Olivier TERRIER

<PAD> tags should be filtered out from the output of the Tagger

17

In a sequence labelling scenario, the internal tag can be present in the output of the Tagger.tag() method. As they are internal they should probably been filtered out I would...

Cannot train a new NER model on cpu-only machine

1

Hello I'm trying to train a TPipeline initialized with a training_config with training_config['gpu'] set to False on a cpu machine (no NVIDIA driver) But the call to the train method...

... based on [UD Arabic NYUAD](https://universaldependencies.org/treebanks/ar_nyuad/index.html) which has a commercial friendly license CC BY-SA 4.0 compared to the [Arabic-PADT UD](https://universaldependencies.org/treebanks/ar_padt/index.html) which is under CC BY-NC-SA 3.0 NYUAD is also much...

For KB concepts the given "valueName" of statements should be the english label instead of the wikipedia page title

4

I find it counter intuitive to set the valueName of statement properties to the title of the wikipedia page for KB concepts Let's take some examples With "human brain" (Q492038)...

Add an option to make a warm-up of lower/upper KB databases at startup

1

Having most of lmdb pages loaded in memory speed up a lot the processing even though it requires a lot of RAM. Maybe we could add an option (in the...

Sometimes entity fishing returns "Invalid id or excluded via caching" as rawName of preferredForm

1

When you call the /service/kb/concept/ endpoint with a valid QID, sometimes the rawName and/or PreferredForm in the response is "Invalid id or excluded via caching" Like for example: ``` {...

Building such an index for all the statements is not manageable due to the explosion of the wikidata recently (more than 90M items) but it could be interesting to restrict...

Make the NGRAM_LENGTH parameter settable at query level

1

Hi, today the maximum length of ngrams used to recognize mentions is hard coded to 6 This is generally sufficient but sometimes we would like to go a little bit...

New ranking model based on BLINK biencoder/crossencoder approach?

2

Hi Patrice, Facebook recently released BLINK https://github.com/facebookresearch/BLINK https://arxiv.org/abs/1911.03814 This is a novel approach for entity linking based on BERT transformers that apparently outperform even the already impressive DeepType (Raiman &...

explain_prediction with OneVsRestClassifier and SVC

1

Hi Thx for this great package It looks like there is an error when you use OneVsRestClassifier with SVCClassifier I have this error ``` File "/home/olivier/.pyenv/versions/suggest/lib/python3.6/site-packages/eli5/sklearn/explain_prediction.py", line 228, in test_explain_prediction_libsvm_linear...