Update matplotlib version

[Radonirinaunimi]

A simple change to address #2162 and the issue reported in #1809 related to the axis transformation for matplotlib >= 3.8. This also seems to fix the non-deterministic failing tests in #2099 (at least for me locally).

PS: Need to be tested further.

[scarlehoff]

Nice! I would be happy with merging it and crossing our fingers since it seems to work also in the ci

[Radonirinaunimi]

Nice! I would be happy with merging it and crossing our fingers since it seems to work also in the ci

I still need to perform further tests but indeed it seems to work. After understanding the issues, I think it happened that magically it worked before v3.8.

[github-actions]

Greetings from your nice fit 🤖 !
I have good news for you, I just finished my tasks:

• Fit Name: NNBOT-dcb268416-2024-10-30
• Fit Report wrt master: https://vp.nnpdf.science/upbwXyHTSNK-7FjKEeb2zw==
• Fit Report wrt latest stable reference: https://vp.nnpdf.science/odhmgItbQx65ag2Vj1g7eQ==
• Fit Data: https://data.nnpdf.science/fits/NNBOT-dcb268416-2024-10-30.tar.gz

Check the report carefully, and please buy me a ☕ , or better, a GPU 😉!

[scarlehoff]

I'd be happy merging this even if it doesn't fix the random failure. If it does, so much the better.

What @RoyStegeman said about the DW variants worries me because those are relatively recent, if it fails again I'll have a deeper look.

Thanks!

[github-actions]

Greetings from your nice fit 🤖 !
I have good news for you, I just finished my tasks:

• Fit Name: NNBOT-028dc2124-2024-10-30
• Fit Report wrt master: https://vp.nnpdf.science/jbCBEMzMQAS9btI69QF3nQ==
• Fit Report wrt latest stable reference: https://vp.nnpdf.science/Ocg3Nxa2QNSGFpqCipAhmQ==
• Fit Data: https://data.nnpdf.science/fits/NNBOT-028dc2124-2024-10-30.tar.gz

Check the report carefully, and please buy me a ☕ , or better, a GPU 😉!

[Radonirinaunimi]

I had to re-run the fitbot because the first one was started before the change in the conda-recipe/meta.yaml. So the last reports are the correct ones.

So , yes, this basically solves the issues with the offset plotting with a still minor issue (which should be easy to solve) that for some very few ratio plots there are points that are cut ways: for e.g. reference vs now.

[scarlehoff]

i see, do you have an idea of how to solve that? Re-axing (if that's a thing?)

[Radonirinaunimi]

i see, do you have an idea of how to solve that? Re-axing (if that's a thing?)

Not yet xd but I will have a look in a while. It shouldn't be difficult.

[RoyStegeman]

Something I didn't realise before: I can reproduce this error locally (while yesterday I said I couldn't), but while it occurs if I run all tests it doesn't if I try to run only the failing tests

Are tests always executed in the same order? Perhaps there is a race condition for an @lru_cache or something like that?

[Radonirinaunimi]

So I haven't found a fancy way to construct the ranges so the $y$ limits are computed manually f7c6595.

Something I didn't realise before: I can reproduce this error locally (while yesterday I said I couldn't), but while it occurs if I run all tests it doesn't if I try to run only the failing tests

Just to confirm that locally, sometimes, I am still experiencing this issue. So updating dependency versions definitely does not magically solve it.

[RoyStegeman]

Do you experience the issue also if you only run the failing tests?

[Radonirinaunimi]

Do you experience the issue also if you only run the failing tests?

I've never experienced it when running separately the failing tests. But also it's not every time that tests are failing when I run everything all at once. It seems to occur randomly when running all the tests altogether...

[RoyStegeman]

Yes same here, I tried two more times running all locally but they just passed

[Radonirinaunimi]

Yes same here, I tried two more times running all locally but they just passed

Hmhmh, this issue deserves a closer look. If we are happy with the last fitbot, we can merge this and investigate the issue in another PR.

[RoyStegeman]

I think the problem might be the caching of parse_new_metadata. I'll have a closer look tomorrow

[scarlehoff]

The latest fitbot actually picked up a problem!

RE the failure, I agree, let's investigate it somewhere else.

[Radonirinaunimi]

The latest fitbot actually picked up a problem!

Yes, the error is that for some instances all the values are all NaNs, so it will be sufficient to simply skip these instances when setting ax.get_ylim. I missed this before because I haven't checked all the datasets that are run/fitted by the bot.

[github-actions]

Greetings from your nice fit 🤖 !
I have good news for you, I just finished my tasks:

• Fit Name: NNBOT-66a0bafcf-2024-11-01
• Fit Report wrt master: https://vp.nnpdf.science/ju590ltLStii5BG9mkW-Hw==
• Fit Report wrt latest stable reference: https://vp.nnpdf.science/h-oztl8KQxyUhtxankT7DQ==
• Fit Data: https://data.nnpdf.science/fits/NNBOT-66a0bafcf-2024-11-01.tar.gz

Check the report carefully, and please buy me a ☕ , or better, a GPU 😉!

[scarlehoff]

There was still a problem with plots with log axes and plots which might have several axis in one figure, this example suffers from both: https://vp.nnpdf.science/ju590ltLStii5BG9mkW-Hw==/matched_datasets_from_dataspecs24_dataset_report_report.html#normalized

I've pushed my solution (+ rebase on top of master). Basically I use the existing ymax/ymin instead of recalculating it, seems to work locally... let's see whether the bot thinks the same thing and we are not just playing whack-a-mole where fixing one plot breaks the following one.

[github-actions]

Greetings from your nice fit 🤖 !
I have good news for you, I just finished my tasks:

• Fit Name: NNBOT-045604d48-2024-11-22
• Fit Report wrt master: https://vp.nnpdf.science/y4iDx_rAQCWMJZowjxvp9g==
• Fit Report wrt latest stable reference: https://vp.nnpdf.science/3L52mzmZR4OggFDOi2J3eg==
• Fit Data: https://data.nnpdf.science/fits/NNBOT-045604d48-2024-11-22.tar.gz

Check the report carefully, and please buy me a ☕ , or better, a GPU 😉!

[Radonirinaunimi]

I've pushed my solution (+ rebase on top of master). Basically I use the existing ymax/ymin instead of recalculating it, seems to work locally... let's see whether the bot thinks the same thing and we are not just playing whack-a-mole where fixing one plot breaks the following one.

This is what I first tried before but didn't work locally because (back then if I remember correctly the issue was) the values of the mins/maxs were computed based on the data cv+/-err and therefore might cut away the predictions. But looking at the reports, everything seems to be perfect now. So, good for it.

The only thing remaining here is updating the regression plot.

[scarlehoff]

I'll try to update the plot later. The change is basically a shift, maybe there's a way of making the test more stable.

[scarlehoff]

Since the python installation is already checking the plot, I put a less restrictive check for the conda installation (I'll increase it until the tests pass).

[scarlehoff]

LGTM! If you are happy with my change to the tests please merge ^^

I've checked all plots in the bot's report and I think they are ok, but at some point the points were dancing around my eyes... I hope I haven't missed anything.

[Radonirinaunimi]

I've checked all plots in the bot's report and I think they are ok, but at some point the points were dancing around my eyes... I hope I haven't missed anything.

I've also gone through all of them before, so let's hope that the parts where the points started dancing in our case did not overlap xD

I'm merging now.