): y_predTotal = y_pred else: y_predTotal = np.append(y_predTotal, y_pred, axis=0) y_predTotal = np.reshape(y_predTotal, (tot_iter, X_test.shape[0])) #plt.scatter(y_train, y_train, color='blue') #plt.show() logging.info(" r fore predicted mean result: %s" % (y_predTotal.mean(axis=0))) #score= mean_squared_error(y_pred, y_test) #mae_score = mean_absolute_error(y_pred, y_test) #logging.info(" randomforest result: %f_%f" %(score, mae_score)) #neural network y_predTotal = np.array([]) idx_test = 0 for time in range(tot_iter): idx_test = idx_test + 1 nn_model = build_NN(data_dim) nn_model.fit(X_train, y_train, epochs=NUM_BPEPOCH, batch_size=BATH_SIZE) y_pred = nn_model.predict(X_test) y_pred.reshape(1, X_test.shape[0]) #logging.info(" neu predict result: %s" %(y_pred)) if (y_predTotal.shape[0] < 1): y_predTotal = y_pred else: y_predTotal = np.append(y_predTotal, y_pred, axis=0) y_predTotal = np.reshape(y_predTotal, (tot_iter, X_test.shape[0])) logging.info(" neu predicted mean result: %s" % (y_predTotal.mean(axis=0))) #score= mean_squared_error(y_pred, y_test) #mae_score = mean_absolute_error(y_pred, y_test) #logging.info(" neural network result: %f_%f" %(score, mae_score)) #AE idx_test = 0 y_predTotal = np.array([]) for time in range(tot_iter): idx_test = idx_test + 1 normal_AE = build_pre_normalAE(data_dim, X_train, epoch_pretrain=NUM_PREEPOCH, hidDim=[140, 280]) normal_AE.fit(X_train, y_train, epochs=NUM_BPEPOCH, batch_size=BATH_SIZE) y_pred = normal_AE.predict(X_test) y_pred.reshape(1, X_test.shape[0]) #logging.info(" ae predict result: %s" %(y_pred)) if (y_predTotal.shape[0] < 1): y_predTotal = y_pred else: y_predTotal = np.append(y_predTotal, y_pred, axis=0) y_predTotal = np.reshape(y_predTotal, (tot_iter, X_test.shape[0])) logging.info(" ae predicted mean result: %s" % (y_predTotal.mean(axis=0))) #score = mean_squared_error(y_pred, y_test) #mae_score = mean_absolute_error(y_pred, y_test) #logging.info(" normal ae result: %f_%f" %(score, mae_score)) #denoise AE idx_test = 0 y_predTotal = np.array([]) for time in range(tot_iter): idx_test = idx_test + 1 denois_AE = build_pre_denoiseAE(data_dim, X_train, epoch_pretrain=NUM_PREEPOCH, hidDim=[140, 280]) denois_AE.fit(X_train, y_train, epochs=NUM_BPEPOCH, batch_size=BATH_SIZE) y_pred = denois_AE.predict(X_test) y_pred.reshape(1, X_test.shape[0]) #logging.info(" denoiseae predict result: %s" %(y_pred)) if (y_predTotal.shape[0] < 1): y_predTotal = y_pred else: y_predTotal = np.append(y_predTotal, y_pred, axis=0) y_predTotal = np.reshape(y_predTotal, (tot_iter, X_test.shape[0])) logging.info(" denoiseae predicted mean result: %s" % (y_predTotal.mean(axis=0))) #score = mean_squared_error(y_pred, y_test) #mae_score = mean_absolute_error(y_pred, y_test) #logging.info(" denoise ae result: %f_%f" %(score, mae_score)) # rbf using sigmoid function, feature should be scaled to -1 and 1 idx_test = 0 for time in range(tot_iter): idx_test = idx_test + 1 scaler = MinMaxScaler() X_train_rbm = scaler.fit_transform(X_train) rbm = build_RBM(NUM_BPEPOCH, NUM_PREEPOCH, batch_size=BATH_SIZE) rbm.fit(X_train_rbm, y_train) X_test_rbm = scaler.transform(X_test) y_pred = rbm.predict(X_test_rbm) logging.info(" dbn predict result: %s" % (y_pred)) #score = mean_squared_error(y_pred, y_test) #mae_score = mean_absolute_error(y_pred, y_test) #logging.info(" dbn result: %f_%f" %(score, mae_score)) #Bi-directional LSTM #t = load_data(False) """