Federated statistics WP1

This repository contains methods for aggregating statistics.

Structure

• examples contains examples
  • standalone contains examples that show how fedstats can be used in general.
  • flame contains examples that show how fedstats can be used within FLAME.
• fedstats contains the code.
  • aggregation module is used for aggregation methods.
  • models statistical models that can be used with the aggregators.
• tests contains tests.

Currently implemented methods

• Aggregation Methods:

  • Meta-analysis aggregation: weighting local results by their variance (also allows the calculation of confidence intervals)
  • Average aggregation: weighting local results by their number of samples
  • Aggregation of Generalized Linear Models
  • Aggregation of p values using Fisher's method

• Models

  • LocalLinearRegression wraps a standard linear regression model from statsmodels for direct use with the aggregators.
  • LocalFisherScoring implements generalized linear models with Fisher scoring.

Contribute

• The repo uses poetry for dependency management. Assuming you have installed poetry: Clone the repo and type poetry install.
• Activate the environment with poetry env activate.
• Testing is done with pytest: poetry run pytest tests/.