v3.2.0: Merge pull request #131 from mila-iqia/development

• switch to python 3.11
• added script to install directly from the web
• added config option to specify how to do early stopping and also how to select the best checkpoint

Release 3.1.1

Minor bug fix.

Release 3.1

• added OmegaConf to allow hierarchical composition of config files
• added Comet and AIMS
• updated pytorch/pytorch lightning
• other minor updates and improvements

Release 3.0.1

Updated LICENSE file, fixed tensorbord version to avoid problem with pytorch lightning.

Release 3.0

• Switched from the cookiecutter syntax to "pure" python. This does not change much for the final cookiecutter user, but it will greatly simplify the process of adding feature/bug fixing of the developers working on the cookiecutter itself.
• Toy experiment changed to fashion mnist, in order to have a classification task instead of a regression one (as it was before).
• Evaluation script was rewritten and now it's separated from the training script.
• Added pytype to Continuous Integration.

Release 2.3

• updated sphinx
• default python interpret version is 3.9
• fixed some links in the library approval file

Release 2.2

• removed tensorflow
• removed MLFlow, added Tensorboard
• updated libraries

Minor updates - Last release with Tensorflow support

• Fixed branch names (master => main) when using github (and github actions).
• By default, creating a project that support a generic SLURM cluster (but optionally it can support Mila cluster + Compute Canada).
• Minor cleaning.

Pytorch-lightning/Keras

The main change is the switch to using pytorch-lightning and keras.
This allows the user to rely on all the features supported by those two higher-level APIs.

Additional CI configuration has also been added (Github actions, Azure).

Codecov has been removed. Currently the test coverage is computed directly in the CI, and printed in the CI logs.

Release 1.1

• Added seed for python/numpy/pytorch/tensorflow.
• Add proper handling of data on the Mila/ComputeCanada server - like, use $SLURM_TMPDIR as a temporary working dir.