def get_callbacks(name):
return [
modeling.EpochDots(),
tf.keras.callbacks.EarlyStopping(monitor='val_categorical_crossentropy',
patience=50, restore_best_weights=True),
# tf.keras.callbacks.TensorBoard(log_dir/name, histogram_freq=1),
# tf.keras.callbacks.ModelCheckpoint(filepath=checkpoint_dir + "/{}/cp.ckpt".format(name),
# verbose=0,
# monitor='val_sparse_categorical_crossentropy',
# save_weights_only=True,
# save_best_only=True),
tf.keras.callbacks.ReduceLROnPlateau(monitor='val_categorical_crossentropy',
factor=0.1, patience=10, verbose=0, mode='auto',
min_delta=0.0001, cooldown=0, min_lr=0),
]
def get_callbacks(self,pat=10):
return [modeling.EpochDots(),
tf.keras.callbacks.EarlyStopping(monitor='val_loss', patience=pat)]
z_test=dtest[:,0:5]
z_scaler=procs.StandardScaler()
z_scaler.fit(z_test)
z_test_norm=z_scaler.transform(z_test)
raw_dataset_test=pd.DataFrame(z_test_norm,columns=names)
w_test=dtest[:,5]
test_low=w_test<5000
w_test=w_test[test_low]
model=build_model()
EPOCHS=1000
for j in range(0,3):
early_stop=keras.callbacks.EarlyStopping(monitor='loss',patience=20)
history=model.fit(raw_dataset,label_dataset,epochs=EPOCHS,validation_split=0.2,
verbose=0,callbacks=[early_stop,tfmod.EpochDots()])
loss,mae,mse=model.evaluate(raw_dataset,label_dataset,verbose=2)
model.summary()
test_predict=model.predict(raw_dataset_test).flatten()
M5=np.size(test_predict)
test_predict=test_predict.reshape(K,1)
test_predict=target_scaler.inverse_transform(test_predict)
test_predict=test_predict.T[0]
print('Datos predecidos (M): {0}'.format(np.shape(test_predict)[0]))
# para la Regression, solo bajas energias
df=pd.DataFrame()
test_out=test_predict[test_low]
df['RealData']=w_test