0.18.0

0.18.0 - 2022-09-26

Added

• feat: allow CLI parameters to be read from a file

CHANGED

• BREAKING CHANGES:

  • the opener only exposes get_data and fake_data methods.
  • the results of the above method is passed under the datasamples keys within the inputs dict arg of all
    tools methods (train, predict, aggregate, score).
  • all method (train, predict, aggregate, score) now takes a task_properties argument (dict) in addition to
    inputs and outputs.
  • The rank of a task previously passed under the rank key within the inputs is now given in the task_properties
    dict under the rank key.

• BREAKING CHANGE: The metric is now a generic algo, replace

import substratools as tools

class MyMetric(tools.Metrics):
    # ...

if __name__ == '__main__':
    tools.metrics.execute(MyMetric())

by

import substratools as tools

class MyMetric(tools.MetricAlgo):
    # ...
if __name__ == '__main__':
    tools.algo.execute(MyMetric())

0.17.0

Changed

• feat: all algo classes rely on a generic algo class

0.16.0

0.16.0 - 2022-09-12

Changed

• Remove documentation as it is not used. It will be replaced later on.
• BREAKING CHANGES: the user must now pass the method name to execute within the dockerfile of both algo and
  metric under the --method-name argument. The method name still needs to be one of the algo or metric
  allowed method name: train, predict, aggregate, score.

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

  shall be replaced by:

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

• BREAKING CHANGES: rename connect-tools to substra-tools (except the github folder)

0.7.0

Features

• Replaces the --data-samples-path arg that was unused by the backend and was expecting the folder in which all data samples would be by --data-sample-paths which is a list of the path to each individual data sample folder.

0.6.1

Features

• Make the 'n_samples' argument mandatory when using fake data
• Add a minimal Docker image, with only contains python3 and substratools to speed up local development

0.6.0

Features

• improved documentation
• new checks making sure that output files were saved in the right place (helpful when developing new assets)
• Algo and AggregateAlgo get an user_models_generator attribute that sets whether to pass models as list or as generators to the train or aggregate method
• New --n-fake-samples arg to be used for training and testing tasks to use in conjunction with --fake-data-samples. The value is passed to the opener.

0.5.0

Features

• Improved logs
• Updated Algo and CompositeAlgo: train methods don't return predictions anymore

0.4.0

Features

• CompositeAlgo: separate serializers for head and trunk models

0.3.0

Features:

• CompositeAlgo base class has been added
• AggregateAlgo base class has been added