Python：基于二分类分解的序分类方法

参考文献：[1] Frank E, Hall M. A simple approach to ordinal classification[C]//European Conference on Machine Learning. Springer, Berlin, Heidelberg, 2001: 145-156.

没有进行类封装，仅是做了函数复现~

分类效果比单独使用多分类‘OVR’效果要好1-2%个精度

import numpy as np
from copy import deepcopy
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import accuracy_score,f1_score
from sklearn.model_selection import StratifiedKFold,KFold
from sklearn import datasets
from collections import OrderedDict



def BDOC_train(X,y):
    rowNum = len(y)
    labelNum = len(set(y))
    labels = np.unique(y)
    ###---------split the label list into K-1 label list-------------###
    labels_ = np.delete(labels,max(labels))
    target = OrderedDict()
    for i,ele in enumerate(labels_):
        target[i] = deepcopy(y)
        for j in range(rowNum):
            if target[i][j] <= ele:
                target[i][j] = ele
            else:
                target[i][j] = ele + 1
    modelDict = OrderedDict()
    ###----------------initial K-1 learning model----------------------###
    for i,ele in enumerate(labels_):
        modelDict[i] = LogisticRegression(solver='newton-cg',penalty='l2')
    ###----------------fit the K-1 learning model----------------------###
    for i,model in modelDict.items():
        model.fit(X,target[i])
    return modelDict

def BDOC_prob(modelDict,test_X):
    probDict = OrderedDict()
    for i,model in modelDict.items():
        probDict[i] = model.predict_proba(test_X)
    return probDict

def BDOC_pred(probDict,len_test,labelNum,labels):
    y_pred = np.zeros(len_test)
    for i in range(len_test):
        prob = np.zeros(labelNum)
        for ele in labels:
            if ele == min(labels):
                prob[ele] = probDict[ele][i][0]
            elif ele == max(labels):
                prob[ele] = probDict[ele-1][i][1]
            else:
                prob[ele] = probDict[ele-1][i][1] - probDict[ele][i][1]
        y_pred[i] = np.argmax(prob)
    return y_pred


if __name__ == '__main__':
    X,y = datasets.load_iris(return_X_y=True)
    labelNum = len(set(y))
    labels = np.unique(y)
    AccList = []
    SKF = StratifiedKFold(n_splits=10,shuffle=True,random_state=0)
    for train_idx, test_idx in SKF.split(X, y):
        train_X, train_y = X[train_idx, :], y[train_idx]
        test_X, test_y = X[test_idx, :], y[test_idx]
        len_test = len(test_y)
        modelDict = BDOC_train(train_X,train_y)
        probDict = BDOC_prob(modelDict,test_X)
        # print(probDict[0])
        # print(probDict[0][1])
        # print(probDict[2][1][0])
        # print(probDict[2][1][1])
        y_pred = BDOC_pred(probDict, len_test, labelNum, labels)
        acc = accuracy_score(y_true=test_y,y_pred=y_pred)
        AccList.append(acc)
    print("Mean Acc =",np.mean(AccList))

 

不懂就问嘛~

想要类封装的版本，读者有需求再写~