Skip to content

GitLab

Closed
Created by Elizaveta Lutsenko@LizLutsenkoOwner0 of 2 tasks completed0/2 tasks

Cover Evaluation subsystem with unit tests

Created by: gkirgizov

Classes introduced during issue #608 (closed) now allow modular unit tests -- so it's time to implement them. This issue is a good introduction into how Fedot evaluates pipelines on data.

Classes requiring testing:

  • MultiprocessingDispatcher -- test that methods return evaluated pipelines with computed fitness, test with/without time limit, with/without multiprocessing, with different kinds of ObjectiveFunction-s (valid/invalid/throwing an exception)
  • PipelineObjectiveEvaluate -- test different kinds of metrics, valid/invalid metric functions, valid/invalid Pipelines, corner cases (empty metrics, empty graph), with/without time constraint, with different data sources

An example of a test and usage of PipelineObjectiveEvaluate is test_cv_multiple_metrics_evaluated_correct. An example of usage of MultiprocessingDispatcher is in EvoGraphOptimiser, see methods __init__ and .optimise().