MPO 581 Class number 11/27 Feb 23, 2011

Topic:

*Scripting (IDL one-liners) for automating plotting calls and EDA

*Correlation and line or curve (model) fitting. Hsieh Ch1 handed out.

*Variance decomposition I: multidimensional averages and deviations

Today's material:

Power tool for homework: if you can type it once, you can loop to build a string and do it N times.
Degrees of freedom, # independent samples, accounting for serial correlation or autocorrelation or oversampling, information content, etc. Fewest number of bins that can hold (most of) the variance.
Decomposition of variance (Analysis of Variance)

into orthogonal categories

Example: Variance decomposition into averages and deviations: 2D (t and p in atm. soundings)
Actually, any model + residual?
Example: sines and cosines

Note: To conserve well, you must use real variance, NOT the population estimator (N-1 version aka "sample variance").
I had to write my own:
    function variance_n, array
        return, mean(double(array)^2) - mean(double(array))^2
    end

    4. Line fitting by least squares vs. LAD applets from wikipedia page on Least absolute deviations (LAD)
    5. Correlation and dependence (wikipedia)
    6. Pearson correlation

HW2 - Multivariate Mayhem! datasets and tools updated.

Open questions, assignments, and loose ends for next class: