Log of /trunk/roofit/roostats/src/AsymptoticCalculator.cxx
Parent Directory
Revision
47326 -
(
view)
(
download)
(
as text)
(
annotate)
-
[select for diffs]
Modified
Thu Nov 15 17:25:55 2012 UTC (2 years, 2 months ago) by
moneta
File length: 57859 byte(s)
Diff to
previous 47005
Fix in Asymptoticcalculator the case when the number of bins in the observables are not enough.
Add a parameter in the constructor to pass a new value of bins for making the Asimov data set
Revision
47005 -
(
view)
(
download)
(
as text)
(
annotate)
-
[select for diffs]
Modified
Fri Nov 2 14:56:00 2012 UTC (2 years, 2 months ago) by
moneta
File length: 57063 byte(s)
Diff to
previous 46094
Fix AsymptoticCalculator to change in RooStats utils to not return always a product in case of one single nuisance pdf.
Try to use in AsymptoticCalculator the same binning for Asimov data set as for observed data sets. This fixes a problem reported by a user
with a binned model.
Revision
45684 -
(
view)
(
download)
(
as text)
(
annotate)
-
[select for diffs]
Modified
Wed Aug 22 10:03:46 2012 UTC (2 years, 5 months ago) by
moneta
File length: 54329 byte(s)
Diff to
previous 45143
fix a bug in the AsymptoticCalculator in setting qmu-tilde for computing limit (thanks to M. Baak who reported this). The bug affects only the first scanned point
Revision
45143 -
(
view)
(
download)
(
as text)
(
annotate)
-
[select for diffs]
Modified
Thu Jul 19 23:16:50 2012 UTC (2 years, 6 months ago) by
moneta
File length: 54344 byte(s)
Diff to
previous 44768
apply patch from Gabriel:
fixes the SimpleLikelihoodRatio by removing constraints terms from pdf. Add in RooStatsUtils funciton to factorize pdf and remove constraint (MakeUNconstrainedPdf)
Remove also some reduntant code in AsymptoticCalculator and use RooStats utils
add in stressRooStats new tests of SimpleLikelihoodRatio and update reference file
Revision
44768 -
(
view)
(
download)
(
as text)
(
annotate)
-
[select for diffs]
Modified
Wed Jun 27 14:26:15 2012 UTC (2 years, 6 months ago) by
moneta
File length: 54981 byte(s)
Diff to
previous 44366
do not call in AsymptoticCalculator SetBackgroundAsAlt(true) since it does not applies in case of hypothesis test
(for limits it is called inside the inverter)
add a tolerance for checking qhen qmu is less than zero for numerical errors in computation of qmu
Revision
44366 -
(
view)
(
download)
(
as text)
(
annotate)
-
[select for diffs]
Modified
Wed May 30 15:32:27 2012 UTC (2 years, 7 months ago) by
axel
File length: 54854 byte(s)
Diff to
previous 44339
Remove #include "Riostream.h" from roofitcore/inc/RooCacheManager.h, which pulls in using namespace std.
Deal with the consequences.
Revision
44266 -
(
view)
(
download)
(
as text)
(
annotate)
-
[select for diffs]
Modified
Wed May 16 12:32:18 2012 UTC (2 years, 8 months ago) by
moneta
File length: 54836 byte(s)
Diff to
previous 44156
apply some Coverity fixes
improve some error messaging in AsymptoticCalculator and automatically set onesideddiscovery if null snapshot is equal to minimum of parameter at construction time
Revision
44156 -
(
view)
(
download)
(
as text)
(
annotate)
-
[select for diffs]
Modified
Mon May 7 16:37:38 2012 UTC (2 years, 8 months ago) by
moneta
File length: 53515 byte(s)
Diff to
previous 44152
- AsymptoticCalculator:: do a fix for bins where pdf is zero when generating Asimov data sets
- HypoTestInverterResult: another fix for lower/upper limit search
Revision
44126 -
(
view)
(
download)
(
as text)
(
annotate)
-
[select for diffs]
Modified
Fri May 4 16:55:47 2012 UTC (2 years, 8 months ago) by
moneta
File length: 53270 byte(s)
Diff to
previous 44101
- fix in AsymptoticCalculator the formulae for the two-sided test statistics for both observed and expected p values
- improve search for upper/lower limit in InverterResult
add also a flag to indicate in the InverterResult if it comes from using a two sided test statistic.
