Skip to content

Fix genetic_relatedness to allow single sample set with self-comparisons #3235

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 1 commit into from
Sep 16, 2025
Merged

Fix genetic_relatedness to allow single sample set with self-comparisons #3235

merged 1 commit into from
Sep 16, 2025

Conversation

@andrewkern
Copy link
Member

@andrewkern andrewkern commented Jun 26, 2025 •
edited
Loading

Description

This PR fixes issue #3055 where genetic_relatedness would fail when computing self-comparisons with a single sample set. As @petrelharp pointed out there is no reason this shouldn't be allowed for certain statistics, including genetic_relatedness

The Problem

Previously, calling genetic_relatedness with a single sample set and indexes referring to that set would raise an error:

ts.genetic_relatedness([[0]], indexes=[(0,0)])
# TSK_ERR_INSUFFICIENT_SAMPLE_SETS: Insufficient sample sets provided

This occurred because the underlying C API validates that at least k=2 sample sets are provided for k-way statistics, even when the indexes only reference a subset of those sets.

This PR

The fix adds an allow_self_comparisons parameter to the internal __k_way_sample_set_stat method, which is only set to True for genetic_relatedness. When enabled:

  1. The method validates that all indexes reference valid sample sets
  2. If fewer than k sample sets are provided, it pads the list with dummy sample sets to satisfy the C API requirement
  3. The dummy sets are never accessed during computation since only the sets referenced by the indexes are used

This ensures that:

  • Only genetic_relatedness behavior is changed
  • Other statistics (like Fst) continue to enforce the minimum sample set requirement
  • The C API contract is satisfied without modifying the C code

Further, this PR sets the stage for allowing for other appropriate k-way stats to be computed with self comparisons by setting a flag.

Testing

Added comprehensive tests for all three computation modes (site, branch, node) to verify:

  • Single sample set with self-comparison works correctly
  • Multiple samples within a single set work correctly
  • The fix doesn't affect other statistics

All existing tests continue to pass, but @petrelharp this would benefit from you specifically looking over what I've done to be sure you're okay with the way I've implemented allow_self_comparisons

Fixes #3055

PR Checklist:

  • Tests that fully cover new/changed functionality.
  • Documentation including tutorial content if appropriate.
  • [] Changelogs, if there are API changes.

I'm not sure if I should touch the changelogs here?

@andrewkern andrewkern mentioned this pull request Jun 26, 2025
Closed
@codecov
Copy link

codecov bot commented Jun 26, 2025 •
edited
Loading

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 89.75%. Comparing base (1d8f453) to head (c173b32).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files 
@@           Coverage Diff           @@
##             main    #3235   +/-   ##
=======================================
  Coverage   89.75%   89.75%           
=======================================
  Files          29       29           
  Lines       32545    32545           
  Branches     5962     5962           
=======================================
  Hits        29211    29211           
  Misses       1890     1890           
  Partials     1444     1444
Flag Coverage Δ
c-tests 86.76% <100.00%> (ø)
lwt-tests 80.38% <ø> (ø)
python-c-tests 88.24% <ø> (ø)
python-tests 98.79% <ø> (ø)
python-tests-no-jit 33.02% <ø> (ø)
python-tests-numpy1 50.71% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines Coverage Δ
c/tskit/trees.c 90.81% <100.00%> (ø)
🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

@andrewkern andrewkern force-pushed the fix-issue-3055-genetic-relatedness branch from 89397d5 to cd830f7 Compare June 26, 2025 22:12
@jeromekelleher
Copy link
Member

I think we'll need to wait for @petrelharp to get informed review on this one.

@jeromekelleher
Copy link
Member

Bumping this one for you @petrelharp

@petrelharp
Copy link
Contributor

Ignore my previous comment (now deleted); I was being dense.

@petrelharp
Copy link
Contributor

Let's see. First, it's quite easy to modify the checking code that produces the error. I went ahead and did that (see this commit: 57ad49f).

However, one can compute 'self' comparisons for other statistics, for instance, this works:

ts.divergence([[0], [1]], indexes=[(0, 0)])

or this

ts.f4([[0],[1],[2],[3]], indexes=[(0,0,0,0)])

These are perfectly valid things to compute, just the thought was they are probably not telling someone what they actually want to know. So, if the intention behind the restriction for "at least k sample sets for a k-way stat" was to keep people from computing things they probably didn't want to compute, then we're not actually doing that. We could require indexes be unique, but I don't think we should actually do that - for instance, f4(..., indexes=[(0,0,1,1)]) is actually a version of f2( ) that people compute sometimes (I think, or maybe it's (0, 1, 0, 1)). So, in the next commit I've just removed the requirement that there be at least k sample sets - instead, requiring there be at least 1 sample set in all cases.

How's that sound, @andrewkern?

@petrelharp petrelharp force-pushed the fix-issue-3055-genetic-relatedness branch from 4b8755f to 7f48542 Compare September 12, 2025 00:58
@petrelharp
Copy link
Contributor

LMK how this looks @andrewkern; if you like it I'll squash these commits down.

@petrelharp petrelharp force-pushed the fix-issue-3055-genetic-relatedness branch from 931ce7e to 34d6826 Compare September 12, 2025 19:11
@petrelharp
Copy link
Contributor

Rebased; added to CHANGELOG; tests are (should be) passing. Ready to merge when you approve, @andrewkern .

andrewkern
andrewkern commented Sep 15, 2025
View reviewed changes
andrewkern
andrewkern commented Sep 15, 2025
View reviewed changes
c/tskit/trees.c
int ret = 0;

if (num_sample_sets < tuple_size) {
if (num_sample_sets < 1) {
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

maybe add a comment here to explain the check?

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It seems pretty self-explanatory to me? Like, how can you compute a statistic with only one sample set? (The comment would be something like // check there is at least one sample set, but we like to avoid this sort of comment that just says the same thing as the code?)

@andrewkern
Copy link
Member Author

looks good to me @petrelharp -- a couple nit picks about adding comments to help explain the code to future us

@petrelharp
Copy link
Contributor

See my comments & changes. Could you "approve" the code in review if you're good with it now? (Or let me know if not?)

benjeffery
benjeffery approved these changes Sep 16, 2025
View reviewed changes
Copy link
Member

@benjeffery benjeffery left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@andrewkern andrewkern self-assigned this Sep 16, 2025
@andrewkern @petrelharp
Fix: genetic_relatedness to allow single sample set with self-compari…
c173b32 
…sons

Previously, genetic_relatedness would fail with TSK_ERR_INSUFFICIENT_SAMPLE_SETS
when given a single sample set and indexes referring to that set, e.g.:
  ts.genetic_relatedness([[0]], indexes=[(0,0)])

This was because the C API requires at least k=2 sample sets for k-way statistics,
even when indexes only reference a single set for self-comparison.

The fix is more general: it removes the restriction that there be at
least k sample sets for k-way stats, because this wasn't a useful or
consistently applied restriction.

Closes tskit-dev#3055
@petrelharp petrelharp force-pushed the fix-issue-3055-genetic-relatedness branch from c98e387 to c173b32 Compare September 16, 2025 15:57
@petrelharp petrelharp added this pull request to the merge queue Sep 16, 2025
Merged via the queue into tskit-dev:main with commit 6f6aa73 Sep 16, 2025
18 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@benjeffery benjeffery benjeffery approved these changes

@petrelharp petrelharp Awaiting requested review from petrelharp

Assignees

@andrewkern andrewkern

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

genetic_relatedness thinks it needs 2 samples sets but is valid with only one

4 participants

@andrewkern @jeromekelleher @petrelharp @benjeffery