Patrice Lopez

Hi Luca ! Shouldn't the tmp directory be defined in `resource-registry.json`? because it is a typical library-level resource (like the default download path). `self.registry` is then already available in the...

The PR is doing two different things: - It adds an output path parameter to the application scripts indicating where to save the models, instead of using the default one...

Thank you @de-code ! I don't think it was discussed before. This is correct, the preprocessor is initialized on the whole training set in the n-fold scenario. At first glance,...

Thank you Olivier! Do you have an example with `` in the output of `Tagger.tag()`? In principle we have: ``` the_tags = list(zip(tokens, tags)) ``` so, if I am not...

Which architecture did you use? Normally there should not be any in list of labelling, so there is another source problem to fix if it's the case!

> The preprocessor list of tag includes ``. Is this normal? yes it is for padding the label vector. Every "channels" will have a `` stuff at index 0 in...

`` cannot pop-out normally when predicting because everything is cut based on the length of the token sequence, which is what I mentioned [above](https://github.com/kermitt2/delft/issues/64#issuecomment-562910590). If we have a `` in...

> While the second makes sense (it's the result of padding not to leave sequence with only one element) the first one does not... If there's a batch with one...

Hi @Protossnam and thanks! That's a very good point, I will prepare some uncased models for NER (with the uncased embedding data).

@Protossnam results are excellent, great ! I was not expecting such good results with lowercase. I've only downloaded the uncased embeddings, but I have not yet started to do some...