交叉验证法 (cross validation): 将数据集D划分成 k 个大小相同或者相似的互斥的子集,每次使用 k - 1 个子集的并集作为训练集,剩余的子集作为测试集,进行k次训练和测试,计算模型在测试集上的准确率,最终返回k个测试结果的平均值 (k 最常用的取值是10)
代码如下:
from sklearn.datasets import load_iris
from random import shuffle
iris = load_iris()
x = iris.data
y = iris.target
dict_class = {}
k = 10
for index_c in range(len(y)):
if y[index_c] in dict_class:
dict_class[y[index_c]].append(index_c)
else:
dict_class[y[index_c]] = [index_c]
for i in dict_class.keys():
shuffle(dict_class[i])
k_fold_test = []
k_fold_train = []
for i in range(k):
for keys in dict_class.keys():
length_sample = len(dict_class[keys])
value = dict_class[keys]
begin = i * length_sample // k
end = (i + 1) * length_sample // k
if i == k - 1:
list_test_index = value[begin:]
list_train_index = value[:begin]
else:
list_test_index = value[begin:end]
list_train_index = value[:begin] + value[end:]
k_fold_test += list_test_index
k_fold_train += list_train_index
print("test ratio: {0:.2f}".format(len(k_fold_test) / (len(k_fold_test) + len(k_fold_train))))
print("train ratio: {0:.2f}".format(len(k_fold_train) / (len(k_fold_test) + len(k_fold_train))))
结果:
test ratio: 0.10
train ratio: 0.90
都看到这,点个赞支持一下咯~