Skip to content

Releases: mila-iqia/cookiecutter-pyml

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

18 Oct 18:54
904af42
Compare
Choose a tag to compare
  • 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

09 Jul 19:26
fd446d2
Compare
Choose a tag to compare

Minor bug fix.

Release 3.1

08 Apr 13:08
03262a2
Compare
Choose a tag to compare
  • 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

21 Aug 19:06
ae3c971
Compare
Choose a tag to compare

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

Release 3.0

17 Mar 18:09
8d2eb42
Compare
Choose a tag to compare
  • 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

15 Aug 16:09
00c2251
Compare
Choose a tag to compare
  • updated sphinx
  • default python interpret version is 3.9
  • fixed some links in the library approval file

Release 2.2

09 Aug 18:38
354f2ad
Compare
Choose a tag to compare
  • removed tensorflow
  • removed MLFlow, added Tensorboard
  • updated libraries

Minor updates - Last release with Tensorflow support

20 Aug 18:31
f5c9701
Compare
Choose a tag to compare
  • 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

28 May 13:58
f139850
Compare
Choose a tag to compare

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

30 Jul 13:00
26dee14
Compare
Choose a tag to compare
  • 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.