[Question] Embeddings for categorical variables

[esbraun]

Greetings,

We have categorical features in our observation data. Are we able to modify the CustomEncoderFactory for the custom neural network to include categorical features to build embeddings?

I'm familiar with how to do with a vanilla torch dataset, i.e.

`class ModelDataset(Dataset):
def init(self, df, cat_fields, cont_fields, y):
self.df = df
self.y = y.astype(np.float32)
cat_values = [c.values for n,c in df[cat_fields].items()]
cont_values = [c.values for n,c in df[cont_fields].items()]
self.cat_features = np.stack(cat_values, 1).astype(np.int64)
self.cont_features = np.stack(cont_values, 1).astype(np.float32)

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

def __getitem__(self, idx):
    cat_val = self.cat_features[idx]
    cont_val = self.cont_features[idx]
    y = self.y[idx]
    return [cat_val, cont_val, y]`

[takuseno]

@esbraun Hi, thanks for the issue. You can build your own custom network for training.
https://d3rlpy.readthedocs.io/en/v2.0.4/tutorials/customize_neural_network.html