[Python] What is the rule to know how many LSTM cells and how many units in each LSTM cell do...

I know that a LSTM cell has a number of ANNs inside.

But when defining the hidden layer for the same problem, I have seen some people using only 1 LSTM cell and others use 2, 3 LSTM cells like this -

model = Sequential()
model.add(LSTM(256, input_shape=(n_prev, 1), return_sequences=True))
model.add(Dropout(0.3))
model.add(LSTM(128, input_shape=(n_prev, 1), return_sequences=True))
model.add(Dropout(0.3))
model.add(LSTM(64, input_shape=(n_prev, 1), return_sequences=False))
model.add(Dropout(0.3))
model.add(Dense(1))
model.add(Activation('linear'))

1. Is there any rule as to how many LSTM cells you should take? Or its just manual experimenting?
2. Another question following this is, how many units you should take in an LSTM cell. Like some people take 256, some take 64 for the same problem.

Continue reading...