This is needed to use the right formula for the asymptotic case.
- Add also possibility to skip and to not draw the points where there is a failing fit or any other error
- Add in Profile likelihood possibility to query if the test statistic is one or two sided
Revision
44101 -
(
view)
(
download)
(
as text)
(
annotate)
-
[select for diffs]
Modified
Thu May 3 16:38:46 2012 UTC (2 years, 8 months ago) by
moneta
File length: 51498 byte(s)
Diff to
previous 44059
fix asymptotic calculator for 2-sided test statistic
and for one sided discovery
- add also method to retrieve the global best fit parameters
Revision
44059 -
(
view)
(
download)
(
as text)
(
annotate)
-
[select for diffs]
Modified
Wed May 2 07:38:48 2012 UTC (2 years, 8 months ago) by
moneta
File length: 48208 byte(s)
Diff to
previous 44034
- BayesianCalculator: do not require anymore to have a prior pdf when using an uniform prior
Make it default case as in case of MCMCCalculator
- AsymptoticC. : fix a comment
Revision
43189 -
(
view)
(
download)
(
as text)
(
annotate)
-
[select for diffs]
Modified
Thu Mar 1 11:16:40 2012 UTC (2 years, 10 months ago) by
moneta
File length: 48326 byte(s)
Diff to
previous 43118
- fix the restoring of log messages in case the NLL was not fitted
- fix generation of Asimov data for simple (not-extended) Gaussian models
Revision
42833 -
(
view)
(
download)
(
as text)
(
annotate)
-
[select for diffs]
Modified
Tue Jan 31 16:07:14 2012 UTC (2 years, 11 months ago) by
moneta
File length: 45309 byte(s)
Diff to
previous 42829
add option in the constructor to use the nominal Asimov data set (i.e. the one where nuisance have nominal values when generating the data set)
Revision
42828 -
(
view)
(
download)
(
as text)
(
annotate)
-
[select for diffs]
Modified
Mon Jan 30 16:55:11 2012 UTC (2 years, 11 months ago) by
moneta
File length: 44273 byte(s)
Diff to
previous 42730
implement a new version of MakeAsimovData where one passes just model and parameter values.
In this case the nuisance parameters are not set to the fitted value from the observed data but they are kept
at their nominal values
Implement the old method using the new one
Revision
42730 -
(
view)
(
download)
(
as text)
(
annotate)
-
[select for diffs]
Modified
Thu Jan 19 14:40:31 2012 UTC (3 years ago) by
moneta
File length: 42521 byte(s)
Diff to
previous 42603
fix checking of status code in case one is running Improve.
Fix also loops on trying the fit when evaluating the PL test statistics to avoid
re-fitting with the same conditions
Revision
42603 -
(
view)
(
download)
(
as text)
(
annotate)
-
[select for diffs]
Modified
Fri Jan 13 08:31:20 2012 UTC (3 years ago) by
moneta
File length: 42070 byte(s)
Diff to
previous 42482
fix a bug in making the Asimov data set (in setting t he global observables to the right value) with the workspace generated with the new version of the Histfactory
In this case the constraint terms are of like Gaussian( glob_obs, func( nuis_param) ), and therefore the global observable must be set to
glob_obs_ASIMOV = func(nuis_param_MLE ) and not glob_ASIMOV = nuis_param_MLE as it was done before
Revision
42482 -
(
view)
(
download)
(
as text)
(
annotate)
-
[select for diffs]
Modified
Tue Dec 13 17:35:39 2011 UTC (3 years, 1 month ago) by
moneta
File length: 38761 byte(s)
Diff to
previous 42324
fix to set back the of poi to non-const after having create the Asimov data set.
The initial Asimov data fit is still done with a fixed poi value (e.g. mu=0)
- when fitfing for finding the best conditional nuisance parameters for the Asimov data set, use the user default minimizer
Revision
42258 -
(
view)
(
download)
(
as text)
(
annotate)
-
[select for diffs]
Modified
Fri Nov 25 20:57:12 2011 UTC (3 years, 1 month ago) by
moneta
File length: 38266 byte(s)
Diff to
previous 41995
- change MakeAsimovData to be a static function which can be used outside the class
- fix a bug when models do not have nuisance parameters
This form allows you to request diffs between any two revisions of this file.
For each of the two "sides" of the diff,
enter a numeric revision.
