def on_epoch_begin(self, epoch, logs={}):
lr = K.get_value(model.optimizer.lr)
print("current learning rate is {}".format(lr))
pred = model.predict(x_test)
predict_all = pred.flatten()
truth_all = y_test.flatten()
mape = eleceval.calcMAPE(predict_all, truth_all)
mae = eleceval.calcMAE(predict_all, truth_all)
mse = eleceval.calcMSE(predict_all, truth_all)
rmse = eleceval.calcRMSE(predict_all, truth_all)
r_2 = eleceval.r2(predict_all, truth_all)
print("After %d training step(s),"
"on test data MAPE = %.4f,MAE = %.4f,MSE = %.4f,RMSE = %.4f,R2 = %.4f"\
% (epoch, mape,mae,mse,rmse,r_2))
if mape <= self.mape_flag:
self.mape_flag = mape
#两个ndarray列合并
#y_con = np.concatenate((truth_all, predict_all), axis=1)
truth_all_reshape = np.reshape(truth_all, [-1, 1])
predict_all_reshape = np.reshape(predict_all, [-1, 1])
y_con = np.concatenate((truth_all_reshape, predict_all_reshape),
axis=1)
#输出真实值和预测值
y_out = pd.DataFrame(y_con, columns=["true_data", "pre_data"])
y_out.to_csv('./result_929_3cluster/steps=%d-MAPE=%.4f,MAE = %.4f,MSE = %.4f,RMSE = %.4f,R2 = %.4f.csv'\
% (epoch,mape,mae,mse,rmse,r_2))
def on_epoch_begin(self, epoch, logs={}):
lr = K.get_value(model.optimizer.lr)
print("current learning rate is {}".format(lr))
pred = model.predict({'input_layer':x_test, 'aux_input':aux_x_test})
pred = np.array(pred)
predict_all = pred.flatten()
truth_all = np.array(y_test)
truth_all = truth_all.flatten()
maape = eleceval.calcMaape(predict_all,truth_all)
mape = eleceval.calcMAPE(predict_all,truth_all)
mae = eleceval.calcMAE(predict_all,truth_all)
mse = eleceval.calcMSE(predict_all,truth_all)
rmse = eleceval.calcRMSE(predict_all,truth_all)
r_2 = eleceval.r2(predict_all,truth_all)
print("After %d training step(s),"
"on test data MAAPE = %.4f,MAPE = %.4f,MAE = %.4f,MSE = %.4f,RMSE = %.4f,R2 = %.4f"\
% (epoch, maape,mape,mae,mse,rmse,r_2))
if maape <= self.maape_flag:
self.maape_flag = maape
#concatenate two arrays
truth_all_reshape=np.reshape(truth_all,[-1,1])
predict_all_reshape=np.reshape(predict_all,[-1,1])
y_con = np.concatenate((truth_all_reshape, predict_all_reshape), axis=1)
#Print true value and predicted value
y_out = pd.DataFrame(y_con, columns=["true_data","pre_data"])
y_out.to_csv('steps=%d-MAAPE=%.4f.csv'\
% (epoch, maape))
def on_epoch_begin(self, epoch, logs={}):
# print(f'x_test[0:1] = {x_test[0:1]}.')
# print(f'y_test[0:1] = {y_test[0:1]}.')
# print(f'pred = {self.model.predict(x_test[0:1])}.')
lr = K.get_value(model.optimizer.lr)
print("current learning rate is {}".format(lr))
pred = model.predict(x_test)
pred = pred[:, 239, :]
print('pred.shape:')
print(pred.shape)
print('y_test.shape:')
print(y_test.shape)
predict_all = pred.flatten()
truth_all = y_test.flatten()
'''
pred_train = model.predict(x_train)
predict_train = pred_train.flatten()
truth_train = y_train.flatten()
mape_train = eleceval.calcMAPE(predict_train,truth_train)
mae_train = eleceval.calcMAE(predict_train,truth_train)
mse_train = eleceval.calcMSE(predict_train,truth_train)
rmse_train = eleceval.calcRMSE(predict_train,truth_train)
r_2_train = eleceval.r2(predict_train,truth_train)
print("After %d training step(s),"
"on test data MAPE_train = %.4f,MAE_train = %.4f,MSE_train = %.4f,RMSE_train = %.4f,R2_train = %.4f"\
% (i*10, mape_train,mae_train,mse_train,rmse_train,r_2_train))
'''
a1 = pred
a2 = y_test
a = pinball_score(a1, a2, 9)
mape = eleceval.calcMAPE(predict_all, truth_all)
mae = eleceval.calcMAE(predict_all, truth_all)
mse = eleceval.calcMSE(predict_all, truth_all)
rmse = eleceval.calcRMSE(predict_all, truth_all)
r_2 = eleceval.r2(predict_all, truth_all)
print("After %d training step(s),"
"on test data MAPE = %.4f,MAE = %.4f,MSE = %.4f,RMSE = %.4f,R2 = %.4f,Pinball loss = %.4f"\
% (epoch, mape, mae, mse, rmse, r_2, a))
if a <= self.pinball_flag:
self.pinball_flag = a
# 两个ndarray列合并
# y_con = np.concatenate((truth_all, predict_all), axis=1)
truth_all_reshape = np.reshape(y_test[:, 0], [-1, 1])
predict_all_reshape = pred
y_con = np.concatenate((truth_all_reshape, predict_all_reshape),
axis=1)
# 输出真实值和预测值0
y_out = pd.DataFrame(y_con)
y_out.to_csv('result1h/steps=%d-pinball=%.4f.csv' % (epoch, a))
def on_epoch_begin(self, epoch, logs={}):
#print(f'x_test[0:1] = {x_test[0:1]}.')
