Official repository: https://github.com/MartinBraquet/ml-digits-recognition.
Test online: https://martinbraquet.com/index.php/solo_page_digits_recognition.
pip install ml-digits-recognition
from ml_digits_recognition import drawing
drawing.run()
For basic usage:
pip install -e "."
For development:
pip install -e ".[dev]"
Click here for a full description.
Visualization of the convolutional neural network:
nn_visualization.ipynb
Train the model and save it as model.pt.
ml_digits_recognition_training.ipynb
Accuracy vs epochs.
Loss vs epochs.
Test in Jupiter Notebook. The model can be loaded from the training above in model.pt or from the
default precise model in model_precise.pt.
ml_digits_recognition_test.ipynb
Test in Python.
python src/ml_digits_recognition/drawing.py
Draw a digit and save it as a PNG file.
user_input_drawing.ipynb
Each commit on the main branch is subject to a CI test.
A new version is released to PyPI when a tag is created. TODO: Set version in pyproject.toml only, then trigger a new PyPI release (and git tag) by updating the version and pushing into main. The CD should thus check on each push on main, if the package is more recent than the last tag, then build and deploy new release. This is faster as one only needs to update locally the version in pyproject.toml in the same commit as the fixes for the new release. The rest is automated, so it can all be done from a local IDE.
The version can be accessible at other places if desired. For example in __init__.py:
import importlib.metadata
__version__ = importlib.metadata.version(__package__)Please open an issue.
Contributions are welcome. Please check the outstanding issues and feel free to open a pull request.