[BLAST_ANAWARE] lrn

From: Chris Crawford (chris2@lns.mit.edu)
Date: Sun Oct 26 2003 - 22:18:28 EST


hi chi & gang,
  don't know if you already beat me to the punch, but i finally got
'lrn' working. it turned out the be the second round of fitting after
pid(), so i just commented it out, and it seems to work fine.
  also, i compared what we have with v2_96, and the results looked very
promising. run 707, up to event 250, v3 got all of the tracks that
v2_96 got, with equal or better (or sometimes very close) chisq.
  finally here is a summary of why it missed some of the events (almost
all low p). In short, we have to be more carefull about cutting on the
direction of segments (they should point to another stub, not the the
origin!); also for some reason, TBLRecon::ScreenTrack() robbed a couple
of newton tracks. Finally, we should limit the number of segments
produced in a single chamber (especially fake stubs), and perhaps we
won't hit the max_track limit. chi, you're also working on ways to
reduce stub/segment multiplicites, right?
  anyways, i think BlastLib2 is pretty stable, now, and we are almost
ready to tag 'v3_0'. mabe just these small fixes, and test crunching a
large sample of runs tocompare statistically with v2_96.
--chris
ps. another enhancement: i modified 'configure.ac' (previously .in) to
work on all directories, ie. BlastLib2~v3, so you can have multiple
versions in your tree. And now BlastLib2 looks for a link to
'Blast_Params' in the 'pwd' before using $BLAST_PARAM, so you don't have
to keep switching the environment variable for each version.

run 707:
    4 both missed dirty (fake stub multiplicity, seg l0,2)
    61 both missed cosmic (traceback theta?)
    77 both missed lowp (Segment.Slope)
    85 both missed lowp (missed stub)
    104 both missed (Recon.ScreenTrack)
    135 both missed lowp (Recon.ScreenTrack, missed best seg l1,2)
    149 both missed lowp (no fake seg l1, missed best stub,seg l2)
    156 both missed p-p (missed fake stub, emap?)
    159 both missed lowp (Segment.Slope)
    189 both missed lowp (Segment.Slope)



This archive was generated by hypermail 2.1.2 : Mon Feb 24 2014 - 14:07:30 EST