1 minute read

This post is written by courtesy of:

以下假设 sample 有 m 个 examples。

The Residual Sum of Squares (RSS) is the sum of the squared residualsPermalink

以下三个概念等价 (我无话可说):

  • RSS: Residual Sum of Squares
  • SSR: Sum of Squared Residuals
  • SSE: Sum of Squared Errors
RSS=i=1mei2=i=1m(yif^(xi))2

The Mean Squared Error (MSE) is the mean of RSSPermalink

MSE=RSSm

The Root Mean Squared Error (RMSE) is the square root of MSEPermalink

RMSE=MSE=RSSm

The Residual Standard Error (RSE) is the square root of RSSdegrees of freedomPermalink

RSE=RSSmp1

where

  • p is the number of predictors
    • i.e. p+1 is the number of right-hand-side variables, including the intercept, in a regression model
  • mp1 denotes the degrees of freedom.
TSS=i=1m(yiy¯)2

where y¯ is the sample mean.

Further we have Var=TSSm1

R2 and Adjusted R2Permalink

R2=1RSSTSSAdjusted R2=1RSS/(mp1)TSS/(m1)=1m1mp1RSSTSS

Chain ReactionPermalink

当趋向 overfitting 时(比如 predictor 增多,模型变 flexible 时):

  • RSS ↓
    • MSE ↓
    • RMSE ↓
    • 如果是 predicator 增多,那么 RSE 无法断定是上升还是下降
  • TSS →
  • R2
  • Adjusted R2 不好说(这正是 adjustment 的体现)

Comments