Hi,
First of all, thanks for your unified framework. I am doing a master thesis and I aim to applying your metric to synthetic text data.
In implementing your code by means of 'metrics.py', I came across something what I think is an error:
You only compute the recall entropy of the reference dataset containing the "real features" instead of what you do in the notebook, where you rightly subtract the entropy of "fake features" before returning the value:
If it is indeed a bug, can you fix that in your code?
Update: I now see that the same holds for recall cross entropy and precision cross entropy.
Hi,
First of all, thanks for your unified framework. I am doing a master thesis and I aim to applying your metric to synthetic text data.
In implementing your code by means of 'metrics.py', I came across something what I think is an error:
You only compute the recall entropy of the reference dataset containing the "real features" instead of what you do in the notebook, where you rightly subtract the entropy of "fake features" before returning the value:
If it is indeed a bug, can you fix that in your code?
Update: I now see that the same holds for recall cross entropy and precision cross entropy.