Difference between example tree output and tree_at_1Mb.svg in documentation

[mayay9]

I have been following the tsinfer usage examples in the documentation (both usage.md and latest usage) using the data file P_dom_chr24_phased.vcf.gz.

I tried running all the steps in the "Data example" section. However, when I generated the tree at 1 Mb, the resulting tree does not match the example tree_at_1Mb.svg provided in the documentation.

Could you clarify whether the example SVG is deterministic, or if there might have been updates to the input data or tskit/tsinfer versions that could cause this difference? Any guidance on reproducing the exact example tree would be appreciated.

[hyanwong]

Hmm, this is very odd. It is definitely deterministic, and the docs should be compiled using the latest version. We changed to tsinfer 0.5 recently, which could give very slightly different results (see https://tskit.dev/news/20251109-tsinfer-0.5.0.html), so perhaps this is the problem.

@duncanMR : do you know what's going on here. I can only investigate later tonight.

[mayay9]

Thank you for your quick response.

I checked my environment, and it turns out I had been using tsinfer 0.4.1.
I updated to the latest version as suggested, but unfortunately the issue still persists — the generated tree is still different from the tree_at_1Mb.svg in the documentation.

[duncanMR]

That is curious indeed! I'll run through the example on my PC and see if I can reproduce the issue.

[duncanMR]

@mayay9, I can reproduce the tree plot you made exactly if I switch to version 0.4.1 as you were using, but on version 0.5.0 I get the same result that is in our documentation. As Yan mentioned, we expect there to be slight differences because we have changed part of the algorithm to improve performance.

What might have happened is that you updated tsinfer to 0.5.0 but the updated module wasn't loaded when you tried to make the plot again. Could you please try running the command below in your shell?

pip freeze | grep tsinfer

The result should be tsinfer==0.5.0.

If it still says 0.4.1 you can update it with pip install --upgrade tsinfer. If it is correct, please try reloading your Python environment (e.g. relaunching a Jupyter notebook) before making the plot.

If you still have issues, let us know!

[mayay9]

Thanks for the support.
I found the cause of the issue and it is now resolved.

Although pip freeze showed tsinfer==0.5.0, the Python environment I was actually running was still loading tsinfer 0.4.1.
After updating the correct environment so that it uses tsinfer 0.5.0, the output now matches the documentation exactly.

Thank you again for your help!