[hist] Fix TH1::KolmogorovTest with option X when one histogram has zero errors #13822
Conversation
|
Starting build on |
Test Results 9 files 9 suites 1d 20h 32m 46s ⏱️ For more details on these failures, see this check. Results for commit 66d3a77. ♻️ This comment has been updated with latest results. |
|
Build failed on windows10/default. Failing tests: |
| h2Expt->Reset(); | ||
| h1Expt->FillRandom(h1_cpy, (Int_t)esum1); | ||
| h2Expt->FillRandom(h1_cpy, (Int_t)esum2); | ||
| if (!afunc1) { |
There was a problem hiding this comment.
Can we introduce a comment here, explaining what happens in the loop over over toys and why we don't need to fill the histograms if afunc1 or afunc2 respectively? And Maybe it's also good to note that they can't be both false, which is checked at the beginning of the function. Otherwise one might wonder why the loop is even implemented also for this case.
hist/hist/src/TH1.cxx
Outdated
| if (afunc1) Copy(*h1Expt); | ||
| if (afunc2) h2->Copy(*h2Expt); |
There was a problem hiding this comment.
| if (afunc1) Copy(*h1Expt); | |
| if (afunc2) h2->Copy(*h2Expt); |
Is this even necessary? The h1Exp is already created as a clone of this, and the h2Exp could also be created as a clone of h2.
By the way, when you create the clones, I would use a lambda to reduce code duplication and also use std::unique_ptr, replacing the block a few lines before with:
auto cloneHist = [&](TH1 const *hist) {
return std::unique_ptr<TH1>{static_cast<TH1 *>((gDirectory ? gDirectory : gROOT)->CloneObject(hist, false))};
};
std::unique_ptr<TH1> h1_cpy = cloneHist(this);
std::unique_ptr<TH1> h1Expt = cloneHist(this);
std::unique_ptr<TH1> h2Expt = cloneHist(this /*or maybe h2 here?*/);There was a problem hiding this comment.
Thanks for the comment. I will simplify the code using just copy and avoid un-neeeded clones of histograms
hist/hist/src/TH1.cxx
Outdated
| } | ||
| if (!afunc2) { | ||
| h2Expt->Reset(); | ||
| h2Expt->FillRandom(h1_cpy, (Int_t)esum2); |
There was a problem hiding this comment.
Why is h2Expt a randomized version of h1_cpy here? Should it not be drawn from h2?
There was a problem hiding this comment.
No, here we need to find the distribution of the KS test statistics, dsEXPT for the NULL, i.e. when we draw two histograms from the same parent distribution.
However, you are right there is an error. If afunc1=TRUE we need to draw from h2. So the logic is not correct.
…ero errors Fix the case of using KS test when one of the histogram has zero errors (i.e is a function). Improve KS test adding possibility to specify number of toys by using option "X=number" for example "X=1000" This fixes the issue root-project#13697
Use as parent distribution for generating the toys for the NULL distribution of the test statistics the histogram with higer statistics. If it is a function use that one as parent. This commit takes into account the comment of Jonas in the PR
lmoneta
force-pushed
the
fix_th1_kstest_x
branch
from
fc7f6dc
to
66d3a77
Compare
|
Starting build on |
Fix the case of using KS test when one of the histogram has zero errors (i.e is a function). Improve KS test adding possibility to specify number of toys by using option "X=number" for example "X=1000"
This PR fixes the issue #13697