Home | News | Documentation | Download

Strage value with RooStats::AsymptoticCalculator


#1

I’m trying the tutorial StandardHypoTestInvDemo using the profile likelihood and the three methods: the frequentist and the hybrid gives similar result, but the hybrid has some problems. I attach the screenshot.

.x StandardHypoTestInvDemo.C(0, "combined", "ModelConfig", "", "obsData", 2, 3)
.x StandardHypoTestInvDemo.C(0, "combined", "ModelConfig", "", "obsData", 1, 3)
.x StandardHypoTestInvDemo.C(0, "combined", "ModelConfig", "", "obsData", 0, 3)



#2

Hi,

Can you please attach the workspace so I can debug your problem ? It is probably due to some failures in fitting the model,

Lorenzo


#3

[quote=“moneta”]Hi,

Can you please attach the workspace so I can debug your problem ? It is probably due to some failures in fitting the model,

Lorenzo[/quote]

I’m using the one provided with the tutorial, it is automatically generated the first time you run the tutorial


#4

Hi,

By running on the tutorial workspace the asymptotic calculator :

.x StandardHypoTestInvDemo.C(0, "combined", "ModelConfig", "", "obsData", 2, 3)

I get a reasanable result (see attached plot) with 5.32 patches or the trunk, but it is true I get a result like yours if I use just 5.32. It is probably due to a bug fix I have applied two days ago.

Best Regards

Lorenzo



#5

[quote=“moneta”]Hi,

By running on the tutorial workspace the asymptotic calculator :

.x StandardHypoTestInvDemo.C(0, "combined", "ModelConfig", "", "obsData", 2, 3)

I get a reasanable result (see attached plot) with 5.32 patches or the trunk, but it is true I get a result like yours if I use just 5.32. It is probably due to a bug fix I have applied two days ago.

Best Regards

Lorenzo[/quote]

thank you. Do you plan a release before Christmas? can you link me the change in the code?


#6

Hi,
After checking in more detail, the problem is still there in the latest version. I was using a workspace done with the old HistFactory version. When using a new workspace done with the new HistFactory of 5.32 the Asymptotic calculator does not work anymore because the fits fail. The workspace are different and they contain in addition some extra nuisance parameters (a statistical error for each bin)
I will investigate this problem,

Best Regards

Lorenzo


#7

[quote=“moneta”]Hi,
After checking in more detail, the problem is still there in the latest version. I was using a workspace done with the old HistFactory version. When using a new workspace done with the new HistFactory of 5.32 the Asymptotic calculator does not work anymore because the fits fail. The workspace are different and they contain in addition some extra nuisance parameters (a statistical error for each bin)
I will investigate this problem,

Best Regards

Lorenzo[/quote]

thank you. About the fitting problem, can this problem be related with my other post? In that case it work with 5.28 and 5.33, but not with 5.32


#8

Hi,

This bug in the AsymptoticCalculator is now fixed in 5.32 patches and trunk. Thank you for reporting this

Best Regards

Lorenzo


#9

thank you very much, can you give me the link to the changeset?

[quote=“moneta”]Hi,

This bug in the AsymptoticCalculator is now fixed in 5.32 patches and trunk. Thank you for reporting this

Best Regards

Lorenzo[/quote]


#10

It is revision 42603 and 42604 in the trunk

root.cern.ch/viewvc?view=rev&revision=42603
root.cern.ch/viewvc?view=rev&revision=42604

Lorenzo