Grigorii Kirgizov

Unify Graph class hierarchy

1

That's an important PR that starts big changes in the core of the core: Graph hierarchy design. My analysis of the current situation with Graphs: There is a lot of...

Because Logger is singleton, the initialization is called only once. But if someone tries to call `Log` with custom file after *someone somewhere at some point* already called init with...

Make logging output not so scary for end-user

2

During Graph verification (`GraphVerifier`) & Pipeline evaluation (`PipelineObjectiveEvaluate` with `EvaluationDispatcher`) usually happen a lot of errors. By default these errors are output to the logger and to the console. They...

DataMerger fails when merging indices with different types

1

Failing example. Here it's seen that `str` index can't be merged with `int` index. Possibly this would also fail for two non-numeric indices. ``` 2022-07-28T15:17:14.1922292Z test/unit/utilities/test_project_import_export.py:88: 2022-07-28T15:17:14.1922714Z _ _ _...

Motivations: 1. Define clear base interfaces for easy extension of Optimiser onto new problem domains 2. Simplify GraphAdapter functionality to avoid subtle bugs with lost fields during .adapt/.restore 3. These...

I have a thought of abstracting evolutionary operators as classes, that will encapsulate all necessary parameters and provide concise functional interface. In other words, I'd like to separate *specification* of...

Fix too restrictive operation filtering in assumptions_builder.py

1

Fixes #767. The test there possibly fails because in it only 3 operations are allowed through cli. So there's not enoughg variability in pipeline generation because of just few operations,...

This tests often fails because of `AttributeError('Too many fitness evaluation errors. Composing stopped.')` in evaluation.py

Requirements - There must be specific points in the system architecture where adapter is applied automatically - Mutations & other evolutionary operators shouldn't worry (or even know) about adapter --...

Graph & OptGraph (with GraphNode & OptNode) completely duplicate each other. The original purpose of separating Optimiser from doman (ML) specific can be acheived by using common Graph interface.