0.40.0-rc1

Added

• Check the Python version used before generating the Dockerfile ([#155])(#155)).

• Python dependencies can be resolved using pip compile during function registration by setting compile to True
  in the Dependency object ([#155])(#155)).

  Dependency(
        pypi_dependencies=["pytest", "numpy"],
        compile=True,
    )

• Dependency objects are now computed at initialization in a cache directory, accessible through the copy_compute_dir method. The cache directory is deleted at the Dependency object deletion. ([#155])(#155))

• Check created wheels name. (#160)

Changed

• BREAKING: Rename generate_wheel.py to manage_dependencies.py (#156)
• BREAKING: Move manage_dependencies.py from remote.register to dependency (#158)
• BREAKING: local_dependencies is renamed local_installable_dependencies (#158)
• BREAKING: local_installable_dependencies are now limited to local modules or Python wheels (no support for bdist, sdist...)([#155])(#155)).

Fixed

• Set, save & load random.seed and np.random.seed along with torch.manual_seed in TorchAlgo(#151)

0.39.1

0.39.1 - 2023-09-05

Fixed

• Keep the last round task output by default (#166)

0.38.1

0.38.1 - 2023-09-05

Fixed

• Keep the last round task output by default (#165)

0.37.1

0.37.1 - 2023-09-04

Fixed

• Keep the last round task output by default (#163)

0.39.0

This is a promotion of 0.39.0rc1

Changed

• BREAKING: Input and output of aggregate tasks are now shared_state. It provides more flexibility to link different type of tasks with each other. To use
  download_aggregate_shared_state on experiments launched before this commit, you can use the following code as a replacement of the function
  (#142).

import tempfile
from substrafl.model_loading import _download_task_output_files
from substrafl.model_loading import _load_from_files
with tempfile.TemporaryDirectory() as temp_folder:
    _download_task_output_files(
        client=<client>,
        compute_plan_key=<compute_plan_key>,
        dest_folder=temp_folder,
        round_idx=<round_idx>,
        rank_idx=<rank_idx>,
        task_type="aggregate",
        identifier="model",
    )
    aggregated_state = _load_from_files(input_folder=temp_folder, remote=True)

Removed

• Function wait in utils. You can use substra.Client.wait_task & substra.Client.wait_compute_plan instead. (#147)

Fixed

• Compatibility with GPU devices when running torch based experiments (#154)
• Pin pydantic to >=1.9.0 & <2.0.0 as pydantic v 2.0.0 has been released with a lot of non backward compatible changes. (#148)

0.39.0rc1

chore: release 0.39.0

Signed-off-by: Olivier Léobal <[email protected]>

0.38.0

0.38.0 - 2023-06-30

Changed

• BREAKING: Rename model_loading.download_shared_state to model_loading.download_train_shared_state (#143)
• BREAKING: Rename model_loading.download_aggregated_state to model_loading.download_aggregate_shared_state (#143)
• Numpy < 1.24 in dependencies to keep pickle compatibility with substra-tools numpy version (#144)

0.38.0-rc1

0.38.0-rc1 - 2023-06-27

Changed

• BREAKING: Rename model_loading.download_shared_state to model_loading.download_train_shared_state (#143)
• BREAKING: Rename model_loading.download_aggregated_state to model_loading.download_aggregate_shared_state (#143)
• Numpy < 1.24 in dependencies to keep pickle compatibility with substra-tools numpy version (#144)

0.37.0

0.37.0 - 2023-06-12

Added

• ComputePlanBuilder base class to define which method are needed to implement a custom strategy in SubstraFL.
  These methods are build_compute_plan, load_local_states and save_local_states. #120
• Check and test on string used as metric name in test data nodes (#122).
• Add default exclusion patterns when copying file to avoid creating large Docker images (#118)
• Add the possibility to force the Dependency editable_mode through the environment variable SUBSTRA_FORCE_EDITABLE_MODE (#131)

Changed

• BREAKING: depreciate the usage of model_loading.download_algo_files and model_loading.load_algo functions. New utils functions are now available. (#125)
  model_loading.download_algo_state to download a SubstraFL algo of a given round or rank.
  model_loading.download_shared_state to download a SubstraFL shared object of a given round or rank.
  model_loading.download_aggregated_state to download a SubstraFL aggregated of a given round or rank.
  The API change goes from:

  algo_files_folder = str(pathlib.Path.cwd() / "tmp" / "algo_files")
  download_algo_files(
    client=client_to_download_from,
    compute_plan_key=compute_plan.key,
    round_idx=round_idx,
    dest_folder=algo_files_folder,
  )
  model = load_algo(input_folder=algo_files_folder).model

  to

  algo = download_algo_state(
    client=client_to_download_from  ,
    compute_plan_key=compute_plan.key,
    round_idx=round_idx,
  )
  model = algo.model

• BREAKING: rename build_graph to build_compute_plan. (#120)

• BREAKING: move schema.py into the strategy module. (#120)

  from substrafl.schemas import FedAvgSharedState
  # Become
  from substrafl.strategies.schemas import FedAvgSharedState

• Way to copy function files (#118)

• download_train_task_models_by_rank uses new function list_task_output_assets instead of using value that has been removed (#129)

Fixed

• New dependencies copy method in Docker mode.(#130)

0.37.0rc2

0.37.0rc2 - 2023-06-12

Added

• ComputePlanBuilder base class to define which method are needed to implement a custom strategy in SubstraFL.
  These methods are build_compute_plan, load_local_states and save_local_states. #120
• Check and test on string used as metric name in test data nodes (#122).
• Add default exclusion patterns when copying file to avoid creating large Docker images (#118)
• Add the possibility to force the Dependency editable_mode through the environment variable SUBSTRA_FORCE_EDITABLE_MODE (#131)

Changed

• BREAKING: depreciate the usage of model_loading.download_algo_files and model_loading.load_algo functions. New utils functions are now available. (#125)
  model_loading.download_algo_state to download a SubstraFL algo of a given round or rank.
  model_loading.download_shared_state to download a SubstraFL shared object of a given round or rank.
  model_loading.download_aggregated_state to download a SubstraFL aggregated of a given round or rank.
  The API change goes from:

  algo_files_folder = str(pathlib.Path.cwd() / "tmp" / "algo_files")
  download_algo_files(
    client=client_to_download_from,
    compute_plan_key=compute_plan.key,
    round_idx=round_idx,
    dest_folder=algo_files_folder,
  )
  model = load_algo(input_folder=algo_files_folder).model

  to

  algo = download_algo_state(
    client=client_to_download_from  ,
    compute_plan_key=compute_plan.key,
    round_idx=round_idx,
  )
  model = algo.model

• BREAKING: rename build_graph to build_compute_plan. (#120)

• BREAKING: move schema.py into the strategy module. (#120)

  from substrafl.schemas import FedAvgSharedState
  # Become
  from substrafl.strategies.schemas import FedAvgSharedState

• Way to copy function files (#118)

• download_train_task_models_by_rank uses new function list_task_output_assets instead of using value that has been removed (#129)

Fixed

• New dependencies copy method in Docker mode.(#130)