EGO (Efficient global optimization) and multiply target EGO method.
References: Jones, D. R., Schonlau, M. & Welch, W. J. Efficient global optimization of expensive black-box functions. J. Global Optim. 13, 455โ492 (1998)
pip install multiego
if __name__ == "__main__":
from sklearn.datasets import load_boston
import numpy as np
from multiego.ego import search_space, Ego
from sklearn.model_selection import GridSearchCV
from sklearn.svm import SVR
#####model1#####
model = SVR() #pre-trained good model with optimized prarmeters for special features
###
X, y = load_boston(return_X_y=True)
X = X[:, :5]
searchspace_list = [
np.arange(0.01, 1, 0.1),
np.array([0, 20, 30, 50, 70, 90]),
np.arange(1, 10, 1),
np.array([0, 1]),
np.arange(0.4, 0.6, 0.02),
]
searchspace = search_space(*searchspace_list)
#
me = Ego(searchspace, X, y, 500, model, n_jobs=6)
re = me.egosearch()
For sklean-type
single model.
- For any user-defined single model, just need offer mean and std of search space.
- For big search space out of memory , just need offer mean and std of search space.
For sklean-type
models.
multiego.base_multiplyego.BaseMultiEgo
- For any user-defined models, just need offer predict_y of search space.
- For big search space out of memory, just need offer predict_y of search space.
More examples can be found in test.
More powerful can be found mipego