gasearch 0.0.4

Genetic algorithm based hyperparameter optimalization.

geneticsearchcv - Genetic algorithm based hyperparameter tuner

geneticsearchcv - Finding hyperparameters the way nature intended

Hyperparameter search isn’t an exact science. We’ve all heard that one.

geneticsearchcv searches the parameter space using genetic algorithm, with multiple solutions in each generation competing for inclusion in the next.

Mutation of the existing solutions introduces new characteristics and crossover helps the beneficial traits spread across the population. At the same time, proportional selection prevents early convergence of algorithm on local minimum, keeping the population diverse and dynamic.

However, since genetic algorithms are heuristic, there can not be any guarantee that the solution delivered is the optimal one for a given problem. Regardless of the number of iterations.

The package follows scikit-learn API conventions and can be readily integrated with existing pipelines.

TODO:

• Implement alternative selection algorithms

• Publish more examples

• Implement alternative crossover operations

• Improve docs

• Optimize, optimize, optimize …