PointerLSTM.py is creating the following issues while running in Keras version 2.2

[mmmaniva]

1. AttributeError: 'super' object has no attribute 'get_constants' - I tried creating a function similar to 'get_initial_states', after that the error resolved.
2. AttributeError: 'PointerLSTM' object has no attribute 'preprocess_input' - I tried creating a function similar to 'get_initial_states', after that the error resolved.
3. AttributeError: 'super' object has no attribute 'step' - but this am not sure how to solve it in Keras version 2.2
   could you please suggest a way to fix the above issues?

Thanks

[Adamwgoh]

Hi, I faced a similar issue as well. I'm new to Keras so I'm not exactly sure what's the situation between pre-Keras2.0 and Keras2.2 (The one I used). However, here are some pointers that I found:

• Point 1 and 2 can be resolved if you inherit Recurrent instead of LSTM. This is because Recurrent is now a legacy layer class object.
• For Point3, Recurrent legacy code only throws a NotImplementedError. I am still trying to understand what was expected from calling LSTM's step function, and if there is an equivalent, that may be the function to call

[mmmaniva]

Thanks a lot for the response @Adamwgoh

[marcvoegele-fhnw]

@Adamwgoh: With using Recurrent. There comes another issue:
('Keyword argument not understood:', 'units')