評估一個預測模型性能一般都有那些指標

對於不一樣類型的模型，會有不一樣的評估指標，那麼咱們從最直接的迴歸和分類這兩個類型，對於結果連續的迴歸問題，

通常使用的大體爲:MSE(均方差),MAE（絕對平均差）,RMSE(根均方差)這三種評估方法，這三種方式公式此處補貼出來。

對於離散的分類問題，咱們通常看ROC曲線，以及AUC曲線，通常好的模型，ROC曲線，在一開始就直接上升到1,而後一直保持1，也就是使得AUC=1.0或者儘量的讓其

接近這個值，這是咱們奮鬥的目標.

摘個實際的例子：--出自《預測分析核心算法》這本書.

 1 #-*-coding:utf-8-*-
 2 __author__ ='gxjun'
 3 import pandas as pd
 4 import matplotlib.pyplot as plt
 5 from pandas import DataFrame
 6 from random import uniform
 7 import math 
 8 import numpy as np
 9 import random 
10 import pylab as pl
11 from sklearn import datasets,linear_model
12 from sklearn.metrics import roc_curve ,auc
13 
14 
15 ##計算RP值
16 def confusionMatrix(predicted ,actual , threshold):
17     if len(predicted) != len(actual):
18         return -1;
19     tp=0.0;
20     fp=0.0;
21     tn=0.0;
22     fn=0.0;
23     for i in range(len(actual)):
24         if actual[i] >0.5:
25             if predicted[i] > threshold:
26                 tp+=1.0;
27             else:
28                 fn+=1.0;
29         else:
30             if predicted[i]<threshold:
31                 tn+=1.0;
32             else:
33                 fp+=1.0;
34     rtn=[tp,tn,fp,fn];
35     return rtn;
36 target_url =("https://archive.ics.uci.edu/ml/machine-learning-databases/undocumented/connectionist-bench/sonar/sonar.all-data")
37 data = pd.read_csv(target_url,header=None,prefix='V');
38 print('-'*80)
39 print(data.head())
40 print('-'*80)
41 print(data.tail())
42 print('-'*80)
43 print(data.describe())
44 print('-'*80)
45 label = [];
46 dataRows = [];
47 
48 for i in range(208):
49     if data.iat[i,-1]=='M':
50         label.append(1.0);
51     else:
52         label.append(0);
53 print label        
54 dataRows=data.iloc[:,0:-1];
55 x_train = np.array(dataRows);
56 y_train = np.array(label);
57 print "x_train shape: {} , y_train shape: {}".format(x_train.shape,y_train.shape);
58 print "x_test shape: {} , y_test shape: {}".format(x_test.shape,y_test.shape);
59 x_test = np.array(dataRows[0:int(208/3)]);
60 y_test = np.array(label[0:int(208/3)]);
61 #train model
62 rockModel = linear_model.LinearRegression();
63 rockModel.fit(x_train,y_train);
64 prob = rockModel.predict(x_train);
65 print('-'*80);
66 confusionMatrain = confusionMatrix(prob,y_train,threshold=0.5);
67 
68 #print confusionMatrain
69 fpr ,tpr,threshold = roc_curve(y_train,prob);
70 roc_auc = auc(fpr,tpr);
71 
72 plt.clf();
73 plt.plot(fpr,tpr,label='ROC curve(area =%0.2f)'%roc_auc);
74 pl.plot([0,1],[0,1],'k-');
75 pl.xlim([0.0,1.0]);
76 pl.ylim([0.0,1.0]);
77 pl.xlabel("FP rate}");
78 pl.ylabel("TP rate}");
79 pl.title("ROC");
80 pl.legend(loc="lower right");
81 pl.show()

結果爲：