-
Notifications
You must be signed in to change notification settings - Fork 9
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.
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
test_somoclu_summarizers.py failing #300
Comments
Yes, I think that this is yet another instance of package developers updating conda and not PyPI, if you do a |
for some reason conda can't find the somoclu package, even though it's listed on conda-forge 🤔 weird, could be a problem with my local setup. I added a comment to the related issue and I'll probably ask them to put 1.7.6 on pypi. |
I think the issue here is primarily that some of the underlying packages aren't up to date for M1 mac. |
the somoclu people updated the package on pip and conda, so I have 1.7.6 working on my M1! but the somoclu tests are still failing 🤔
|
Yeah, this is a generic issue with ML algorithms. When they change the code version, the values you get change slightly.
Maybe check with ziang that the new value is reasonable and then loosen the np.isclose() requirement.
…-e
On Mar 6, 2023, at 4:19 PM, Max West ***@***.***> wrote:
the somoclu people updated the package on pip and conda, so I have 1.7.6 working on my M1! but the somoclu tests are still failing 🤔
# test loading model by name rather than via handle
summarizer2 = summarizer_class.make_stage(
name=key,
model=f"tmpsomoclu_"+key+".pkl",
aliases={
"model": "somoclu_summarize_test2_model",
"input": "somoclu_summarize_test2_input",
"spec_input": "somoclu_summarize_test2_spec_input",
},
)
_ = summarizer2.summarize(phot_data, spec_data)
fid_ens = qp.read(
summarizer2.get_output(
summarizer2.get_aliased_tag("single_NZ"), final_name=True
)
)
meanz = fid_ens.mean().flatten()
> assert np.isclose(meanz[0], 0.14414913252122552)
E assert False
E + where False = <function isclose at 0x104591550>(0.13940019753973246, 0.14414913252122552)
E + where <function isclose at 0x104591550> = np.isclose
tests/test_somoclu_summarizers.py:62: AssertionError
—
Reply to this email directly, view it on GitHub <https://github.com/LSSTDESC/RAIL/issues/300#issuecomment-1457258826>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADRIGIVN4JSUFLI5PIY76STW2Z5HVANCNFSM6AAAAAAU2Z7S5I>.
You are receiving this because you commented.
|
I think this same issue popped up with pzflow, and if you look at the pzflow test in tests/test_algos.py you can see that we have a np.isclose with an atol=0.05 because there is a minor numerical instability that led to some small changes in the results when some of the normalizing flow dependencies were being updated to newer versions. My guess is that we should just do the same thing here. |
overtaken by events a while back |
Running an a 2021 macbook M1 (so apple silicon, which might be the cause of my woes...)
when trying to build and run the RAIL package locally, I get the following errors on the somoclu test:
there seems to be a related issue tracked in the somoclu repo noting this on windows machines from 4 years ago. It seems like they pushed a
1.7.5.1
update to fix it but the pip install doesn't actually work locally, due to a metadata issue:They also have a 1.7.6 release out from ~2 years ago but it's not on pypi 🙁
The text was updated successfully, but these errors were encountered: