Training on datasets without y attribute (ex: unsupervised-learning) throws AttributeError

[arunppsg]

To Reproduce

import csv
import deepchem as dc
from deepchem.models.torch_models import InfoGraphModel

# Generate sample dataset without y value
with open('smiles.csv', 'w', newline='') as csvfile:
    fieldnames = ['smiles']
    writer = csv.DictWriter(csvfile, fieldnames=fieldnames)
    writer.writeheader()
    for i in range(100):
        writer.writerow({'smiles': 'Cc1[nH]ccc1C(=O)N[C@@H](C)CNC(=O)c1cccc2[nH]ccc21'})

featurizer = dc.feat.MolGraphConvFeaturizer(use_edges=True)

loader = dc.data.CSVLoader([], feature_field='smiles', featurizer=featurizer)
dataset = loader.create_dataset('smiles.csv', data_dir='temp-dir')

# Fit on model
loss = model.fit(dataset, nb_epoch=1)
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
Cell In[4], line 1
----> 1 loss = model.fit(dataset, nb_epoch=1)

...

File ~/deepchem/models/torch_models/torch_model.py:921, in <listcomp>(.0)
    918 input_tensors = [torch.as_tensor(x, device=self.device) for x in inputs]
    919 if labels is not None:
    920     labels = [
--> 921         x.astype(np.float32) if x.dtype == np.float64 else x
    922         for x in labels
    923     ]
    924     label_tensors = [
    925         torch.as_tensor(x, device=self.device) for x in labels
    926     ]
    927 else:

AttributeError: 'NoneType' object has no attribute 'dtype'

Potential Solution

Adding a check for y values before this line

deepchem/deepchem/models/torch_models/torch_model.py

Line 985 in ae7880e

yield ([X_b], [y_b], [w_b])

to return None for y when it is None instead of returning [None].