Wrong super call in RecurrentSequential __init__

[JulianStj]

In the init function for RecurrentSequential, it calls

super(RecurrentModel, self).init(**kwargs)

But RecurrentSequential is inherited from RecurrentModel. The call should be

super(RecurrentSequential, self).init(**kwargs)

But the arguments don't match. I'm not quite sure how to fix this the "right" way, but calling the base class constructor is not correct for RecurrentSequential

[kklemon]

I also encountered this problem while trying to run the Nesting RecurrentSequentials demo. Indeed that super call in RecurrentSequential looks suspicious and should probably call RecurrentModel's constructor instead of itself, but this also wouldn't really make sense.

In opposite to Keras' Sequential model which updates itself with every add(...) call and hence doesn't need to be build in the end explicitly, RecurrentSequential doesn't and thus you need to call build(...) on the instance before you can access the internal model.

For example the mentioned demo would go down to

rnn1 = RecurrentSequential()
rnn1.add(LSTMCell(16, input_dim=30))
rnn1.add(Dense(20))
rnn1.build((None, None, 30))  # batch len, seq len, seq dim

rnn1_cell = rnn1.get_cell()

...

then.