#### Description BaggingClassifier with base_estimator=LinearSVC(), n_estimators=10, n_jobs=10, max_samples=0.1 takes the same time to train as LinearSVC(). #### Steps/Code to Reproduce ```python from sklearn.svm import LinearSVC from sklearn.ensemble import BaggingClassifier base_estimator = LinearSVC(random_state=42, tol=1e-6) n_estimators = 10 max_samples = 1.0 / n_estimators clf = BaggingClassifier(base_estimator, n_estimators=n_estimators, n_jobs=10, max_samples=max_samples) clf.fit(X_trn, y_trn) ``` #### Expected Results I expected it to train about 10 times faster #### Actual Results Trains for the same amount of time as its base estimator LinearSVC(). Produces exactly the same accuracy as LinearSVC(), which is also strange. When I monkey-patched base estimator's fit() method: ```python # Pretend that base_estimator.fit() doesn't support "sample_weight" def fit_no_sample_weight(estimator, X, y): return estimator._original_fit(X, y) base_estimator._original_fit = base_estimator.fit base_estimator.fit = fit_no_sample_weight.__get__(base_estimator, LinearSVC) ``` it trained as expected: about 10 times faster and accuracy was lower. #### Versions System: python: 3.6.8 |Anaconda, Inc.| (default, Dec 30 2018, 01:22:34) [GCC 7.3.0] executable: /foo/anaconda3/bin/python machine: Linux-3.16.36begun-x86_64-with-centos-7.3.1611-Core BLAS: macros: SCIPY_MKL_H=None, HAVE_CBLAS=None lib_dirs: /foo/anaconda3/lib cblas_libs: mkl_rt, pthread Python deps: pip: 19.0.3 setuptools: 40.8.0 sklearn: 0.21.2 numpy: 1.16.2 scipy: 1.2.1 Cython: 0.29.5 pandas: 0.24.1