AttributeError: 'ecole.core.observation.NodeBipartiteObs' object has no attribute 'variable_features'

[ZhiweiShan]

Describe the bug

When I run the branching-imitation example, it was fine until:

observation = train_data[0].to(DEVICE)

logits = policy(
    observation.constraint_features,
    observation.edge_index,
    observation.edge_attr,
    observation.variable_features,
)
action_distribution = F.softmax(logits[observation.candidates], dim=-1)

print(action_distribution)

I got the error report:

AttributeError                            Traceback (most recent call last)
<ipython-input-13-4842d4bcbfd3> in <module>()
----> 1 observation = train_data[0].to(DEVICE)
      2 
      3 logits = policy(
      4     observation.constraint_features,
      5     observation.edge_index,

/home/dryi/anaconda3/envs/ecole2/lib/python3.6/site-packages/torch_geometric/data/dataset.py in __getitem__(self, idx)
    196                 or (isinstance(idx, np.ndarray) and np.isscalar(idx))):
    197 
--> 198             data = self.get(self.indices()[idx])
    199             data = data if self.transform is None else self.transform(data)
    200             return data

<ipython-input-10-c9da1c62e7ff> in get(self, index)
     66         edge_indices = sample_observation.edge_features.indices.astype(np.int32)
     67         edge_features = np.expand_dims(sample_observation.edge_features.values, axis=-1)
---> 68         variable_features = sample_observation.variable_features
     69 
     70         # We note on which variables we were allowed to branch, the scores as well as the choice

AttributeError: 'ecole.core.observation.NodeBipartiteObs' object has no attribute 'variable_features'

Setting

• OS: Ubuntu 20.04.3 LTS
• Python version: 3.6
• Ecole version: 0.7.3 (conda install -c conda-forge ecole automatic installation) I am using torch 1.8.1+cu101

Thank you for any help!

[AntoinePrv]

Hi @ZhiweiShan ,

We recently released Ecole 0.8.0, so the branching tutorial you are running is probably using that new version. You can either update Ecole, or rollback to the v0.7.3 tag on Github.

[ZhiweiShan]

Hi Antoine,

Thank you very much. It helps.