2022-02-11-ner_deid_subentity_fr

[jsl-models]

No description provided.

[vkocaman]

@luca-martial is this the updated model ? did you retrain ?

[luca-martial]

@vkocaman yes, this is retrained on new data here is a summary of what I've added:

• used the 90%+ coverage embeddings (dim300)
• created extra synthetic data from my medical records for train set
• removed all synthetic data from test set
• there's no overlap between train/test
• added WikiNER french dataset, including only shorter sentences with less than 10 Os in them
• separated LOC into COUNTRY, CITY, STREET
• for WikiNER added annotations for dates

I checked the model outputs on sample text and it's much more robust imo

[vkocaman]

@vkocaman yes, this is retrained on new data here is a summary of what I've added:

• used the 90%+ coverage embeddings (dim300)
• created extra synthetic data from my medical records for train set
• removed all synthetic data from test set
• there's no overlap between train/test
• added WikiNER french dataset, including only shorter sentences with less than 10 Os in them
• separated LOC into COUNTRY and CITY
• for WikiNER added annotations for dates

I checked the model outputs on sample text and it's much more robust imo

looks nice ! good job.. what about the generic version ?

[luca-martial]

looks nice ! good job.. what about the generic version ?

Currently training - it's coming!

[vkocaman]

Conflicts spotted in PR

[josejuanmartinez]

@vkocaman @luca-martial Same thing I was reporting: Please close any PR until Pavel fixes the md naming and adds the PySpark version to the md file

[muhammetsnts]

Since spark24 version of this model is merged, I will close this not to overwrite on the md file of the spark30 md file. @luca-martial please re-upload this model tomorrow.