Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Robust estimation methods • Robust estimation methods are relatively insensitive to mean X “bad” data. median XM • Example: using median rather than mean: N 2 • Mean X minimizes the sum X satisfies (Xi A) 0 A i1 of squared errors (SSE): • Median XM minimizes the N XM satisfies sum of absolute errors Xi A 0 A i1 (SSE): • Proof: use Heaviside step 1, x > 0 function: 1 –1 H(x) 0, x 0 1, x 0 H' (x) 2 (x) How the median minimises SAE • Need to show that the median minimises the expected value of the sum of absolute errors: X X M X X M H X X M X M X XM X M X XM H X X M H X X M X X M H' X X M P(X X M ) P(X X M ) 0 when X M is the median. 0 (since X-XM=0 when H’ = 1 and H’=0 for all other values of X - XM) Mean and median • The median is less sensitive to outliers than the mean. Mean Median • Although the median is unbiased it is not a minimum-variance estimator. • Note how standard deviation of the median varies with sample size in comparison to standard deviation of the mean. Median Mean Evaluation of median without sorting Xi XM Xi XM 1 0 -1 • A useful application: N Since i1 Xi XM 0, first make a guess at XM . Xi XM N Then estimate a new XM i1 N i1 and iterate to convergence. Xi Xi XM , 1 Xi XM Median filtering and sigma-clipping • Median filtering: take window of N points • Replace central point by median of the N points. Window • Sigma-clipping: – – – – Fit all points by minimising 2 Set threshold K and check for outliers at ± K or more Repeat fit omitting largest outlier Iterate until set of rejected points converges. Reject Reject Using 2min to reject models • Suppose we fit M parameters to N data points: 2 min yi f (xi ; P1 ...PM ) 2 2 ~ N M i i1 2 NM N NM ( N2 M ) 2(N M) • We use N-M because an N - parameter fit should fit N points exactly. • If model is good, then the best-fit 2min should be: 2 min N M 2(N M) What if 2min is too high? • Several possibilities to consider: – Statistical fluke – use 2 distribution to estimate probability – Wrong model – use 2 distribution to reject model – Right model, but additional (nuisance) parameters not correctly chosen: x – Error bars too small. Re-scale. • The third possibility adds a constant to 2min . • Can still use 2min +1 to set 1- confidence intervals on parameter values. x Diagnosis of 2min ≠ N-M ±√2(N-M) N–M 2=1 2min N–M 2= 2min /(N–M) >1 2= 2min /(N–M) < 1 2min N–M ˆ 2min > N–M due to unimportant parameters omitted from model ˆ 2min > N–M due to underestimation of error bars 2min < N–M due to overestimation of error bars If this happens use these values to estimate errors on parameters