1 / 15

Modeling Data

Modeling Data. Greg Beckham. Bayes. Fitting Procedure should provide Parameters Error estimates on the parameters A statistical measure of goodness of fit Without a statistical measure of goodness of fit, the model is most likely useless. Bayes ’ Theorem.

weldon
Download Presentation

Modeling Data

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Modeling Data Greg Beckham

  2. Bayes • Fitting Procedure should provide • Parameters • Error estimates on the parameters • A statistical measure of goodness of fit • Without a statistical measure of goodness of fit, the model is most likely useless

  3. Bayes’ Theorem • Bayes’ Theorem relates the conditional probability of two events • P(A|B) = P(A) * (P(B|A)/P(B)) • P(A|B) is the probability of A given B • P(A|B) = (P(A) ^ P(B))/P(B) • 0 <= P(A|B) <= 1 • P(A) and P(B) are unconditional probabilities

  4. Example • Suppose population of 60% boys and 40% girls • Girls wear pants and skirts with equal probability, boys always wear pants • If a student is wearing pants (B), what is the probability the student is a girl (A)? • P(B|A) = .5 • P(A) = .4 • P(B) = .5 * .4 + 1 * .6 = .8 • P(A|B) = P(B|A)P(A)/P(B) = (.5 * .4)/.8 = .25

  5. Bayes’ Theorem • Defs • H, some hypothesis • I, collective background data • Can assign some plausibility to H even before we have explicit data • P(H|I) • Bayesian Prior

  6. Bayes’ Theorem • When data is added • When more data is added • Using the chain rule Shows you get the same answer if the data came together

  7. Questions

  8. Linear Least Squares • Fitting N data points (xi, yi), t = 0, …, N-1 to a model that has M adjustable parameters • Functional relationship between measured independent and dependant variables • Minimize of a0…aM-1

  9. Linear Least Squares • How to derive linear least squares • Start by asking Given a particular set of parameters, what is the probability that the data set should have occurred • Plus or minus some fixed Δy • Each data point yi has a measurement error that is independently random and distributed as a normal distribution around true model y(x) • Standard deviations of these are σ

  10. Linear Least Squares • Probability of the data set is the product of the probabilities at each point 15.1.2

  11. Least Squares • Notice the Δy in each term • So the goal is to minimize the error • Where P(model) = P(a0…aM-1) is our prior probability distribution on al models

  12. Least Squares • Take logarithm of 15.1.2 to get • Since N, Δy, σ are constants, minimizing this is equivalent to minimizing 15.1.2

  13. Non-Linear Fit

  14. Non-linear Fit • Fitting depends non-linearly on the set of unknown parameters ak, k = 0,1,…,M-1 • Define a merit function χ2 (chi-squared) and determine best-fit parameters by minimizing the function • Minimization must be done iteratively until it stops decreasing (only changes by some small amount)

  15. Non-linear Fit • Where d is an M-vector and D is an MxM matrix • D is the second derivative of χ2 at any a. • If we are close then • If it is a poor approximation then step down the gradient

More Related