class cGAN_LSTMConv3Architecture ill defined

[frMitola]

class cGAN_LSTMConv3Architecture (here the link)
has pool_and_stride = round((x.shape[1] + 1) / (self._seq_len + 1)) at line 422. This causes dimension mismatch for data with dimension different from a power of 2 (i tried with data with dimension 24 and 128, the former doesn't work, the latter works).
I suggest to change with pool_and_stride = math.ceil((x.shape[1] + 1) / (self._seq_len + 1)) from class cGAN_Conv4Architecture, which works fine for me.