Probably wrong indexing about indrnn

[Luonic]

@batzner i was just reading your implementation, not actually running code. Thank you for pointing out that input is 2D because I missed that)
I am going to give a try to your IndRNN for character-level sentiment analysis and will notify if any problems will occur

[batzner]

The IndRNNCell code is an adaptation from TensorFlow's BasicRNNCell, which also uses inputs_shape[1]. I think it doesn't make a difference whether you use 1 or -1, because the inputs must be 2-dimensional as specified a few lines above with self.input_spec = base_layer.InputSpec(ndim=2).

Did you get the ValueError while running a script that uses the IndRNNCell? If so, could you provide the code, so that I can try to reproduce it?