Yet Another Session for Training Artificial Intelligence

This is a framework for easy development of new AI Research Projects.

• Anaconda Envirment.
• AI project structure.
• Tensorflow, Keras and Scikit-learn.
• HPC jobs manager.
• Pip Package.
• Weight and Bias easy setup.
• Notebooks.
• Nose unit tests.
• Singularity container.
• Easy experiment reproducibility.

This is a guide to understand the zen of this repository.

There are 5 folder for develop you new project:

AI folder is your AI project package, for more details check AI/README.md

If you want to rename the IA folder with your project name, you must also edit theses files:

• setup.py: replace ia_name variable in line 7.
• __version__.py inside AI folder.
• Maybe you want to edit the conda envirment file yastai.yml.
• Replace from AI import * in every script in train/.
• Replace from AI import * in every nootebook in notebooks/.
• Replace from AI import * in every nootebook in tests/.
• Replace AI from files in docs/.
• HPC train scripts maybe
• conda environment maybe

Collections of scripts for setup your data, config preprocessing pipes, and train/evaluate your models.

In these folder you can find also:

• experiments/: Folder with experiments configurations and results. More details and name conventions in train/experiments/README.md.
• debug.py: Utils functions for check GPU support, profiling keras/tensorflow models and plots models.
• hpc/: Script and instructions for run experiments in HPC environments.

More details for train/ folder in train/README.md.

Contains notebook version of the scripts in \train. More details in notebooks/README.md.

Handy for Google Colab.

Collections of user defined unit tests with nose.

Here a tutorial for nose oython testing.

More details in tests/README.md.

Different options for consume a trained model.

More details in api/README.md.

TODO

Check anaconda documentation here.

• environment.yml

conda env create -f environment.yml --prefix ./.env

TODO Manually add packages to enviroment.yml and Pipfile.

TODO

TODO

Requiere PyPi and TestPyPi accounts

doc

conda activate ./.env
python setup.py sdist bdist_wheel
twine check ./*

twine upload --repository testpypi dist/*

twine upload dist/*