* RUNS
*** Updated to Xavi's latest cvs on cluster and BeamformIt on 10/2/06

    | corpus | data set |   run | comment                                                          |
    |--------+----------+-------+------------------------------------------------------------------|
    | rt04s  | del4xs   |   63+ |                                                                  |

    _Runs below with new code: [[CVS tag of my code w/ new PCA]]_ 
    | corpus | data set    |   run | comment                                                     |
    |--------+-------------+-------+-------------------------------------------------------------|
    | rt05se | xc5         | 1598+ | 1416, no NR                                                 |
    | rt04s  | NRall.xc1.0 | 1655+ | 1.0s XC, 0.5s ER, new PCA all NR                            |
    | rt05se | NRall.xc1.0 | 1677+ | 1.0s XC, 0.5s ER, new PCA all NR                            |
    _below, removed feat3/4 merge stream feats so that remStuckFrms.m will get called_
    | devDD | broken speech tools 7226 | XC1ref, *, NR on, in case this matters for low dim pca input (dfw=1)              |
    | tstDD |              broken 7262 | XC1ref, *, NR on, in case this matters for low dim pca input (dfw=1)              |
    |-------+--------------------------+-----------------------------------------------------------------------------------|
    | dtDD  |                     6474 | dimPick, ent1G, entNLG                                                            |
    | dtDD  |                     6535 | dimPick, many new methods                                                         |

* Reoptimize
*** TODO correlation manifold kernel support vectors from combinations of diariation speaker location models
*** Split CCA w/ do_feat_weights==1				    :conclusion:
    
    _Conclusions:_ 
    1.) best perf is 13.42%, 
    2.) beats baseline, 
    3.) doesn't beat PCA-only dim reduction 
    4.) but no gain over CCA-ER or XC dim reduction

      Total Diariz Error vs. Run [merged 5720, dev_DD, merged 5737, test_DD]

|      x |   feat |                 |       |      | dev,5720 |   | tst,5737 |       |      |      x |
| runNum | 3_nmix | feat3_pairFeats | worst | best |      tot |   |      tot | worst | best | runNum |
|--------+--------+-----------------+-------+------+----------+---+----------+-------+------+--------|
|  05703 |      1 | XCfCCA.CC.1     | 40.84 | 3.04 |    18.21 |   |     15.5 | 40.58 | 5.42 |  05720 |
      
*** Summary Table

|        |       |      |       |   |       |       |      |        |                         |
|--------+-------+------+-------+---+-------+-------+------+--------+-------------------------|
|    dev |       |      |   dev |   |   tst |       |      |        | dev                     |
| runNum | worst | best |   tot |   |   tot | worst | best | runNum | note                    |
|--------+-------+------+-------+---+-------+-------+------+--------+-------------------------|
|  04677 | 49.26 | 2.75 | 18.04 |   | 17.29 | 46.08 | 2.96 |  04680 | baseline, dfw==0        |
|  05739 | 44.36 |  3.3 | 16.01 |   | 15.03 | 40.15 | 1.96 |  05746 | baseline, dfw==1        |
|  05386 | 36.36 | 2.58 | 15.17 |   | 13.03 |  24.3 | 3.65 |  05373 | XC-PCA                  |
|  05454 |  44.2 | 6.29 | 19.73 |   | 20.26 | 42.99 | 5.88 |  05455 | ER/XC                   |
|  05716 | 36.39 | 2.36 | 15.22 |   | 13.42 | 30.47 | 2.34 |  05733 | XCfCCA.dMI.3            |