使用sklearn对多分类的每个类别进行指标评价操作
时间：2020-10-03 10:16:38 | 栏目：Python代码 | 点击：次
今天晚上，笔者接到客户的一个需要，那就是：对多分类结果的每个类别进行指标评价，也就是需要输出每个类型的精确率（precision），召回率（recall）以及F1值（F1-score）。
对于这个需求，我们可以用sklearn来解决，方法并没有难，笔者在此仅做记录，供自己以后以及读者参考。
我们模拟的数据如下：
y_true = ['北京', '上海', '成都', '成都', '上海', '北京', '上海', '成都', '北京', '上海']
y_pred = ['北京', '上海', '成都', '上海', '成都', '成都', '上海', '成都', '北京', '上海']
其中y_true为真实数据，y_pred为多分类后的模拟数据。使用sklearn.metrics中的classification_report即可实现对多分类的每个类别进行指标评价。
示例的Python代码如下：
# -*- coding: utf-8 -*- from sklearn.metrics import classification_report y_true = ['北京', '上海', '成都', '成都', '上海', '北京', '上海', '成都', '北京', '上海'] y_pred = ['北京', '上海', '成都', '上海', '成都', '成都', '上海', '成都', '北京', '上海'] t = classification_report(y_true, y_pred, target_names=['北京', '上海', '成都']) print(t)
输出结果如下：
precision recall f1-score support
北京 0.75 0.75 0.75 4
上海 1.00 0.67 0.80 3
成都 0.50 0.67 0.57 3
accuracy 0.70 10
macro avg 0.75 0.69 0.71 10
weighted avg 0.75 0.70 0.71 10
需要注意的是，输出的结果数据类型为str，如果需要使用该输出结果，则可将该方法中的output_dict参数设置为True，此时输出的结果如下:
{‘北京': {‘precision': 0.75, ‘recall': 0.75, ‘f1-score': 0.75, ‘support': 4},
‘上海': {‘precision': 1.0, ‘recall': 0.6666666666666666, ‘f1-score': 0.8, ‘support': 3},
‘成都': {‘precision': 0.5, ‘recall': 0.6666666666666666, ‘f1-score': 0.5714285714285715, ‘support': 3},
‘accuracy': 0.7,
‘macro avg': {‘precision': 0.75, ‘recall': 0.6944444444444443, ‘f1-score': 0.7071428571428572, ‘support': 10},
‘weighted avg': {‘precision': 0.75, ‘recall': 0.7, ‘f1-score': 0.7114285714285715, ‘support': 10}}
使用confusion_matrix方法可以输出该多分类问题的混淆矩阵，代码如下：
from sklearn.metrics import confusion_matrix y_true = ['北京', '上海', '成都', '成都', '上海', '北京', '上海', '成都', '北京', '上海'] y_pred = ['北京', '上海', '成都', '上海', '成都', '成都', '上海', '成都', '北京', '上海'] print(confusion_matrix(y_true, y_pred, labels = ['北京', '上海', '成都']))
输出结果如下：
[[2 0 1] [0 3 1] [0 1 2]]
为了将该混淆矩阵绘制成图片，可使用如下的Python代码：
# -*- coding: utf-8 -*-
# author: Jclian91
# place: Daxing Beijing
# time: 2019-11-14 21:52
from sklearn.metrics import confusion_matrix
import matplotlib.pyplot as plt
import matplotlib as mpl
# 支持中文字体显示, 使用于Mac系统
zhfont=mpl.font_manager.FontProperties(fname="/Library/Fonts/Songti.ttc")
y_true = ['北京', '上海', '成都', '成都', '上海', '北京', '上海', '成都', '北京', '上海']
y_pred = ['北京', '上海', '成都', '上海', '成都', '成都', '上海', '成都', '北京', '上海']
classes = ['北京', '上海', '成都']
confusion = confusion_matrix(y_true, y_pred)
# 绘制热度图
plt.imshow(confusion, cmap=plt.cm.Greens)
indices = range(len(confusion))
plt.xticks(indices, classes, fontproperties=zhfont)
plt.yticks(indices, classes, fontproperties=zhfont)
plt.colorbar()
plt.xlabel('y_pred')
plt.ylabel('y_true')
# 显示数据
for first_index in range(len(confusion)):
for second_index in range(len(confusion[first_index])):
