kaggle 各种评价指标之一 :Error Metrics for Regression Problems 回归问题错误度量

基本上必须看一遍，顺便简单翻译一下：

1 MAE

统计上，MAE是一个用来衡量预测和最终结果之间的接近程度的数量

In statistics, the mean absolute error (MAE) is a quantity used to measure how close forecasts or predictions are to the eventual outcomes. The mean absolute error is given by

MAE=1n∑i=1n|yi−yi^|=1n∑i=1n|ei|.

Where

AE=|ei|=|yi−yi^|

Actual=yi

Predicted=yi^

R code:

MAE <- sum(abs(y-y_pred)) / length(y)
language: ruby

Competitions using this metric:

• https://www.kaggle.com/c/how-much-did-it-rain-ii
• https://www.kaggle.com/c/allstate-claims-severity

2,Weighted Mean Absolute Error 带权重的MAE

This is the weighted average of absolute errors:

WMAE=1n∑i=1nwi|yi−y^i|

Example Solution/Submission: 这个例子似乎有问题

Example solution file:

id,val,weight,indicator
1,10,.1,Public
2,200,.01,Public
3,3000,.001,Public
4,40000,.0001,Public
5,5,1,Private
6,0.6,10,Private
7,.07,100,Private
language: php

Example submission:

id,val
1,11
2,200
3,3300
4,44000
5,5.5
6,0.66
7,0.077
language: ruby

Then:

PublicScore = (1.0/(4.0))*((.1*(11 - 10)) + (.01*(200 - 200)) + (.001*(3300 - 3000)
PrivateScore = (1.0/(3.0))*(1*(5.5 - 5.0) + 10*(0.66 - 0.6) + 100*(0.077 - .07));
language: perl

Implementation of this metric:

Code

Competitions using this metric:

• https://www.kaggle.com/c/benchmark-bond-trade-price-challenge

3  Root Mean Squared Error

未定义

4   Root Mean Squared Logarithmic Error

The RMSLE is calculated as  计算公式

ϵ=1n∑i=1n(log(pi+1)−log(ai+1))2−−−−−−−−−−−−−−−−−−−−−−−−−−√

Where:

ϵ  is the RMSLE value (score)

n is the total number of observations in the (public/private) data set,

pi is your prediction,

andai is the actual response for i. log(x) is the natural logarithm of x

Notes:

• RMSLE penalizes an under-predicted estimate greater than an over-predicted estimate
  
• RMSLE 惩罚  under-predicted estimate 大于over-predicted estimate
  

Competitions using this metric:

https://www.kaggle.com/c/yelp-recruiting/details/evaluation

https://www.heritagehealthprize.com/c/hhp/details/Evaluation

https://www.kaggle.com/c/online-sales/details/Evaluation

https://www.kaggle.com/c/walmart-recruiting-sales-in-stormy-weather

https://www.kaggle.com/c/bike-sharing-demand

https://www.kaggle.com/c/caterpillar-tube-pricing

https://www.kaggle.com/c/grupo-bimbo-inventory-demand