Hi professor,
I have used this method to compute the 'removed_trend' and 'removed_std', but the result was not same as you. Can you tell me your method.
my method:
df['mean'] =( df.groupby(['day_of_week','hour','minute'])['pickups'].transform('mean'))

My email: [email protected]

Hi, Professor, when i read your paper, i have an idea that is to add weather(eg precipitation) into the model, then I have done something to the DeepJMQR model as following:

weather data processing

weather_feats = weather['precipitation'].values[:,np.newaxis]

---------------- Build lags

weather_feats[:-NUM_LAGS,:] = weather_feats[NUM_LAGS:,:]
weather_feats_train = weather_feats[:i_train,:]
weather_feats_val = weather_feats[i_train:i_val,:]
weather_feats_test = weather_feats[i_val:i_test,:]
#深度学习模型for mean

---------------- Independent deep learning models for mean and quantiles

def tilted_loss(q,y,f):
e = (y-f)
# The term inside k.mean is a one line simplification of the first equation
return K.mean(q*e + K.clip(-e, K.epsilon(), np.inf), axis=-1)

def build_model(loss="mse", num_outputs=1):
model = Sequential()

model.add(ConvLSTM2D(filters=100, kernel_size=(3, 3),
               input_shape=(None, nx, ny, 1),
               padding='same', return_sequences=False))
model.add(BatchNormalization())
model.add(Dropout(0.75))

model.add(Dense(units=num_outputs, 
                kernel_regularizer=regularizers.l2(0.0001),
         ))
model.add(Activation("linear"))

model.compile(loss=loss, optimizer='nadam')

return model

print ("\nTraining NN for mean...")

checkpoint best model

checkpoint = ModelCheckpoint("weights_mean.best.hdf5", monitor='val_loss', verbose=1, save_best_only=True, mode='min')

fit model to the mean

model_mean = build_model(loss="mse", num_outputs=1)
model_mean.summary()
model_mean.fit(
[X_train[:,:,:,:,np.newaxis],weather_feats_train[:,:]],
y_train[:,:,:,np.newaxis],
batch_size=128,
epochs=N_EPOCHS, #100
validation_data=([X_val[:,:,:,:,np.newaxis],weather_feats_val[:,:]],y_val[:,:,:,np.newaxis]),
callbacks=[checkpoint],
verbose=0)

But there is a error:ValueError: Error when checking model input: the list of Numpy arrays that you are passing to your model is not the size the model expected. Expected to see 1 array(s), but instead got the following list of 2 arrays

I have tried to solve this problem, but failed. Could you give me a good idea?
Thank you very much！