Matthew Middlehurst

It would be good to cover the partials if possible, but I'm not going to demand 100%. Keep in mind testing is more than coverage, there probably not a lot...

Hi are there any issues with this? It is still a draft.

I was more referring to this image.  I would reduce the name of channels from 5 used in the notebook to 3, may be confusing to have the same...

This is still incorrect, looks like the series length and number of channels are swapped.

Yes probably best to transpose the matrix array. Does the documentation mention the dimensions of the output?

I think maybe just leave the function as it is for now and input a univariate series (1d numpy array). The current image does not display 10 time points.

Running this now. will take a look and get back to you.

So, it is crashing on `print(afc.predict(data))` for me. But this makes sense since it is trained on a dataset with 20 series length then `data` has 200 series length. Thats...

Yeah we can fix this on our side. Though may take a while to make a release, since we have a lot of other changes currently that we don't want...

Though saying that, it runs the below without issues (I think). Maybe we can configure it to accept unequal length relatively easily. We really need to revamp this code 🙂....