Questions about using DGLLife with DGL

[code4luck]

Thank you for your ongoing support. I have a question regarding an issue I'm facing when using the GINConv model from DGL with data features for small molecules processed by DGLLife. I use
`
class ESIDataset(data.Dataset):
def init(self, list_IDs, df):
self.list_IDs = list_IDs
self.df = df
self.atom_featurizer = CanonicalAtomFeaturizer()
self.bond_featurizer = CanonicalBondFeaturizer(self_loop=True)
self.fc = partial(smiles_to_bigraph, add_self_loop=True)

def __len__(self):
    return len(self.list_IDs)

def __getitem__(self, index):
    index = self.list_IDs[index]
    v_d = self.df.iloc[index]['SMILES']
    v_d = self.fc(smiles=v_d, node_featurizer=self.atom_featurizer, edge_featurizer=self.bond_featurizer) 
    return v_d

for the SMEILS datasets. While I use from dgl.nn.pytorch import GINConv to make a model, I'm wondering if this approach to building the model is compatible with the data features in DGLLife. node_feats = batch_graph.ndata.pop('h') node_feats = gin(batch_graph, node_feats) ,Thanks!

[mufeili]

Could you be more specific about what did you mean by "compatible"?

[code4luck]

Could you be more specific about what did you mean by "compatible"?

Thanks for your reply. I want to know whether my dataset features(code) are correct for the model inputs? because I use the DGL GinConv module to make a GIN model, which is not the same as GIN model in DGLLife. So I don't know my code is correct.🙏

[mufeili]

As long as your inputs along with the GINConv input requirements, you should be fine. In general, GinConv takes as input a graph and its node features, which can be for example a molecular graph and atom features.