sksearchspace 2020.8.0.0.23.1

Defines search spaces for scikit-lean estimators

sksearchspace

Scikit-learn Search Space Configurations with curated search spaces for scikit-learn estimators.

Usage

from sksearchspace import SearchSpace
from sklearn.tree import DecisionTreeClassifier

estimator_space = SearchSpace.for_sklearn_estimator(DecisionTreeClassifier, seed=42)
estimator_space.sample()
# {'criterion': 'entropy','min_samples_leaf': 15, 'min_samples_split': 11}

estimator_space.sample()
# {'criterion': 'entropy', 'min_samples_leaf': 12, 'min_samples_split': 4}

sksearchspace uses ConfigSpace for sampling. The ConfigSpace configuration can be accessed through an attribute:

estimator_space.configuration
# Configuration space object:
# Hyperparameters:
#   criterion, Type: Categorical, Choices: {gini, entropy}, Default: gini
#   min_samples_leaf, Type: UniformInteger, Range: [1, 20], Default: 1
#   min_samples_split, Type: UniformInteger, Range: [2, 20], Default: 2

A json file can be loaded as follows:

with open("search_space.json", "r") as f:
    estimator_space = SearchSpace(f.read())

License

Copyright (c) 2020 Thomas J. Fan

Distributed under the terms of the MIT license, pytest is free and open source software.