Nikolay Nikitin
Nikolay Nikitin
1. Modify quick start guide: add step-by-step explanations of typical FEDOT scenaries. 2. Add doc devoted to installation issues. 3. Add instruction to the FEDOT's bug reporting. 4. Extend Q&A...
There are several issues that can cause the descrease in quality metrics obtained with FEDOT: * encoder-related fit problems - #785 * unstable fitness in time - #783 * data...
The restart feature can be useful during experiments: 1) The run of optimiser produces the files with full saved state 2) Then, the optimisation can be continued from the saved...
The following errors occurs in the log for classification tasl:  This problem should be sovled and covered by tests. #785 can be also relevent here.
- [Fixed] List of models for GPU present was not obtaining correctly - [Added] Test for GPU preset Example of solved problem: 
API parameter 'tuner_metric' can be set as Callable or str. However, 'str' or MetricTypesEnum are causing erorr in current version. Also, there is no unit tests that covers this functionality.
In the current implementation, the export of chains designed for text, images, and multi-modal data is not stable. Trasformations of several spefic models raises errors during export to JSON. The...
Basic integration with https://mlflow.org/ should be implemented.
- Derive oversampling/undersampling functions from existing libaries - Implement automated data preprocessing for InputData as a part of Fedot API - Run experiments to prove the efficiency of oversampling/undersampling for...
Must be solved: - https://github.com/nccr-itmo/FEDOT/issues/752 (however, the tests and examples are not the part of PyPI release) Can be included: - https://github.com/nccr-itmo/FEDOT/pull/724 - https://github.com/nccr-itmo/FEDOT/pull/663 - https://github.com/nccr-itmo/FEDOT/pull/707 - https://github.com/nccr-itmo/FEDOT/issues/753