Test exact identity of reco references, use converter-less muon reco, don’t require PV filter

• The MR changes the hlt1 reco baseline test to test that the result matches a reference.
• Adds MuonID algorithm without converters (needs LHCb!2111 (merged) and Rec!1689 (merged))
• Adds test for deprecated path with converters.
• Adds test for tracking efficiencies in Hlt1 reconstruction.
• Adds test for tracking efficiencies in Hlt2 reconstruction.
• Remove requirement of having at least one PV from standalone reconstruction.

Closes #79 (closed).

The refs/hlt1_reco_baseline.ref file is missing

but why, what do you really want to assert with this test? you know I'm allergic to refs

I want to test that nothing changes in the reconstruction and if there is a change that must be acknowledged by a change in the references.

added 1 commit

• c2c35fe1 - Add muon reconstruction path without converters

Compare with previous version

If you don't mind, I make the two changes, changing the test and changing muon ID in one MR. Otherwise the juggling of the two becomes a bit cumbersome.

changed title from WIP: Test exact identity of references of hlt1 reco baseline test to WIP: Test exact identity of references of hlt1 reco baseline test and use MuonID algorithm without converters in reco baseline

changed the description

added 1 deleted label

mentioned in merge request LHCb!2111 (merged)

• [2019-10-03 00:14] Validation started with lhcb-lcg-dev4#1050
• [2019-10-03 13:34] Validation started with lhcb-lcg-dev4#1051
• [2019-10-04 00:10] Validation started with lhcb-lcg-dev4#1052
• [2019-10-05 00:11] Validation started with lhcb-lcg-dev4#1053
• [2019-10-06 00:10] Validation started with lhcb-lcg-dev4#1054
• [2019-10-07 00:10] Validation started with lhcb-lcg-dev4#1055

added 1 commit

• 6a330c73 - fix tests

Compare with previous version

added 1 commit

• c712037c - Fixed formatting

Compare with previous version

added 1 commit

• 08738765 - Add PrChecker efficiency tests for Hlt1

Compare with previous version

changed the description

added 3 commits

• 0a083f18 - Fixed formatting
• 0614c8f5 - Adapt reference
• 1580efe4 - Remove PV filter from reco sequence, this should be decided line by line

Compare with previous version

added 1 commit

• 9457e107 - Some cleanup and make data flow more explicit

Compare with previous version

added 1 commit

• df63f65f - Add Hlt2 tracking efficiency checks

Compare with previous version

assigned to @apearce

changed the description

mentioned in issue #86 (closed)

added 1 commit

• 94368449 - Update reference files, add tests for Brunel-like reconstruction

Compare with previous version

added 1 commit

• 583a0e23 - Fixed formatting

Compare with previous version

@apearce the tests are passing for me locally. I stop now adding new stuff as the MR is becoming big.

added hlt1 hlt2 ~1664 labels and removed 1 deleted label

added reconstruction [DEPRECATED] + 1 deleted label and removed hlt1 hlt2 ~1664 labels

added hlt1 hlt2 ~1664 labels and removed 1 deleted label

added 1 deleted label

To have different references depending on the platform you need to suffix the filename with the platform, e.g. hlt1_reco_baseline.ref.x86_64+avx2+fma-centos7-gcc8-opt. You can have a single reference match multiple platforms by omitting some of the tokens split by -. For instance hlt1_reco_baseline.ref.x86_64 would match all but the avx2 platform and hlt1_reco_baseline.ref.x86_64+avx2+fma would match only the avx2 platform. The general logic is that the platform where you run is split into tokens with - as a separator and the same is done with the last part of the ref filename. The filename where the most tokens are a subset of the platform's tokens is the reference used for the validation, see here.

Several tests fail for me locally.

The following tests FAILED:
	 13 - RecoConf.hlt1_reco_baseline (Failed)
	 14 - RecoConf.hlt1_reco_baseline_deprecated (Failed)
	 15 - RecoConf.hlt1_reco_baseline_with_mcchecking (Failed)
	 18 - RecoConf.hlt2_reco_baseline_with_mcchecking (Failed)
	 20 - RecoConf.hlt2_reco_brunelesque_with_mcchecking (Failed)

After some discussion, this is because @sstahl made reference files with an avx2 build, and I ran without avx2. It's expected that there are some differences between these two on the level of numerical precision, but the VELO reconstruction has other 'known' differences that cascade down and cause larger-than-numerical differences later in the reconstruction.

These types of tests are important so we want them in Moore ASAP. We haven't been able to think of any nice solutions that will get all nightly slots green; the best we have is to add separate references for each platform (hopefully only two for now: with and without avx2). It's possible to specify platform-specific reference files by appending components of the targeted platform to the end of the filename, e.g. some_reference.ref.x86_64+avx2+fma will target avx2 and some_reference.ref.dbg will target all dbg platforms.

I've added this MR to the lhcb-tdr-test nightly, and tomorrow @sstahl can download the non-avx2 references and add them here.

• [2019-10-08 00:12] Validation started with lhcb-tdr-test#697