Ticket #697 (closed defect: fixed)
Opened 2012-04-09T15:34:35-05:00
Last modified 2012-06-15T18:42:51-05:00
Sometimes TRI2 and SLIMPlugin results agree, except for chi-square
| Reported by: | aivar | Owned by: | aivar |
|---|---|---|---|
| Priority: | major | Milestone: |
|
| Component: | slim-plugin | Severity: | serious |
| Keywords: | Cc: | ||
| Blocked By: | Blocking: | #674 |
Description
Sometimes the results are in agreement, but the reduced chi-square is not in agreement.
Change History
comment:1 Changed 2012-04-09T15:41:39-05:00 by aivar
comment:2 Changed 2012-04-09T17:30:14-05:00 by aivar
Another point (155, 230) exhibits the same behavior, results agree except for chisquare for Poisson Data and Poisson Fit.
comment:3 Changed 2012-04-09T18:18:08-05:00 by aivar
Note this bug covers fitting without an Excitation curve loaded.
comment:5 Changed 2012-06-12T11:21:50-05:00 by aivar
100611-YG-256.sdt at (146, 230) cursors 0.0/2.15/11.23 Poisson Data:
SLIM Plugin: A 71.28562 T 2.092129 Z 0.99059045 X2 0.32557997 TRI2: A 69.08 T 2.13 Z 0.92 X2 1.007875
In other words for TRI2 cursors 0.0/2.1484/11.2305 and 0.0/2.15/11.23 are equivalent, for SLIM Plugin they are not. Suggests a rounding technique discrepancy.
comment:6 Changed 2012-06-12T11:28:12-05:00 by aivar
Comment to last comment:
I went from 0.0/2.1484/11.2305 to 0.0/2.15/11.23 because TRI2 is now truncating cursors to 2 places. (Odd because I used to enter more digits--this is with my XP VM and TRI2 Version 2.5.2.1 FI_port.)
comment:7 Changed 2012-06-12T15:19:46-05:00 by aivar
With my Windows 7 VM and TRI2 version 2.731 I get:
0/2.15/11.23 A 71.286 T 2.092 Z 0.991 X2 1.015688
0/2.1484/11.2305 A 69.081 T 2.135 Z 0.922 X2 1.007875
So "for TRI2 cursors 0.0/2.1484/11.2305 and 0.0/2.15/11.23 are equivalent" from two comments above only applies for version 2.5.2.1 (on XP). They are not equivalent for 2.731 (on Windows 7). In other words latest TRI2 is consistent with SLIM Plugin, except for this chisquare being off issue.
comment:8 Changed 2012-06-12T16:43:33-05:00 by aivar
- Summary changed from Sometime TRI2 and SLIMPlugin results agree, except for chi-square to Sometimes TRI2 and SLIMPlugin results agree, except for chi-square
comment:9 Changed 2012-06-15T18:42:51-05:00 by aivar
- Status changed from assigned to closed
- Resolution set to fixed
"Chisquare_delta" was off and causing these problems. Cursor rounding discrepancies do not appear to exist with recent SP and TRI2 2.7.3.1.
For 100611-YG-256.sdt which I got from Jimmy Fong, with cursors at 0.0/2.1484/11.2305 for point (146, 230) RLD/LMA fit I get:
Gaussian noise is good
SLIM Plugin A 73,23523 T 2.0873063 Z 0.94988865 X2 0.85290354
TRI2 A 73.24 T 2.09 Z 0.95 X2 0.852904
Poisson Data is bad
SLIM Plugin A 69.08094 T 2.1349077 Z 0.9221971 X2 0.32471284 X
TRI2 A 69.08 T 2.13 Z 0.92 X2 1.007875 X
Poisson Fit is bad
SLIM Plugin A 75.977646 T 2.0332031 Z 1.039639 X2 0.33459756 X
TRI2 A 75.98 T 2.03 Z 1.04 X2 0.851691 X
Maximum Likelihood Estimation is good
SLIM Plugin A 73.182625 T 2.0885024 Z 0.9476004 X2 0.9309332
TRI2 A 73.19 (bit of a rounding error?) T 2.09 Z 0.95 X2 0.930933 TRI2