Thanks for the wonderful code. I looked into your code and have run the Keras program and was able to replicate your results. However, I have a doubt regarding the channel layer. I will be extremely grateful if you clear this doubt.
In the channel layer implementation in the file channel.py
[1] Line L43, channel_matrix = self.activation(self.kernel), are the softmax activation applied across the channel_matrixrow-wise or column-wise ?
Can you please clarify selecting one over the other? I thought since we are using the confusion matrix as the initialization step, it should be row stochastic?
[2] Line L57, is the dot product of the 'baseline model' outputs done over the 'rows' or 'columns' of the channel matrix?
Hi
Thanks for the wonderful code. I looked into your code and have run the Keras program and was able to replicate your results. However, I have a doubt regarding the channel layer. I will be extremely grateful if you clear this doubt.
In the channel layer implementation in the file
channel.py[1] LineL43,channel_matrix = self.activation(self.kernel), are thesoftmaxactivation applied across thechannel_matrixrow-wiseorcolumn-wise? Can you please clarify selecting one over the other? I thought since we are using theconfusion matrixas the initialization step, it should be row stochastic?[2] Line
L57, is the dot product of the 'baseline model' outputs done over the 'rows' or 'columns' of the channel matrix?Thanks in advance! Devraj