While running test tasks on 'mol_building_env', I encountered 'invalid_losses'. After investigating, I discovered that the shape of non_edge_index was incorrect. This issue can be easily fixed by setting non_edge_index = np.zeros((0, 2), dtype=np.int64). The correct shape is needed for line 382, where non_edge_index.shape[0] is used.
https://github.com/recursionpharma/gflownet/blob/f106cdeb6892214cbb528a3e06f4c721f4003175/src/gflownet/envs/mol_building_env.py#L355
While running test tasks on 'mol_building_env', I encountered 'invalid_losses'. After investigating, I discovered that the shape of
non_edge_indexwas incorrect. This issue can be easily fixed by settingnon_edge_index = np.zeros((0, 2), dtype=np.int64). The correct shape is needed for line 382, wherenon_edge_index.shape[0]is used.