Features

How does FEDOT avoid data leakage and bias?

Before the usage of AutoML, the dataset is usually splitting to training and test part. Then, the cross-validation is used to estimate the value of fitness functions. The default number of folds is 5. It can be specified manually using cv_folds parameter.

To deal with potential bias, additional metrics can be passed to optimiser using metric parameter.

Which algorithms are available as building block for optimiser?

The names and metadata of available ML models can be obtained from JSON file, as well as data operations file.

What hyperparameter options were available to each algorithm?

The ranges for tuning of hyperparameters are presented in search_space.py file. The default values of hyperparameters are available in JSON file.