English to Twi translation library/system being put together by the Ghana NLP team.
Install Kasa directly from online link:
pip install git+https://github.com/GhanaNLP/kasaWhen editing Kasa locally, it is helpful to install from a local copy of Kasa instead:
git clone https://github.com/GhanaNLP/kasa
cd kasa
pip install .
The examples/ folder of the repo constains examples you can directly run to test various functionalities, e.g.,
python load_and_preprocess_parallel_dataset.py
This may require obtaining some data files and accordingly passing the right paths to the methods in load_and_preprocess_parallel_dataset.py. All this should be self-explanatory if you look at the code.
You will need to download two corpus (English and Twi) into a data folder on your local machine in order to run the examples, using the link below.
https://www.kaggle.com/azunre/jw300entw
Please first clone this repo to your local machine, using a command line tool such as Cygwin or Anaconda Prompt:
git clone https://github.com/GhanaNLP/kasa
Create a branch for your contributions, and check it out:
git branch <your-branch-name>
git checkout <your-branch-name>
Try to pick a branch name that described what you are planning to add to the library, see current branch names in this repo for ideas.
Write your code, test it and then push to your branch:
git push origin <your-branch-name>Create a pull request using the online github repo page, making sure to include a senior member of the team to review your work before merging it into the master branch of the repo.
Never push changes to master!!!
Usually, we start our work by implementing it directly as a Jupyter Python notebook and host it on Kaggle runtime so team members can play around with ideas. Below is a list of the notebooks we have worked on so far, as well as links to where they are hosted on Kaggle so . They are located in the notebooks/ subfolder of this repo, and the hope is to convert all useful methods they contain into refined methods within the Kasa library.
*)The data_processing.ipynb preprocessing notebook is also loaded on kaggle @ https://www.kaggle.com/azunre/ghananlp-kasa-preprocessing-word2vec-v0-1
*)The ghananlp-kasa-retrieval-v0-1.ipynb retrieval "proof-of-concept" notebook is also loaded on kaggle @ https://www.kaggle.com/azunre/ghananlp-kasa-retrieval-v0-
*)The eng_twi_transformer.ipynb transformer-based english-to-twi NMT model notebook is also loaded on kaggle @
https://www.kaggle.com/azunre/ghananlp-kasa-nmt-transformer-v0-1
It is helpful to know the function of the following files when contributing:
*)HISTORY.md -- what functionality was added during each version/release, and what is planned to be added next/in the near future.
*)setup.py -- specify dependencies for added methods here
)__init__.py -- import specification file, does not exist for now...
*)MANIFEST.in -- (yet to be added) this will be used to make sure required files are copied over when the library is installed (no such requirement yet)
The method dir is very useful for debugging your classes and methods, example:
from Kasa import Preprocessingdir(Preprocessing)If you need further information, either contact Ghana NLP members directly, or see a repo on github like this one: