0.28.0

0.28.0 - 2022-09-12

Fixed

• Throw an error if pytorch 1.12.0 is used.
  There is a regression bug in torch 1.12.0, that impacts optimizers that have been pickled and unpickled. This bug occurs for Adam optimizer for example (but not for SGD). Here is a link to one issue covering it: pytorch/pytorch#80345

Changed

• Removing classic-algos from the benchmark dependencies
• NOTABLE CHANGES due to breaking changes in connect-tools: the user must now pass the method name to execute from the
  tools defined class within the dockerfile of both algo and metric under the --method-name argument:

  ENTRYPOINT ["python3", "metrics.py"]

  shall be replaced by:

  ENTRYPOINT ["python3", "metrics.py", "--method-name", "score"]

• Use the new Susbtra sdk features that return the path of the downloaded file. Change the model_loading.pyimplementation and the tests.

Features

• In the PyTorch algorithms, move the data to the device (GPU or CPU) in the training loop and predict function so that the user does not need to do it.
• Rename connect-tools docker images to substra-tools
• Benchmark:
  • use public data hosted on Zenodo for the benchmark

Tests

• Fix the GPU test to the last breaking changes, and unskip the use_gpu=False case

Documentation

• Update the NpIndexGenerator docstrings to add information how to use it as a full epoch index generator.