make Keras 'None' batch size unchanged, using tf.scatter_nd

Question

I need to input a pooling module to the LSTM decoder, and I\'m constructing this using a custom layer with the encoder LSTM states and Keras Input layer as inputs. In this custo