0.23.1

Substra 0.23.1 - 2022-11-24

Main changes

• BREAKING CHANGE: replace the tasks traintuple, aggregatetuple, predicttuple, testtuple, composite_traintuple with a single task.

  task_key = client.add_task(
      substra.schemas.TaskSpec(
          algo_key=algo_key,
          worker=client.organization_info().organization_id,  # org on which the task is executed
          inputs=[
              {
                  'identifier': 'datasamples',
                  'asset_key': datasample_key
              },
              {
                  'identifier': 'opener',
                  'asset_key': dataset_key
              }
          ],
          outputs= {
              'example_output': {
                  'permissions': {
                      'public': False,
                      'authorized_ids': ['org1'],
                  },
                  'is_transient': True,
              }
          }
      )
  )

  task = client.get_task(task_key)
  tasks = client.list_task()

  # Compute plan changes
  compute_plan = client.add_compute_plan(
      substra.schemas.ComputePlanSpec(
          name = 'my compute plan',
          tasks = [
              schemas.ComputePlanTaskSpec(
                  task_id=uuid.uuid4(),
                  algo_key=algo_key,
                  worker=client.organization_info().organization_id,  # org on which the task is executed
                  inputs=[
                      {
                          'identifier': 'datasamples',
                          'asset_key': datasample_key
                      },
                      {
                          'identifier': 'opener',
                          'asset_key': dataset_key
                      }
                  ],
                  outputs= {
                      'example_output': {
                          'permissions': {
                              'public': False,
                              'authorized_ids': ['org1'],
                          },
                          'is_transient': True,
                      }
                  }
              )
          ]
      )
  )

SubstraFL

• The metric registration is simplified. The user can now directly write a metric function within their script, and directly register it by specifying the right dependencies and permissions. The metric function must have (datasamples, predictions_path) as signature.

Example of new metric registration:

  metric_deps = Dependency(pypi_dependencies=["numpy==1.23.1"])
  permissions_metric = Permissions(public=True)

  def mse(datasamples, predictions_path):
      y_true = datasamples["target"]
      y_pred = np.load(predictions_path)
      return np.mean((y_true - y_pred)**2)


  metric_key = add_metric(
      client=substra_client,
      permissions=permissions_metric,
      dependencies=metric_deps,
      metric_function=mse,
  )

• The round 0 is now exposed. Possibility to evaluate centralized strategies before any training (FedAvg, NR, Scaffold). The round 0 is skipped for single org strategy and cannot be evaluated before training.
• Add support for Python 3.10.
• Local dependencies are installed in one pip command to optimize the installation and avoid incompatibilities error.
• Fix error when installing current package as local dependency.

Substra:

• Add Windows compatibility for Docker mode.

Web application:

• Add documentation link in user menu.
• Removed task categories from the frontend.
• Renamed any tuple thing into a task thing.
• Check for last admin when editing a user.
• Build error logs are now accessible given the right permissions.