Directed, Rapid and Optimized Search Of Fitting HYPErparameters.
- GitHub repo: https://github.com/jbdelfau/Drosofhype
- Free software: GNU license
A genetic algorithm that searches for the optimal hyperparameters of a machine-learning model.
This library requires a machine-learning model and a json specifying the genetic search configuration. 2 examples are provided in the files test_svm.py and test_xgbregressor.py.
Let's start with the parameters of the genetic search:
- population_size: size of each generation.
- n_parents: number of best performing individuals kept to generate children.
- mutation_rate: occurence of children mutations(between 0 and 1).
- additional_properties: information that must be included in the output table as additional columns.
In addition, the json must contain an additional key - hyperparameters - describing which parameters of the model have to be varied and in what range. See the above files for more examples.