This repository was archived by the owner on Feb 28, 2024. It is now read-only.
This repository was archived by the owner on Feb 28, 2024. It is now read-only.
Scikit-optimize master GaussianProcessRegressor is incompatible with scikit-learn 1.0 #1061
Description
scikit-learn 0.19 depreciated the y_train_mean attribute of GaussianProcessRegressor. At the time it was made private, and so scikit-optimize continued to rely on it. #467 was never merged to properly fix that.
Scikit-learn 1.0 removed the attribute, breaking scikit-optimize.
Version
python==3.9.2scikit-learn==1.0scikit-optimize==0.9.dev0(installed withpip install git+https://github.com/scikit-optimize/scikit-optimize.git)
How to reproduce
mkdir test
cd test
python3 -m venv venv
source venv/bin/activate
pip install "scikit-learn==1.0" git+https://github.com/scikit-optimize/scikit-optimize.gitWrite the following script to test.py:
import scipy
import sklearn
import skopt
from sklearn import model_selection, svm
from sklearn.datasets import load_digits
print("scipy:", scipy.__version__)
print("scikit-learn:", sklearn.__version__)
print("scikit-optimize:", skopt.__version__)
X, y = load_digits(n_class=10, return_X_y=True)
X_train, X_test, y_train, y_test = model_selection.train_test_split(
X, y, train_size=0.75, test_size=0.25, random_state=0
)
opt = skopt.BayesSearchCV(
svm.SVC(),
{
"C": (1e-6, 1e6, "log-uniform"),
"gamma": (1e-6, 1e6, "log-uniform"),
"kernel": ["linear", "rbf"],
},
n_iter=32,
cv=3,
)
opt.fit(X_train, y_train)
print("Best train score:", opt.best_score_)
print("Tests score:", opt.score(X_test, y_test))and run it in the test venv:
python test.py