我設(shè)置了一個(gè)簡(jiǎn)單的實(shí)驗(yàn)，以在運(yùn)行sklearnGridSearchCV時(shí)檢查多核CPU的重要性KNeighborsClassifier。我得到的結(jié)果令我感到驚訝，我想知道我是否誤解了多核的好處，或者我做得不好。2-8個(gè)工作之間的完成時(shí)間沒(méi)有差異。怎么來(lái)的 ？我已經(jīng)注意到“ CPU性能”選項(xiàng)卡上的差異。在第一個(gè)單元運(yùn)行時(shí)，CPU使用率約為13％，而最后一個(gè)單元?jiǎng)t逐漸增加到100％。我期望它能更快完成。也許不是線(xiàn)性地更快，也就是8個(gè)工作將比4個(gè)工作快2倍，但要快一點(diǎn)。這是我的設(shè)置方式：我正在使用jupyter-notebook，單元格是指jupyter-notebook單元格。我已經(jīng)加載了MNIST，并在中使用0.05了3000數(shù)字的測(cè)試大小X_play。from sklearn.datasets import fetch_mldatafrom sklearn.model_selection import train_test_splitmnist = fetch_mldata('MNIST original')X, y = mnist["data"], mnist['target']X_train, X_test, y_train, y_test = X[:60000], X[60000:], y[:60000], y[60000:]_, X_play, _, y_play = train_test_split(X_train, y_train, test_size=0.05, random_state=42, stratify=y_train, shuffle=True)在下一個(gè)單元格中，我將進(jìn)行設(shè)置KNN并設(shè)置一個(gè)GridSearchCVfrom sklearn.neighbors import KNeighborsClassifierfrom sklearn.model_selection import GridSearchCVknn_clf = KNeighborsClassifier()param_grid = [{'weights': ["uniform", "distance"], 'n_neighbors': [3, 4, 5]}]然后，我為8個(gè)n_jobs值完成了8個(gè)單元格。我的CPU是具有4核8線(xiàn)程的i7-4770。grid_search = GridSearchCV(knn_clf, param_grid, cv=3, verbose=3, n_jobs=N_JOB_1_TO_8)grid_search.fit(X_play, y_play)結(jié)果Parallel(n_jobs=1)]: Done  18 out of  18 | elapsed:  2.0min finishedParallel(n_jobs=2)]: Done  18 out of  18 | elapsed:  1.4min finishedParallel(n_jobs=3)]: Done  18 out of  18 | elapsed:  1.3min finishedParallel(n_jobs=4)]: Done  18 out of  18 | elapsed:  1.3min finishedParallel(n_jobs=5)]: Done  18 out of  18 | elapsed:  1.4min finishedParallel(n_jobs=6)]: Done  18 out of  18 | elapsed:  1.4min finishedParallel(n_jobs=7)]: Done  18 out of  18 | elapsed:  1.4min finishedParallel(n_jobs=8)]: Done  18 out of  18 | elapsed:  1.4min finished第二次測(cè)試隨機(jī)森林分類(lèi)器的使用要好得多。測(cè)試大小為0.5，30000圖片。from sklearn.ensemble import RandomForestClassifierrf_clf = RandomForestClassifier()param_grid = [{'n_estimators': [20, 30, 40, 50, 60], 'max_features': [100, 200, 300, 400, 500], 'criterion': ['gini', 'entropy']}]