Skip to content

Latest commit

 

History

History
156 lines (94 loc) · 5.05 KB

CHANGES.md

File metadata and controls

156 lines (94 loc) · 5.05 KB
Raw

Releases

0.0.39 (2021/05/20)

  • (models/fro) Updated the model

0.0.38 (2021/05/20)

  • (deps) Moved to PaPie

0.0.37 (2021/04/14)

  • (models/fro) Fixed regression for apostrophes

0.0.36 (2021/04/12)

  • (models/fro) Added [REF:...] excluder

0.0.35 (2021/04/12)

  • (models/Freem) Update model to handle morph
  • (models/Freem+Fr) Added reference excluder
  • (models/Freem+Fr) Excluders updated to use CharRegistry
  • (pipeline/excluders) reworked ApostropheExcluder and the like to use CharRegistry

0.0.34 (2021/04/02)

  • (CLI) Allows to specify --max-tokens at the CLI level

0.0.33 (2021/04/02)

  • (model/lasla) Updated model to use multi-part pie model

0.0.32 (2021/03/22)

  • (model/grc) Added a [REF:.*] excluder
  • (Tagger) Passed the argument for lower to the main Tagger (might not change a thing)

0.0.31 (2021/02/17)

  • (requirements) Upgrade pie version because I (@ponteineptique) messed up.

0.0.30 (2021/02/17)

  • (requirements) Upgrade pie version

0.0.29 (2021/02/11)

  • (models/grc) Added morphology tags and updated to 0.0.2

0.0.28 (2021/02/11)

  • (requirements) Upgrade pie version

0.0.27 (2021/02/01)

  • (models/lasla) Support ignoring character tokens through [IGN:char]
  • (pipeline/excluders) Made sure excluder would use the same replacement character through a CharRegistry dictionary

0.0.26 (2021/01/14)

  • (models/lasla) Use model LASLA+ from 0.0.5b trained on PyTorch 1.3.1

0.0.25 (2021/01/13)

  • Added a max_tokens per sentence limit in DataIterators.

0.0.24 (2020/12/14)

  • (models/fro) Updated model fro to 0.3.0 using multiple tasks

0.0.23 (2020/12/04)

  • (models/dum) Added a new model with Middle Dutch thanks to Mike Kestemont
  • (tokenizers) Added a SimpleTokenizer based on length

0.0.22 (2020/12/02)

  • (models/lasla) Apply unidecode
  • (models/lasla) Use model LASLA+ from 0.0.5alpha trained on PyTorch 1.3.1
  • (models/lasla) Updated the abbreviation list
  • (CI) Added Github Actions
  • (Documentation) Added a warning about supported python versions
  • (Documentation) Fixed the example
  • (pipeline) Created AbbreviationsRemoverExcluder
  • (dependencies) Cleaned the version requirements due to pip update

0.0.21 (2020/09/24)

  • (models/LASLA) Fixed a bug where clitics are not split correctly after nouns
  • Fixed multiple typos in CHANGES.md in version numbers

0.0.20 (2020/09/22)

  • New Latin model which handles capitalized input, entities and better disambiguation.

0.0.19 (2020/09/18)

  • (Latin Model) Fixed a long standing bug where Latin would not tag Gender because I forgot it in the GlueProcessor... Big Facepalm

0.0.18 (2020/09/08) - Release named A2M

  • Fixed the way the DataIterator deals with documents ending with a sentence formed of excluded tokens only.
  • Fixed a typo in an import pattern
  • (Latin Model) Dealt with some weird Unicode numerals which unexpectedly broke our .isnumeric() usage (e.g. ↀ )

0.0.17 (2020/07/26)

  • Added a way to tag texts where word are already tokenized: new lines are word separator, double new lines are sentence separator
  • Reworked the way preprocessing of special chars is done prior to sentence tokenization and after it. Creation of the class Excluder (pie_extended.pipeline.tokenizers.utils.excluder)
    • Allows for more code sharing across models.
  • Fixed a typo that would prevent to tag with FREEM (and nobody saw that ! ;) )

0.0.16 (2020/06/22)

  • Fixed Early Modern French Model (reusing processor and tokenizer of FR model)
  • Added Ancient Greek Model (Very basic addition, need more work probably ?)

0.0.15 (2020/06/22)

  • Added Early Modern French Model (reusing processor and tokenizer of FR model)

0.0.14 (2020/04/24)

  • Hotfixed columns order in tsv output

0.0.13 (2020/04/24)

  • Hotfixed lowercase for latin model

0.0.12 (2020/04/24)

Unfilled TODO

0.0.11 (2020/04/24)

Unfilled TODO

0.0.10 (2020/04/24)

  • PIE_EXTENDED_DOWNLOADS environment variable can be used to set up a non default directory for models and linked data.
    • eg. PIE_EXTENDED_DOWNLOADS=~/PieData pie-extended download fr

0.0.9 (2020/04/24)

  • (Breaking) Postprocessors now must return a list of dict instead of a dict with .get_dict() methods (#c8be021)
  • Added a better tokenizer for Classical French
    • Keeps aujourd'hui intact
    • Keeps union dash with pronouns for -le in sentence such as mange-le.
    • Keeps the -t euphonique together with the pronoun: mange-t-il becomes mange and -t-il
      • Its removed from lemmatization until a new model is trained (old model had -t on the verb)
      • Elision work as intended for non euphonique such as "Va-t'en" -> va, -t', en
  • Updated Classical French models (#15)
  • Added a post-processor to split tokens (#17)