scikit学习GridSearchCV弃用警告

Question

我正在使用scikit-learn 0.14的GridSearchCV，但总是得到以下警告:

/Library/Frameworks/epd 64 . framework/Versions/7.2/lib/python 2.7/site-packages/sk learn/grid _ search . py:706:deprecation warning:忽略GridSearchCV的附加参数！params参数将在0.15中被删除。弃用警告)

有没有人确切知道哪些参数被忽略了？

代码(x和y从文件中读取):

def balanced_accuracy (ground_truth, predictions):
    f00 = 1. * ((ground_truth == 1) & (predictions == 1)).sum() / (ground_truth == 1).sum()
    f11 = 1. * ((ground_truth == 2) & (predictions == 2)).sum() / (ground_truth == 2).sum()
    return 0.5* (f00 + f11)

bc_score = make_scorer(balanced_accuracy, greater_is_better=True)

C_range = 10. ** np.arange(-3, 3)
gamma_range = 10. ** np.arange(-3, 3)
r_range = np.concatenate((np.array([0]), 10.0 ** np.arange(-1, 3)))
kernel = "poly"
deg = 2
cw = "auto"

param_grid = dict(C=C_range, coef0 = r_range, gamma=gamma_range)
ss = ShuffleSplit(len(y), 10, test_size = 1000, train_size = 1000)

grid = GridSearchCV(svm.SVC(kernel = kernel, max_iter = 1000000, degree = deg, class_weight = cw), param_grid=param_grid, cv=ss, scoring = bc_score)
grid.fit (x, y, sample_weight = sw)

提前感谢！

Answer 1

< code>grid.fit的< code>sample_weight参数已被否决。这在文档中以及通过查看scikit-learn源代码的0.14.X发布分支都很明显。

在这种情况下，sample_weight在0.14版本中也会被忽略，因此grid.fit（x， ysample_weight=sw）等价于grid.fit（x， y）。

作为附带说明，按照pep-8编写是个好主意。