@kostajh, we published datasets for all 12/16 models that passed the evaluation in this round.

The conclusion on the backtesting results is that most of the languages look fine besides:

• wuuwiki, zh_classicalwiki, and zh_yuewiki which have extremely low precision and recall compared to the recommended threshold of 0.75 and 0.2.

Model evaluation has been completed and below are the backtesting results:

15/16 models were trained successfully in the 14th round of wikis.

The Chinese Wikipedia (zhwiki) training pipeline is also throwing this error:

File "wikipedia2vec/dictionary.pyx", line 231, in wikipedia2vec.dictionary.Dictionary.build
File "wikipedia2vec/dump_db.pyx", line 124, in wikipedia2vec.dump_db.DumpDB.is_disambiguation
File "wikipedia2vec/dump_db.pyx", line 125, in wikipedia2vec.dump_db.DumpDB.is_disambiguation
File "wikipedia2vec/dump_db.pyx", line 126, in wikipedia2vec.dump_db.DumpDB.is_disambiguation
UnicodeEncodeError: 'utf-8' codec can't encode characters in position 2-3: surrogates not allowed

@kostajh, we published datasets for all 23/24 models that passed the evaluation in this round.

The conclusion on the backtesting results is that most of the languages look fine besides:

• piwiki's precision and recall are nil.
• orwiki (0.71) and pawiki (0.74) have a precision slightly lower than the recommended one (0.75).
• nqowiki has a slightly low precision (0.73) and low recall (0.11).

Model evaluation has been completed and below are the backtesting results:

24/24 models were trained successfully in the 13th round of wikis.

@kostajh, we published datasets for all 21/23 models that passed the evaluation in this round.

The conclusion on the backtesting results is that most of the languages look fine besides:

• mnwwiki's precision and recall are bad. Both are nil.
• mnwiki (0.72) and newiki (0.74) have a precision slightly lower than the recommended one (0.75).
• mlwiki's precision (0.69) and recall (0.14) are lower than the recommended one (0.75 and 0.2 respectively).
• mywiki has a low precision (0.63) and very low recall (0.06).

@kostajh and @Sgs, thank you for letting me know about cewiki missing in the wikis.txt. When we trained this wiki the process completed successfully and we published the datasets here:

@Sgs, yes, the number of links in a wiki affects how the model performs.

Model evaluation has been completed and below are the backtesting results:

23/23 models were trained successfully in the 12th round of wikis.

@kostajh, we published datasets for all 21/22 models that passed the evaluation in this round.

The conclusion on the backtesting results is that most of the languages look fine besides:

Model evaluation has been completed and below are the backtesting results:

The lrcwiki pipeline was failing during the spark job with the message poinValueError: RDD is empty.

21/22 models were trained successfully in the 11th round of wikis.

@kostajh, we published datasets for all 17/19 models that passed the evaluation in this round.

kywiki has been added to wikis that will be deployed in the 11th round T308136

The Armenian sentence tokenization bug has been fixed in T327371#8631149

The conclusion on the backtesting results is that most of the languages look fine besides:

• klwiki (0.74), and kmwiki (0.70) have a precision that is slightly lower than the recommended one (0.75).
• krcwiki has a low precision (0.65).
• koiwiki has a low recall (0.13).

Model evaluation has been completed and below are the backtesting results:

18/19 models were trained successfully in the 10th round of wikis.

@MGerlach, thank you for the recommendations. I have tested the fix locally and the hywiki training pipeline completed successfully.

A unique URL has been added for each WikiGPT search result.

Thank you for working on this, @isarantopoulos!

@isarantopoulos, I could not find Chris' username on Toolforge.

An explainability section has been added to WikiGPT.

A login page has been added to WikiGPT.

WikiGPT is now up and running. You can see it here: https://wiki-gpt.toolforge.org/

@kostajh, we published datasets for all 19/21 models that passed the evaluation in this round.

Thank you for filing this @kostajh. I will work on periodically updating the link recommendation models.

@kostajh, yes @MGerlach and I will work on this.

The conclusion on the backtesting results is that most of the languages look fine besides:

Model evaluation has been completed and below are the backtesting results: