Repository navigation
Clamp fracDefining to 1.0 in perform_bootstrap to avoid numpy binomial "p > 1" crash - #321
Merged
mariaelf97 merged 1 commit intoSep 4, 2026
Conversation
…l "p > 1" crash fracDefining = fracDepths.sum() is mathematically 1.0 but can float-round one ULP above 1.0, which numpy.random.Generator.binomial rejects as p>1. Clamp to 1.0. Fixes andersen-lab#320
Collaborator
|
Hi @pditommaso, thanks for bringing this up and for providing a quick solve! We hadn't observed this previously, but it makes sense that it'd crash like this if p>1. Your change should solve this issue nicely. We'll get this incorporated into main shortly! Best, |
Collaborator
|
Thank you @pditommaso for this quick fix! |
Collaborator
|
Sure! In transit atm, so if you have a moment free to de-lint that'd be awesome! thanks @mariaelf97 |
Contributor
Author
|
Thanks both! much appreciated! |
This was referenced Sep 15, 2026
pditommaso
added a commit
to pditommaso/modules
that referenced
this pull request
Sep 17, 2026
Bumps freyja from 2.0.3 to 2.0.4 across boot, demix, variants and update (container + conda pins). Freyja 2.0.4 fixes a floating-point rounding bug in perform_bootstrap (andersen-lab/Freyja#321): fracDefining could round one ULP above 1.0 and make numpy's binomial raise "p < 0, p > 1 or p is NaN", intermittently crashing FREYJA_BOOT (nf-core/viralrecon#618). Verified present in the container: sample_deconv.py now clamps `fracDefining = min(fracDepths.sum(), 1.0)`. Snapshots are intentionally unchanged: freyja 2.0.4 still reports "version 2.0.3" from `freyja --version` (upstream did not bump the version string), so the captured version and `freyja variants` output are identical to 2.0.3. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Hi, and thanks for maintaining Freyja! 🙏
Context. We run nf-core/viralrecon on a nightly basis across a few cloud backends, and noticed
freyja bootoccasionally failing on a small fraction of samples (~0.5–4% in our runs). Tracing it led to a floating-point edge case inperform_bootstrap; this PR proposes a small fix. More detail in #320.What happens.
fracDefining = fracDepths.sum()is mathematically the identitytotalDepth / totalDepth = 1.0, but floating-point accumulation can land one ULP above 1.0 (e.g.1.0000000000000002). Because that value is passed as thepargument torng.binomial(totalDepth, fracDefining, numBootstraps), numpy rejectsp > 1:The change. Clamp
fracDefiningto its mathematical maximum of 1.0 before it's used as a probability:When the value is already ≤ 1.0 nothing changes; when it rounds above 1.0,
binomialnow receivesp = 1.0and returnstotalDepth(the intended result).fracDepths_adj = fracDepths / fracDefiningis unaffected (division by 1.0).Verification.
rng.binomial(1684431, 1.0000000000000002, 10)raises theValueErrorbefore the change and returns[1684431, …]after; a real SARS-CoV-2 sample that triggered the crash runsfreyja bootto completion with the patch applied.I didn't include a unit test since the trigger is a float-rounding artifact of specific (depth vector × barcode set) inputs that's tricky to reproduce portably — but I'd be glad to add one if you can suggest a stable way to force it, and of course happy to adjust the approach however you'd prefer.
Fixes #320.
Note: this PR was prepared with the help of an AI agent (Claude Code), and reviewed before submission.