import相關工具箱：

from sklearn.decomposition import PCA
from sklearn.preprocessing import normalize

L2-normalization（可選）:

def l2_norm(data):
    for i in range(data.shape[0]):
        data[i, :] = normalize(data[i, :].reshape((1, -1)), norm='l2')  # L2歸一化
    return data
train_annotated_feature = l2_norm(train_annotated_feature)
train_candidates_feature = l2_norm(train_candidates_feature)
test_annotated_feature = l2_norm(test_annotated_feature)
test_candidates_feature = l2_norm(test_candidates_feature)
 

用train_annotated_feature資料訓練PCA模型，然後對其他資料進行主成分分析：

pca = PCA(n_components=300, copy=True, whiten=False)    # 降維到300dimensions
pca.fit(train_annotated_feature)

pca_train_annotated_feature = pca.transform(train_annotated_feature)
pca_train_candidates_feature = pca.transform(train_candidates_feature)
pca_test
 
_annotated_feature = pca.transform(test_annotated_feature)
pca_test_candidates_feature = pca.transform(test_candidates_feature)