Principal Component Analysis

• An intro talk to PCA: ~ccohalan/references/PCA_mini.pdf or .sxi (to open a presentation in openoffice, type ooimpress filename at the command line)

• The PCA code that was used:
  • ~ccohalan/wtl_timestream_pipeline/timestream_processing/pca_tc.pro. I think there's enough comments in the code to give you an idea of what's what
  • Things like wtl.../timestream_processing/pca_prep.pro and wtl.../utilities/create_2d_from_3d.pro can help if you want to perform tests. Note that create_bad_pca_array (in pca_prep) will be changed to create_scans once this program is ready.

• Results: All plots can be found in ~ccohalan/plots/PCA-tests. Most notably:
  • 405-pca_1-1-1_wbolos.ps reproduce Ruediger's plots (http://bolo.berkeley.edu/internal/apexsz/apex_online_log/index.php/2006-05-25, first link (PS file)) Note that the difference between our plots coincides with glitches. I'm working on that.
  • Tests on simulated data:
    • 8 bolos, 2 different atmospheres (4 bolos each atmo), signal in 1 bolo: sim_data_0to3pc.ps
    • 9 bolos, lag in atmosphere change (groups of 3 bolos, so 2 lags from reference group) : sim_data_lag.pro
  • 405, Replacing giltches by other values:
    • Replace the glitches by zeros: 405-1pca-glitchreplacedbyzero.ps
    • Replace the glitches by a line joining both tips of the glitch: 405-1pca-glitchreplacedbyav.ps
    • Replace glitches at 3 different times : 405-glitchdiff-3types.ps ---> Explanation: (1st column : right at first, put glitched data = ADC count right before glitch, 2nd column : put glitch = 0, go through PCA, then put spikes (due to glitch=0) = ADC count right before glitch, 3rd column : go through PCA with glitch, then put timesteps with glitch = ADC count before glitch)

-- ClaireCohalan - 21 Jun 2006

---

This topic: Main > TWikiUsers > ClaireCohalan > PrincipalComponentAnalysis Topic revision: r1 - 2006-06-21 - ClaireCohalan

---