Accessing properties in `.db` files, tutorials 1, 2.

[CalmScout]

Dear maintainers of SchNetPack,

I am following the tutorials from your repo and adapting it to my data. From tutorial 1 we have a custom_data object of the class schnetpack.data.datamodule.AtomsDataModule that was generated for a set of SMILES and corresponding bioactivities labels like:

for smi, bio in zip(smiles, bioactivities):
    try:
        atoms_list.append(smi2ase(smi))
        properties = {'bioactivity': bio}
        property_list.append(properties)
    except Exception as e:
        # write to log
        with open("papyrus1k-exceptions.log", "a") as file:
            file.write(f"Exception {type(e)} {e} for smi: {smi}\nbio: {bio}\ntarget: {target}\n")

new_dataset = ASEAtomsData.create(
    str(path_to_db / f"{target}.db"), 
    distance_unit='Ang',
    property_unit_dict={'bioactivity':'bool'}
)

new_dataset.add_systems(property_list, atoms_list)

So, new_dataset supposedly has a "bioactivity" property. After saving it to the .db file and loading with the command:

custom_data = spk.data.AtomsDataModule(
    str(path_db), 
    batch_size=10,
    distance_unit='Ang',
    split_file=None, # supresses the warning message regarding the split sizes mismatch
    num_train=1500, # total number of samples is 1881
    num_val=381,
    transforms=[
        trn.ASENeighborList(cutoff=5.),
        trn.CastTo32()
    ],
    load_properties=["bioactivity"],
    # property_units={'bioactivity':'bool'},
    num_workers=1,
    pin_memory=True, # set to false, when not using a GPU
)
custom_data.prepare_data()
custom_data.setup()

I am trying to get an access to the "bioactivity" property. How should I do that? I want to use this label as U0 is used in the tutorial 2 in the "Setting up the model" section.

I understand that the question probably should be addressed to the authors of ase package, but I appreciate it if you could provide any hint on how I can adapt your package (and tutorials in particular) to my data.

Best regards, Anton.

[jnsLs]

Dear Anton,

you can iterate through any subset of your data using the dataloader.

For the test set this could look like:

dataloader = custom_data.test_dataloader()

for sample_idx, sample in enumerate(tqdm(dataloader)):
    print(sample)

The fact that you defined bool as property unit might be a problem. I would recommend using an empty string "" instead. You should use those empty units when you define the dataset and when you load the dataset.

Best, Jonas

[CalmScout]

Dear Jonas,

I am figuring out how to use dataloaders in schnetpack. Meanwhile, while just copy paste the code you suggested above I received the following exception:

(pyg) (mambaforge) popova@mi001pc014:~/Projects/citre-benchmarking/scripts$ python tmp.py 
- bioactivity
  0%|          | 0/760 [00:00<?, ?it/s]
Traceback (most recent call last):
  File "/home/popova/Projects/citre-benchmarking/scripts/tmp.py", line 42, in <module>
    for sample_idx, sample in enumerate(tqdm(dataloader)):
  File "/home/popova/.pyenv/versions/mambaforge/envs/pyg/lib/python3.12/site-packages/tqdm/notebook.py", line 250, in __iter__
    for obj in it:
  File "/home/popova/.pyenv/versions/mambaforge/envs/pyg/lib/python3.12/site-packages/tqdm/std.py", line 1181, in __iter__
    for obj in iterable:
  File "/home/popova/.pyenv/versions/mambaforge/envs/pyg/lib/python3.12/site-packages/torch/utils/data/dataloader.py", line 631, in __next__
    data = self._next_data()
           ^^^^^^^^^^^^^^^^^
  File "/home/popova/.pyenv/versions/mambaforge/envs/pyg/lib/python3.12/site-packages/torch/utils/data/dataloader.py", line 1346, in _next_data
    return self._process_data(data)
           ^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/popova/.pyenv/versions/mambaforge/envs/pyg/lib/python3.12/site-packages/torch/utils/data/dataloader.py", line 1372, in _process_data
    data.reraise()
  File "/home/popova/.pyenv/versions/mambaforge/envs/pyg/lib/python3.12/site-packages/torch/_utils.py", line 705, in reraise
    raise exception
AttributeError: Caught AttributeError in DataLoader worker process 0.
Original Traceback (most recent call last):
  File "/home/popova/.pyenv/versions/mambaforge/envs/pyg/lib/python3.12/site-packages/torch/utils/data/_utils/worker.py", line 308, in _worker_loop
    data = fetcher.fetch(index)  # type: ignore[possibly-undefined]
           ^^^^^^^^^^^^^^^^^^^^
  File "/home/popova/.pyenv/versions/mambaforge/envs/pyg/lib/python3.12/site-packages/torch/utils/data/_utils/fetch.py", line 51, in fetch
    data = [self.dataset[idx] for idx in possibly_batched_index]
            ~~~~~~~~~~~~^^^^^
  File "/home/popova/.pyenv/versions/mambaforge/envs/pyg/lib/python3.12/site-packages/schnetpack/data/atoms.py", line 269, in __getitem__
    props = self._get_properties(
            ^^^^^^^^^^^^^^^^^^^^^
  File "/home/popova/.pyenv/versions/mambaforge/envs/pyg/lib/python3.12/site-packages/schnetpack/data/atoms.py", line 347, in _get_properties
    torch.tensor(row.data[pname].copy()) * self.conversions[pname]
                 ^^^^^^^^^^^^^^^^^^^^
AttributeError: 'int' object has no attribute 'copy'

Changed the property unit from bool to an empty string as was suggested, thank you.

Best regards, Anton.

[jnsLs]

Hi Anton,

this probably means that your target (bioactivity) does not have the suitable dimensions. You could try to expand the properties before writing it to the dataset. It should be a list of np.arrays with dimension (1,)

Best, Jonas