Interaction Datasets

[mcneela]

The DES370K interaction dataset has been created. The data dictionary elements additionally have keys "mol0" and "mol1" to make getting the individual dimers easier for users.

P.S. The current DES dataset we are using pulls the geometries from .mol files which were not created by the original dataset authors and appear to be correct. The correct atomic coordinates can be obtained from the original csv file as I am doing.

[mcneela]

@FNTwin

[prtos]

I think we only need this. a lot of the information can in the sub dict can be retrieved only with the info below!

        item = dict(
            energies=energies,
            subset=np.array(["DES370K"]), #In dess they have subsets for each monomer no? so mabe the subset here can be "subset1.subset2"
            n_atoms=np.array([natoms0 + natoms1], dtype=np.int32),
            n_atoms_first=np.array([natoms0], dtype=np.int32),
            atomic_inputs=atomic_inputs, # with n_atoms_first we can resplit this so we can leave this and split in the getitem
            name=name, # already smiles1 and smiles2 can be 
        )

[FNTwin]

Remove yaml dependency

[mcneela]

Thank you Danny for all the work!

I would remove the commented parts of the code and add some docstrings for the different classes/new datasets like we have in the other datasets.

Also lots of read_raw_entries are pretty long and handling multiple functions internally so I would refactor them a little bit into multiple function to make them more readable.

Last thing, is some dataset can probably be written with a little bit more inheritance to make everything cleaner and have less clutter in the interaction dataset folder , as an example:

class des370k(...):
    ....

class des5m(des370k):
    ....

If you can add a dummy class as a test for the new object return type

Comments have been deleted and docstrings have been added! 🧨

[mcneela]

Thank you Danny for all the work!
I would remove the commented parts of the code and add some docstrings for the different classes/new datasets like we have in the other datasets.
Also lots of read_raw_entries are pretty long and handling multiple functions internally so I would refactor them a little bit into multiple function to make them more readable.
Last thing, is some dataset can probably be written with a little bit more inheritance to make everything cleaner and have less clutter in the interaction dataset folder , as an example:

class des370k(...):
    ....

class des5m(des370k):
    ....

If you can add a dummy class as a test for the new object return type

Comments have been deleted and docstrings have been added! 🧨

Refactoring of DES370K/5M is now done as well

[mcneela]

Remove yaml dependency

Fixed now

[FNTwin]

Just do double check, is every datasets in Hartree and Angstrom units?

[FNTwin]

I think this should be uncommented.
Also I don't see the X40 and Splinter downloads in the config_factory, is there a particular reason why we didn't add them there?

[mcneela]

I downloaded those datasets manually. There were multiple files and such so it seemed difficult to use the config_factory.

[FNTwin]

For now this is fine but we need to add more informations about these datasets in the readme

[FNTwin]

Remove the print

[mcneela]

Just do double check, is every datasets in Hartree and Angstrom units?

I'm not positive, I suggest we merge and then do a new PR if there are any issues.

[FNTwin]

Just do double check, is every datasets in Hartree and Angstrom units?

I'm not positive, I suggest we merge and then do a new PR if there are any issues.

I'm ok merging (as soon as you remove the print statement) but I'll leave it to you correcting the units in case they are wrong