用于非数字值的SKLL和Kappa

Question

我正在尝试用scikit learn库计算非数字值的Cohen's kappa。有没有办法将一组标签"happy"，"sad"，"happy“转换成一个浮点数？

from skll.metrics import kappa
y_true = ["HAPPINESS","OTHER","NONE","FEAR","NONE","NONE","NONE","ANGER"]
y_pred = ["HAPPINESS","NONE","NONE","FEAR","NONE","NONE","NONE","NONE"] 
kappa_val = kappa(y_true, y_pred)

现在我得到了这个错误：

ValueError: could not convert string to float: "HAPPINESS"

Answer 1

您需要使用LabelEncoder

from sklearn import preprocessing
from sklearn.metrics import classification
import numpy as np
y_true = ["HAPPINESS","OTHER","NONE","FEAR","NONE","NONE","NONE","ANGER"]
y_pred = ["HAPPINESS","NONE","NONE","FEAR","NONE","NONE","NONE","NONE"] 
enc = preprocessing.LabelEncoder()
enc.fit(np.hstack((y_pred, y_true))) #have to give it all possible labels
kappa_val = classification.cohen_kappa_score(enc.transform(y_true), enc.transform(y_pred))
print kappa_val

0.578947368421