Peter Shevcnenko
Peter Shevcnenko
- Refactored default graph representations by matplotlib + networkx. - Added 'pyvis' visualization engine for interactive graph views via html. - Added 'graphviz' visualization engine (requires [Graphviz](https://graphviz.org/) and [pygraphviz](https://pygraphviz.github.io/)). -...
The key idea is to observe all historical pipelines' performance on an additional data and compare it to "fitness". "Additional data" in this context means any data that was not...
A custom Serializer is needed that implies knowledge of OptHistory's inner structure. It must not contain or dump paths to classes used in OptHistory, thus it is to be immune...
When `InputData` is retrieved from a source with explicit feature and target class names (e.g. `.csv` file or Pandas `DataFrame`), it should remember their labels. The labels should also be...
An attempt to minimize data exchange between processes by changing evaluation interface.
This PR is addressed to solve the following issues: - #900 - #846 Also makes a step towards #866 (labeled generations are planned, possible with custom values). Work in progress:...
- Replaced `while` -> `for`, as it is significantly faster. - Moved some conditions out of cycles. - Made copies of unique elements from the lists to avoid duplicates check....
Some convenience improvements should be performed at any time: - [ ] Allow `Fedot.fit` to get path to `features` and `target` via `Path` objects. - [ ] Write explicit docstring...
Currently, individuals in `OptHistory` designed to be uncopyable and share the same object by unique `uid`. Evaluation made parallel by `joblib` returns a list of new implicitly copied individuals and...
This is a meta-issue to collect all related problems. - [ ] #896 - [ ] #900