Feedback on new docs

[JonathanSchmidt1]

uncertainty supporting models are not mentioned in https://lab-cosmo.github.io/upet/latest/models.html is this intentional
UPET_UQ_SUPPORTED_MODELS = [
"pet-mad-s-v1.0.2",
"pet-mad-xs-v1.5.0",
"pet-mad-s-v1.5.0",
]
neither dos
PET-MAD DOS
PET_MAD_DOS_LATEST_STABLE_VERSION = "1.0"
PET_MAD_DOS_AVAILABLE_VERSIONS = ["1.0"]

[JonathanSchmidt1]

should this be fixed:
metatomic/torch/ase_calculator.py:1482: UserWarning: compute_requested_neighbors_from_options is deprecated and will be removed in a future version. Please use neighbor_lists_for_model to get the calculators and call them directly.
vesin.metatomic.compute_requested_neighbors_from_options(

[abmazitov]

This should be fixed on a metatomic level, but please feel free to just add this to the ignore

[Luthaf]

This will be fixed by the next release of metatomic_ase

[JonathanSchmidt1]

Should we have examples for e.g. DOS, uncertainty?

[abmazitov]

I think that in general it would be super-nice to make the cookbook examples on PET embedded into the current docs somehow, I just don't know if that is technically possible with sphinx

[Luthaf]

If nothing else, we should be able to add links to the tutorial through intersphinx

[abmazitov]

If nothing else, we should be able to add links to the tutorial through intersphinx

I think that's a brilliant idea, I'd be happy to see how it works

[sofiia-chorna]

so do we want it to be embedded or just a link with intersphinx?

[abmazitov]

Ideally - embedded, but that can be hard to achieve. Otherwise I see it as if we had neat links with preview thumbnails that lead to atomistic cookbook

[abmazitov]

uncertainty supporting models are not mentioned in https://lab-cosmo.github.io/upet/latest/models.html is this intentional UPET_UQ_SUPPORTED_MODELS = [ "pet-mad-s-v1.0.2", "pet-mad-xs-v1.5.0", "pet-mad-s-v1.5.0", ] neither dos PET-MAD DOS PET_MAD_DOS_LATEST_STABLE_VERSION = "1.0" PET_MAD_DOS_AVAILABLE_VERSIONS = ["1.0"]

That's a good catch, I didn't want to mention the availability of UQ there, but I agree that it might be nice to mention this list somewhere near the UQ block in the docs https://lab-cosmo.github.io/upet/latest/usage/ase.html#uncertainty-quantification.

[sandrasaade]

For the Usage, ASE interface, I am not sure exactly about the pre-trained sketch-map dimensionality reduction, I feel it is not clear enough. Is this shown in the code snippet below?

[sandrasaade]

Discussed with Sofiia, perhaps worth linking the example in chemiscope, otherwise IMO it may be confusing

[abmazitov]

Yeah, I agree, the current example is quite minimalistic. Please feel free to adapt it and make it more clear! I would be also very much in favor of embedding the chemiscope here!

[JonathanSchmidt1]

I noticed that usage completely replaces examples and adresses my issues. can we just remove examples?

[abmazitov]

Shall we remove the current examples entirely and replace them with embedded parts of the atomistic cookbook, as @Luthaf has suggested above?

[Luthaf]

That would be ideal IMO, this way we don't have to maintain multiple copies of the same thing

[sandrasaade]

Another question regarding the Batched evaluation with metatrain, Step 2: It mentions the dataset in extxyz format, however in the code snippet the dataset is passed as an .xyz file?

[abmazitov]

Yes, this is normal, ASE treats xyz as extxyz. We mention extxyz just because the plain xyz doens't allow to store anything but positions

[Luthaf]

The docs seems to be missing API documentation, ideally generated with autodoc to have all functions/submodules/classes visible!

[sofiia-chorna]

fair point, happy to fix it

[abmazitov]

Yes, please! Good catch!

[sofiia-chorna]

fixed in #140

[sandrasaade]

For the Frequently Asked Questions: When mentioning XS, XL or S model, it would be worth stating somewhere the list of hyperparameters associated with each model size, and usually what kind of applications they are suitable for. Otherwise, I have not heard about these labels before

[abmazitov]

Maybe we can put the link to Filippo's paper and say that the full list of hypers available here https://arxiv.org/abs/2601.16195 - it's in the Section 5 of the SI.

[sandrasaade]

Yep exactly, that would work as well!

[abmazitov]

Alternatively we can just put this table in the docs, maybe that will be even better

[JonathanSchmidt1]

Pr is open #139

[abmazitov]

Pr is open #139

Approved

[GardevoirX]

This page is not linked in the new doc, it's intended?
https://github.com/lab-cosmo/upet/blob/df94f29600d6bdcde75ccd3b4a09006f66d61100/docs/SPEED.md

[sofiia-chorna]

i think it would be nice to have a speed page in the docs

[abmazitov]

Yes, very good idea! In general we should probably add some scripts to replicate this plot on a local hardware, but for now let's add it to the docs

[GardevoirX]

Here #141