history = model.fit(
{
'tempInput': xtrain[0],
'humdInput': xtrain[1],
'presInput': xtrain[2]
}, {'out': ytrain},
validation_split=0.2,
epochs=300,
batch_size=400,
verbose=2,
callbacks=[checkpoint, earlystop, board])
return history
x, y = getXY('./data/raw.csv',
features=['Temperature', 'Td dew point', 'StnPres'])
xtrain, xtest, ytrain, ytest = train_test_split(x, y, test_size=0.25)
print(xtrain.shape, ytrain.shape, xtest.shape, ytest.shape)
xtrainin = [
xtrain[:, :, 0][:, :, np.newaxis], xtrain[:, :, 1][:, :, np.newaxis],
xtrain[:, :, 2][:, :, np.newaxis]
]
xtestin = [
xtest[:, :, 0][:, :, np.newaxis], xtest[:, :, 1][:, :, np.newaxis],
xtest[:, :, 2][:, :, np.newaxis]
]
print('xtrain after sep feature: ', xtrainin[0].shape, xtrainin[1].shape,
xtrainin[2].shape)
hist = CNNsepFeature(xtrainin, ytrain) # start training
#callbacks
checkpoint = ModelCheckpoint('./models/LSTMmodel.hdf5',
monitor='val_loss',
save_best_only=True,
verbose=1)
earlystop = EarlyStopping(monitor='val_loss', patience=10)
board = TensorBoard(log_dir='./logs')
model.compile(loss='mse', optimizer='adam')
model.fit(xtrain,
ytrain,
epochs=300,
batch_size=256,
verbose=2,
validation_split=0.2,
callbacks=[checkpoint, earlystop])
return model
x, y = getXY('./data/raw.csv')
xtrain, xtest, ytrain, ytest = train_test_split(x, y, test_size=0.25)
print(xtrain.shape, ytrain.shape, xtest.shape, ytest.shape)
hist = LSTMmodel(xtrain, ytrain) # start training
m = load_model('./models/LSTMmodel.hdf5') # load best model
ypred = m.predict(xtest)
print('mae: ', mae(ytest, ypred))