Added check for non-finite copy ratios in ModelSegments pipeline.

[samuelklee]

@LeeTL1220 This should fail earlier in situations like the one you ran into with the low coverage test data. I don't think there should be any unintended side effects. I'm fine if this doesn't make it in before the point release if you want to run some sanity checks on real data with it (since users should be able to figure out what is going on relatively quickly if they run into the issue), but I'll leave it up to you.

[codecov-io]

Codecov Report

Merging #4292 into master will increase coverage by 0.188%.
The diff coverage is 100%.

@@              Coverage Diff               @@
##              master    #4292       +/-   ##
==============================================
+ Coverage     79.082%   79.27%   +0.188%     
- Complexity     16655    17036      +381     
==============================================
  Files           1050     1050               
  Lines          59681    60796     +1115     
  Branches        9754     9946      +192     
==============================================
+ Hits           47197    48193      +996     
- Misses          8696     8787       +91     
- Partials        3788     3816       +28

Impacted Files	Coverage Δ	Complexity Δ
.../tools/copynumber/denoising/SVDDenoisingUtils.java	79.221% <100%> (+0.273%)	46 <2> (+2)	⬆️
...er/tools/copynumber/formats/records/CopyRatio.java	76% <100%> (+2.087%)	9 <0> (ø)	⬇️
...broadinstitute/hellbender/utils/read/GATKRead.java	50% <0%> (-18.75%)	20% <0%> (+7%)
...er/tools/spark/sv/evidence/EvidenceTargetLink.java	70.513% <0%> (-6.05%)	18% <0%> (+3%)
...ender/engine/datasources/ReferenceMultiSource.java	76.471% <0%> (-0.452%)	15% <0%> (+4%)
...lbender/utils/read/SAMRecordToGATKReadAdapter.java	92.027% <0%> (-0.433%)	270% <0%> (+134%)
...ute/hellbender/utils/read/ArtificialReadUtils.java	93.909% <0%> (-0.186%)	122% <0%> (+52%)
...itute/hellbender/engine/spark/GATKRegistrator.java	100% <0%> (ø)	4% <0%> (+2%)	⬆️
...utils/test/ReadsPreprocessingPipelineTestData.java	1.25% <0%> (+0.403%)	1% <0%> (ø)	⬇️
...tructuralVariationDiscoveryArgumentCollection.java	97.143% <0%> (+0.476%)	0% <0%> (ø)	⬇️
... and 15 more

[LeeTL1220]

Can you just add a bit more for a user to diagnose?

[LeeTL1220]

@samuelklee I'm fine without this one in next release.

[samuelklee]

@LeeTL1220 OK, tweaked the message a bit. I think I'm OK with this going in for the next point release. This is the sort of thing for which it will be nice to have the automatic validations, as a sanity check.

[LeeTL1220]

@samuelklee Is it possible to expand the error message a bit so that users get something actionable. For example, "Does the interval list match what was covered in the bam file? Does this bam have coverage in all intervals?"

[samuelklee]

I'd rather keep the message more generic, and think of the check as simply defining what a valid CopyRatio object can be: an interval associated with a finite double value. One might imagine that someone would try to create such an object that does not originate from a BAM (perhaps for test data, or for imputing missing values in pre-existing data, etc.). This check says that they must create it with some finite value.

A more appropriate place for the sort of message you suggest is in the relevant denoising method. In the edge case you encountered, you used a BAM that was almost completely uncovered in all bins at the specified resolution, resulting in a sample median of zero. Since one of the steps in standardization is dividing by the sample median, this results in a divide by zero. I've added the corresponding check.

[samuelklee]

@LeeTL1220 Is this good to go?

[LeeTL1220]

Go for it.

…

On Feb 21, 2018 16:51, "samuelklee" ***@***.***> wrote: @LeeTL1220 <https://github.com/leetl1220> Is this good to go? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#4292 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ACDXk1jxHtfgMbCXExlN3x5i6jPZYECkks5tXI_AgaJpZM4RxGuu> .