[Bioc-devel] IRanges findOverlaps Result Different for Recent Update
Hervé Pagès
hpages at fredhutch.org
Thu Jan 15 21:44:35 CET 2015
Hi Michael,
On 01/15/2015 11:59 AM, Michael Lawrence wrote:
> My concern is mostly in user code not seen in Bioc svn.
I understand but the fate of that code is to get out of sync
sooner or later. And sooner rather than later if it relies on
undocumented behavior.
> But perhaps the
> partial sorting (by query) is sufficient for many of those.
It seems to be sufficient for more than 99.5% of the packages in
BioC svn :-)
Note that keeping Hits objects partially sorted instead of fully
sorted not only speeds up findOverlaps() but also basic operations
on Hits objects like union(), t(), etc...
Since we are on it, I should also mention that new in BioC 3.1 is a
Hits() constructor function which takes care of partially sorting the
hits, selectHits() for selecting hits in the same way the 'select'
arg of findOverlaps() does, and all the comparison operations (==, <=,
order, sort, rank, etc..., see ?`Hits-comparison` in S4Vectors).
Cheers,
H.
>
> On Thu, Jan 15, 2015 at 11:34 AM, Hervé Pagès <hpages at fredhutch.org
> <mailto:hpages at fredhutch.org>> wrote:
>
> Hi guys,
>
> Indeed, the Hits object returned by findOverlaps() is not fully
> sorted anymore. Now it's sorted by query hit *only* and not by query
> hit *and* subject hit. Fully sorting a big Hits object has a high
> cost, both in terms of time and memory footprint. The partial
> sorting is *much* cheaper: it's done using a "tabulated sorting"
> algo implemented in C that works in linear time.
>
> The partial sorting is important: it allows a very common
> transformation like as(hits, "List") to be super fast. But the
> full sorting was overkill and generally not needed. Also note that
> the full sorting was never enforced via the validity method for
> Hits objects (and t(hits) was breaking that order in BioC < 3.1).
> Now the validity method for Hits enforces the partial sorting and
> t(hits) preserves it.
>
> There were only 3 or 4 packages that broke in devel because of
> that change (typically the change broke their unit tests). I fixed
> them (except Repitools, but it's still on my list). The fix is easy:
> if having the hits fully sorted matters, just use sort() on the Hits
> object. The man page for ?findOverlaps will soon be updated to
> reflect these changes.
>
> Cheers,
> H.
>
>
>
> On 01/15/2015 06:42 AM, Kasper Daniel Hansen wrote:
>
> Has it ever been documented that the return object is sorted in
> a specific
> way? I just want to make sure we think about whether that is
> something we
> want to enforce giving the possibility of using a different
> algorithm in
> the future.
>
> We could also address this by implementing (perhaps it already
> exists) a
> sort() method for the return object. That would still break
> existing code
> though.
>
> Best,
> Kasper
>
> On Wed, Jan 14, 2015 at 11:13 PM, Michael Lawrence <
> lawrence.michael at gene.com <mailto:lawrence.michael at gene.com>> wrote:
>
> I bet there is a lot of code that depends on having the hits
> (conveniently)
> ordered by query,subject index, so we should try to restore
> the previous
> behavior.
>
> On Wed, Jan 14, 2015 at 8:00 PM, Dario Strbenac <
> dstr7320 at uni.sydney.edu.au <mailto:dstr7320 at uni.sydney.edu.au>>
> wrote:
>
> Hello,
>
> For an identical query, the matrix results are in a
> different order.
> Consider the subject hits of the last two rows :
>
> mapping # R Under development (unstable)
> (2015-01-13 r67453) and
>
> IRanges 2.1.35
> queryHits subjectHits
> [1,] 1 1
> [2,] 1 4
> [3,] 2 2
> [4,] 4 1
> [5,] 4 4
> [6,] 6 7
> [7,] 6 6
>
> mapping # R Under development (unstable)
> (2015-01-13 r67453) and
>
> IRanges 2.0.1
> queryHits subjectHits
> [1,] 1 1
> [2,] 1 4
> [3,] 2 2
> [4,] 4 1
> [5,] 4 4
> [6,] 6 6
> [7,] 6 7
>
> This causes some values to be extracted in a different
> order by our
> annotationLookup function, and causes an error for the
> development
>
> version
>
> of Repitools on a test case which uses all.equal to
> compare a list to a
> correct list, but not for the release version which uses
> the release
> version of IRanges. Should I update the test case to
> have a new expected
> result, or is this new characteristic of findOverlaps
> likely to revert to
> the previous output soon ?
>
> The two sets of intervals to produce this result are
> anno and probesGR,
> defined in the tests.R file in the Repitools package.
>
> ------------------------------__--------
> Dario Strbenac
> PhD Student
> University of Sydney
> Camperdown NSW 2050
> Australia
> _________________________________________________
> Bioc-devel at r-project.org
> <mailto:Bioc-devel at r-project.org> mailing list
> https://stat.ethz.ch/mailman/__listinfo/bioc-devel
> <https://stat.ethz.ch/mailman/listinfo/bioc-devel>
>
>
> [[alternative HTML version deleted]]
>
> _________________________________________________
> Bioc-devel at r-project.org <mailto:Bioc-devel at r-project.org>
> mailing list
> https://stat.ethz.ch/mailman/__listinfo/bioc-devel
> <https://stat.ethz.ch/mailman/listinfo/bioc-devel>
>
>
> [[alternative HTML version deleted]]
>
> _________________________________________________
> Bioc-devel at r-project.org <mailto:Bioc-devel at r-project.org>
> mailing list
> https://stat.ethz.ch/mailman/__listinfo/bioc-devel
> <https://stat.ethz.ch/mailman/listinfo/bioc-devel>
>
>
> --
> Hervé Pagès
>
> Program in Computational Biology
> Division of Public Health Sciences
> Fred Hutchinson Cancer Research Center
> 1100 Fairview Ave. N, M1-B514
> P.O. Box 19024
> Seattle, WA 98109-1024
>
> E-mail: hpages at fredhutch.org <mailto:hpages at fredhutch.org>
> Phone: (206) 667-5791 <tel:%28206%29%20667-5791>
> Fax: (206) 667-1319 <tel:%28206%29%20667-1319>
>
>
--
Hervé Pagès
Program in Computational Biology
Division of Public Health Sciences
Fred Hutchinson Cancer Research Center
1100 Fairview Ave. N, M1-B514
P.O. Box 19024
Seattle, WA 98109-1024
E-mail: hpages at fredhutch.org
Phone: (206) 667-5791
Fax: (206) 667-1319
More information about the Bioc-devel
mailing list