Add PhayaThaiBERT engine with new features [WIP]

[pavaris-pm]

What does this changes

According to #868, i have made a new PR by added new folder named phayathaibert since it is a new Thai language model, with that, i decided to treat it to be the same as we treat wangchanberta because it also has their own folder as well. Apart from introducing a new model, i currently made an experiment and added new features into pythainlp including its test cases as well (you can see the list in this PR description). Clearly note here that this PR also solved the unsync forked version in #871 as well.

Will resolve #871 and fix #868.

List of new added features from PhayaThaiBERT [WIP] 🚧 👷🏻‍♂️

⚠️ [NOTE] i will ask for a review after all tasks in the below list completed

Here is the task which I found that PhayaThaiBERT can be integrated in PyThaiNLP after reading a paper. The list below here is the current progress (check mark means that I already added in the source code and will ask for your review after i complete all of them krub):

• Part-of-speech tagging
• Named-entity-recognition
• Tokenization
• Data Augmentation (Text)

Upcoming features that can be added soon (futurePR)

• Word Correction (Under heavy development)

etc ... (I will keep add more into the list based on what I have found during an experiment)

Your checklist for this pull request

🚨Please review the guidelines for contributing to this repository.

• Passed code styles and structures
• Passed code linting checks and unit test

[pep8speaks]

Hello @pavaris-pm! Thanks for updating this PR. We checked the lines you've touched for PEP 8 issues, and found:

There are currently no PEP 8 issues detected in this Pull Request. Cheers! 🍻

Comment last updated at 2023-12-11 14:00:19 UTC

[coveralls]

coverage: 85.515% (-0.9%) from 86.41%
when pulling e7ef6ce on pavaris-pm:dev
into 6514bb8 on PyThaiNLP:dev.

[pavaris-pm]

@bact @wannaphong i've already add all features that i've been found from phayathaibert into the source code and already fix pep8 format.

As of today, here is new features added

1. Part-of-speech tagging model trained by @MpolaarbearM (already made co-authored commit 👍🏻 )
2. Named-entity-recognition model trained by @pavaris-pm
3. Tokenization (even it used the same tokenizer as WangchanBERTa, clearly note here that Vocabulary Expansion was integrated into it) added by @pavaris-pm
4. Data Augmentation (Text) by @pavaris-pm

Upcoming features that can be added soon (future PR)

1. Word Correction (i've found a nice paper for Two-stage Thai Misspelling Correction based on Pre-trained Language Models and need further research on it)
   

   pythainlp/pythainlp/spell/wanchanberta_thai_grammarly.py

   Lines 4 to 11 in 75bbc07

   	"""
   	Two-stage Thai Misspelling Correction based on Pre-trained Language Models
   	:See Also:
   	* Paper: \
   	https://ieeexplore.ieee.org/abstract/document/10202006
   	* GitHub: \
   	https://github.com/bookpanda/Two-stage-Thai-Misspelling-Correction-Based-on-Pre-trained-Language-Models

According to this, i think that adding these 4 completed features first, and the upcoming features (e.g. word correction) can be added later with the next PR because it is better to bring the state-of-the-art Thai encoder based model into production asap. With that, you can review this PR and suggest for further development of them. If you're ok with this, you can approve and merge it krub.

[pavaris-pm]

@MpolaarbearM kindly inform here that co-authored commit already made krub. You can check it 😄

[bact]

There were few error test suite (not related to your PR).
I have fixed them, do you mind to sync your code with the latest from dev? So the test can be run again.
thanks.

[pavaris-pm]

There were few error test suite (not related to your PR).

I have fixed them, do you mind to sync your code with the latest from dev? So the test can be run again.

thanks.

Roger that. I'll do it krub.

[pavaris-pm]

@bact I'm done with syncing already

[sonarcloud]

Kudos, SonarCloud Quality Gate passed!   

0 Bugs
0 Vulnerabilities
0 Security Hotspots
0 Code Smells

No Coverage information
No Duplication information