k 最近邻，对准确度分数和混淆矩阵进行交叉验证

Question

我有以下数据，对于每一列，带有数字的行是输入，字母是输出。

A,A,A,B,B,B
-0.979090189,0.338819904,-0.253746508,0.213454999,-0.580601104,-0.441683968
-0.48395313,0.436456904,-1.427424032,-0.107093825,0.320813402,0.060866105
-1.098818173,-0.999161692,-1.371721698,-1.057324962,-1.161752652,-0.854872591
-1.53191442,-1.465454248,-1.350414216,-1.732518018,-1.674040715,-1.561568496
2.522796162,2.498153298,3.11756171,2.125738509,3.003929536,2.514411247
-0.060161596,-0.487513844,-1.083513761,-0.908023322,-1.047536921,-0.48276759
0.241962669,0.181365373,0.174042637,-0.048013217,-0.177434916,0.42738621
-0.603856395,-1.020531402,-1.091134021,-0.863008165,-0.683233589,-0.849059931
-0.626159165,-0.348144322,-0.518640038,-0.394482485,-0.249935646,-0.543947259
-1.407263942,-1.387660115,-1.612988118,-1.141282747,-0.944745366,-1.030944216
-0.682567673,-0.043613473,-0.105679403,0.135431139,0.059104888,-0.132060832
-1.10107164,-1.030047313,-1.239075022,-0.651818656,-1.043589073,-0.765992541

我正在尝试执行 KNN LOOCV 以获得准确度得分和混淆矩阵。

from sklearn.neighbors import KNeighborsClassifier
from sklearn.metrics import accuracy_score
from sklearn.metrics import confusion_matrix
from sklearn.model_selection import LeaveOneOut
import pandas as pd

def main():
  csv = 'data.csv'
  df = pd.read_csv(csv)
  X = df.values.T
  y = df.columns.values
  clf = KNeighborsClassifier()
  loo = LeaveOneOut()
  for train_index, test_index in loo.split(X):
    X_train, X_test = X[train_index], X[test_index]
    y_train, y_test = y[train_index], y[test_index]
    clf.fit(X_train, y_train)
    y_true = y_test
    y_pred = clf.predict(X_test)
    ac = accuracy_score(y_true, y_pred)
    cm = confusion_matrix(y_true, y_pred)
    print ac
    print cm

if __name__ == '__main__':
  main()

但是我的结果都是0。我哪里错了？

Answer 1

我认为你的模型没有得到正确的训练，因为它只需要猜测一个值，所以它没有得到正确的结果。我可以建议切换到 KFold 或 StratifiedKFold。 LOO 的缺点是对于大样本它变得非常耗时。这是我在 X 数据上实施 StratifiedKFold 并进行 3 次拆分时发生的情况。我用 0 和 1 随机填充了 y，而不是使用 A 和 B，并且没有转置数据所以它有 12 行：

from sklearn.neighbors import KNeighborsClassifier
from sklearn.metrics import accuracy_score
from sklearn.metrics import confusion_matrix
from sklearn.model_selection import StratifiedKFold
import pandas as pd

csv = 'C:\df_low_X.csv'
df = pd.read_csv(csv, header=None)
print(df)

X = df.iloc[:, :-1].values
y = df.iloc[:, -1].values

clf = KNeighborsClassifier()
kf = StratifiedKFold(n_splits = 3)

ac = []
cm = []

for train_index, test_index in kf.split(X,y):
    X_train, X_test = X[train_index], X[test_index]
    y_train, y_test = y[train_index], y[test_index]
    print(X_train, X_test)
    clf.fit(X_train, y_train)
    y_pred = clf.predict(X_test)
    ac.append(accuracy_score(y_test, y_pred))
    cm.append(confusion_matrix(y_test, y_pred))
print(ac)
print(cm)

# ac
[0.25, 0.75, 0.5]

# cm
[array([[1, 1],
       [2, 0]], dtype=int64), 

array([[1, 1],
       [0, 2]], dtype=int64),

 array([[0, 2],
       [0, 2]], dtype=int64)]

k 最近邻，对准确度分数和混淆矩阵进行交叉验证

k nearest neighbors with cross validation for accuracy score and confusion matrix

python

pandas

machine-learning

scikit-learn

cross-validation