#print(f'y_test[0:1] = {y_test[0:1]}.')
#print(f'pred = {self.model.predict(x_test[0:1])}.')
lr = K.get_value(model.optimizer.lr)
print("current learning rate is {}".format(lr))
pred = model.predict({'input_layer': x_test, 'aux_input': aux_x_test})
#print('pred.shape:')
#print(pred.shape)
#print('y_test.shape:')
#print(y_test.shape)
predict_all = pred.flatten()
truth_all = y_test.flatten()
'''
pred_train = model.predict(x_train)
predict_train = pred_train.flatten()
truth_train = y_train.flatten()
mape_train = eleceval.calcMAPE(predict_train,truth_train)
mae_train = eleceval.calcMAE(predict_train,truth_train)
mse_train = eleceval.calcMSE(predict_train,truth_train)
rmse_train = eleceval.calcRMSE(predict_train,truth_train)
r_2_train = eleceval.r2(predict_train,truth_train)
print("After %d training step(s),"
"on test data MAPE_train = %.4f,MAE_train = %.4f,MSE_train = %.4f,RMSE_train = %.4f,R2_train = %.4f"\
% (i*10, mape_train,mae_train,mse_train,rmse_train,r_2_train))
'''
mape = eleceval.calcMAPE(predict_all, truth_all)
mae = eleceval.calcMAE(predict_all, truth_all)
mse = eleceval.calcMSE(predict_all, truth_all)
rmse = eleceval.calcRMSE(predict_all, truth_all)
r_2 = eleceval.r2(predict_all, truth_all)
print("After %d training step(s),"
"on test data MAPE = %.4f,MAE = %.4f,MSE = %.4f,RMSE = %.4f,R2 = %.4f"\
% (epoch, mape,mae,mse,rmse,r_2))
if mape <= self.mape_flag:
self.mape_flag = mape
#两个ndarray列合并
#y_con = np.concatenate((truth_all, predict_all), axis=1)
truth_all_reshape = np.reshape(truth_all, [-1, 1])
predict_all_reshape = np.reshape(predict_all, [-1, 1])
y_con = np.concatenate((truth_all_reshape, predict_all_reshape),
axis=1)
#输出真实值和预测值
y_out = pd.DataFrame(y_con, columns=["true_data", "pre_data"])
y_out.to_csv(
'/media/dzf/data/data/MSFNN_train_result/0_test_causal_3datasplit_seqoutput/steps=%d-MAPE=%.4f.csv'
% (epoch, mape))
def on_epoch_begin(self, epoch, logs={}):
#print(f'x_test[0:1] = {x_test[0:1]}.')
#print(f'y_test[0:1] = {y_test[0:1]}.')
#print(f'pred = {self.model.predict(x_test[0:1])}.')
lr = K.get_value(model.optimizer.lr)
print("current learning rate is {}".format(lr))
pred = model.predict(x_test)
predict_all = pred.flatten()
truth_all = y_test.flatten()
mape = eleceval.calcMAPE(predict_all, truth_all)
mae = eleceval.calcMAE(predict_all, truth_all)
mse = eleceval.calcMSE(predict_all, truth_all)
rmse = eleceval.calcRMSE(predict_all, truth_all)
r_2 = eleceval.r2(predict_all, truth_all)
pinball = pinball_score(y_test, pred, 9)
prediction = pred
truth = y_test[:, 0]
winkler40 = winkler_score(truth, prediction[:, 2], prediction[:, 6],
0.4)
winkler80 = winkler_score(truth, prediction[:, 0], prediction[:, 8],
0.8)
pcip40 = PCIP(truth, prediction[:, 2], prediction[:, 6], 0.4)
pcip80 = PCIP(truth, prediction[:, 0], prediction[:, 8], 0.8)
ace40 = pcip40 - 0.4
ace80 = pcip80 - 0.8
ecr40 = ECR(prediction[:, 2],
prediction[:, 6]) / (np.max(truth) - np.min(truth))
ecr80 = ECR(prediction[:, 0],
prediction[:, 8]) / (np.max(truth) - np.min(truth))
print("After %d training step(s),"
"on test data MAPE = %.4f,MAE = %.4f,MSE = %.4f,RMSE = %.4f,R2 = %.4f,piball_loss = %.4f,winkler40 = %.4f,winkler80 = %.4f,pcip40 = %.4f,pcip80 = %.4f,ace40 = %.4f,ace80 = %.4f,ecr40 = %.4f,ecr80 = %.4f"\
% (epoch, mape,mae,mse,rmse,r_2,pinball,winkler40,winkler80,pcip40,pcip80,ace40,ace80,ecr40,ecr80))
if pinball < self.pinball_flag:
self.pinball_flag = pinball
#两个ndarray列合并
#y_con = np.concatenate((truth_all, predict_all), axis=1)
truth_all_reshape = np.reshape(truth_all, [-1, 1])
predict_all_reshape = np.reshape(predict_all, [-1, 1])
y_con = np.concatenate((truth_all_reshape, predict_all_reshape),
axis=1)
#输出真实值和预测值
y_out = pd.DataFrame(y_con, columns=["true_data", "pre_data"])
y_out.to_csv('./result_929_3cluster/steps=%d-MAPE=%.4f,MAE = %.4f,MSE = %.4f,RMSE = %.4f,R2 = %.4f,piball_loss = %.4f,winkler40 = %.4f,winkler80 = %.4f,pcip40 = %.4f,pcip80 = %.4f,ace40 = %.4f,ace80 = %.4f,ecr40 = %.4f,ecr80 = %.4f.csv'\
% (epoch,mape,mae,mse,rmse,r_2,pinball,winkler40,winkler80,pcip40,pcip80,ace40,ace80,ecr40,ecr80))