plt.text(first_index, second_index, confusion[first_index][second_index])
# 显示图片
plt.show()
生成的混淆矩阵图片如下：
补充知识：python Sklearn实现xgboost的二分类和多分类
二分类：
train2.txt的格式如下：
import numpy as np
import pandas as pd
import sklearn
from sklearn.cross_validation import train_test_split,cross_val_score
from xgboost.sklearn import XGBClassifier
from sklearn.metrics import precision_score,roc_auc_score
min_max_scaler = sklearn.preprocessing.MinMaxScaler(feature_range=(-1,1))
resultX = []
resultY = []
with open("./train_data/train2.txt",'r') as rf:
train_lines = rf.readlines()
for train_line in train_lines:
train_line_temp = train_line.split(",")
train_line_temp = map(float, train_line_temp)
line_x = train_line_temp[1:-1]
line_y = train_line_temp[-1]
resultX.append(line_x)
resultY.append(line_y)
X = np.array(resultX)
Y = np.array(resultY)
X = min_max_scaler.fit_transform(X)
X_train,X_test, Y_train, Y_test = train_test_split(X,Y,test_size=0.3)
xgbc = XGBClassifier()
xgbc.fit(X_train,Y_train)
pre_test = xgbc.predict(X_test)
auc_score = roc_auc_score(Y_test,pre_test)
pre_score = precision_score(Y_test,pre_test)
print("xgb_auc_score:",auc_score)
print("xgb_pre_score:",pre_score)
多分类：有19种分类其中正常0，异常1~18种。数据格式如下：
# -*- coding:utf-8 -*-
from sklearn import datasets
from sklearn.multiclass import OneVsRestClassifier
from sklearn.svm import LinearSVC
from sklearn.cross_validation import train_test_split,cross_val_score
from sklearn.svm import SVC
from sklearn.linear_model import LogisticRegression
from xgboost.sklearn import XGBClassifier
import sklearn
import numpy as np
from sklearn.preprocessing import OneHotEncoder
from sklearn.metrics import precision_score,roc_auc_score
min_max_scaler = sklearn.preprocessing.MinMaxScaler(feature_range=(-1,1))
resultX = []
resultY = []
with open("../train_data/train_multi_class.txt",'r') as rf:
train_lines = rf.readlines()
for train_line in train_lines:
train_line_temp = train_line.split(",")
train_line_temp = map(float, train_line_temp) # 转化为浮点数
line_x = train_line_temp[1:-1]
line_y = train_line_temp[-1]
resultX.append(line_x)
resultY.append(line_y)
X = np.array(resultX)
Y = np.array(resultY)
#fit_transform(partData)对部分数据先拟合fit，找到该part的整体指标，如均值、方差、最大值最小值等等（根据具体转换的目的），然后对该partData进行转换transform，从而实现数据的标准化、归一化等等。。
X = min_max_scaler.fit_transform(X)
#通过OneHotEncoder函数将Y值离散化成19维，例如3离散成000000???100
Y = OneHotEncoder(sparse = False).fit_transform(Y.reshape(-1,1))
X_train,X_test, Y_train, Y_test = train_test_split(X,Y,test_size=0.2)
model = OneVsRestClassifier(XGBClassifier(),n_jobs=2)
clf = model.fit(X_train, Y_train)
pre_Y = clf.predict(X_test)
test_auc2 = roc_auc_score(Y_test,pre_Y)#验证集上的auc值
print ("xgb_muliclass_auc:",test_auc2)