def model_fit(train_X, train_y, test_X, test_y):
model_save_file = ''
model_save = {}
test_regressor = ['GBDT']
regressors = {
'GBDT': gbdt_classifier,
}
print('reading training and testing data...')
# train_x, train_y, test_x, test_y = read_data()
for regressor in test_regressor:
print('******************* %s ********************' % regressor)
start_time = time.time()
model = regressors[regressor](train_X, train_y)
print('training took %fs!' % (time.time() - start_time))
t0 = time.strftime('%Y%m%d%H%M%S', time.localtime(time.time()))
pickle.dump(model, open('../model/GBDT_model_%s.h5' % t0, 'wb'))
predict = model.predict(test_X)
# print('test_y: {}\npredict: {}'.format(test_y, predict))
# score = model.score(test_x, test_y)
calMetrix(__file__, predict, test_y)
plt.figure()
plt.plot(np.arange(len(predict)), test_y, 'go-', label='true value')
plt.plot(np.arange(len(predict)),
predict,
'ro-',
label='predict value')
plt.title('%s' % regressor)
plt.legend()
plt.show()
def model_fit(train_X, train_y, test_X, test_y):
# define model
model = Sequential()
# model.add(TimeDistributed(cnn))
model.add(TimeDistributed(Convolution1D(128, 4, border_mode='same'), input_shape=train_X.shape[1:]))
model.add(TimeDistributed(MaxPooling1D(pool_size=2)))
model.add(TimeDistributed(Flatten()))
model.add(LSTM(128, return_sequences=True, name="lstm_layer0"))
model.add(LSTM(128, return_sequences=False, name="lstm_layer1"))
# model.add(LSTM(100, return_sequences=True, name="lstm_layer2"))
model.add(Dense(output_dim, activation='sigmoid'))
# model.add(GlobalAveragePooling1D(name="global_avg"))
model.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy'])
# %%
model.fit(train_X, train_y,batch_size=batch_size, nb_epoch=nb_epoch, verbose=0, validation_data=(test_X, test_y))
test_y = test_y.reshape(test_y.size, 1)
predict_y = model.predict(test_X)
predict_y = predict_y.reshape(predict_y.size, 1)
predict_y = classifyRes(predict_y)
calMetrix(__file__, predict_y, test_y)
t0 = time.strftime('%Y%m%d%H%M%S', time.localtime(time.time()))
model.save('../model/CNN_LSTM_model_%s.h5'%t0)
def model_fit(train_X, train_y, test_X, test_y):
# train_X, train_y, test_X, test_y = read_data()
outdim = train_y.shape[1]
validation_number = int(0.9 * train_X.shape[0])
validation_X = train_X[validation_number:]
validation_y = train_y[validation_number:]
train_X = train_X[:validation_number]
train_y = train_y[:validation_number]
model = Sequential()
model.add(LSTM(100, input_shape=(train_X.shape[1], train_X.shape[2]), return_sequences=True, dropout=0.25))
# print(model.layers)
model.add(LSTM(100, return_sequences=True))
model.add(LSTM(100, return_sequences=True))
model.add(LSTM(50, return_sequences=False))
# model.add(Dense(outdim))
model.add(Dense(outdim, activation='sigmoid'))
# model.add(Dense(outdim, activation='softmax'))
# sigmoid and softmax are activation functions used by the neural network output layer
# for binary discrimination and multi-class discrimination
# binary cross-entropy and categorical cross-entropy are corresponding loss functions
# model.add(Activation('linear'))
# model.compile(loss='mse', optimizer='rmsprop')
model.compile(optimizer='adam',
loss='binary_crossentropy',
# loss='categorical_crossentropy',
metrics=['accuracy']
)
# fit network
history = model.fit(train_X, train_y, epochs=100, batch_size=72, validation_data=(validation_X, validation_y),
verbose=0,
shuffle=True)
# summarize performance of the model
# scores = model.evaluate(train_X, train_y, verbose=0)
# print(model.metrics_names)
# print("model loss: %.2f%%" % (scores*100))
# plot history
# plt.plot(history.history['loss'], label='train')
# plt.plot(history.history['val_loss'], label='validation')
# plt.legend()
# plt.grid()
# plt.savefig('loss.png')
# plt.clf()
np.savetxt('loss.csv', history.history['loss'])
np.savetxt('val_loss.csv', history.history['val_loss'])
t0 = time.strftime('%Y%m%d%H%M%S', time.localtime(time.time()))
model.save('../model/LSTM_model_%s.h5'%t0)
# make a prediction
y_pre = model.predict(test_X)
# yhat = yhat.reshape(train_y.shape[1])
print(y_pre.shape)
y_pre = classifyRes(y_pre)
calMetrix(__file__, y_pre, test_y)
# inv_yhat = yhat
# inv_test_y = test_y
np.savetxt('forecast.csv', y_pre)
np.savetxt('actual.csv', test_y)