fixing preprocessing issue with gpu

setup.py

@@ -0,0 +1,10 @@
+# pip install -U Flask
+# pip install -U flask-cors
+# conda install -c anaconda h5py
+# pip install PeptideBuilder
+# pip install requests
+# git clone https://github.com/JeppeHallgren/pytorch-crf.git
+# cd pytorch-crf/
+# pip install -r requirements.txt
+# pip install -e .
+

util.py

@@ -139,9 +139,13 @@ def calculate_dihedral_angles_over_minibatch(atomic_coords_padded, batch_sizes,
  atomic_coords = atomic_coords_padded.transpose(0, 1)
  for idx, _ in enumerate(batch_sizes):
  angles.append(calculate_dihedral_angles(atomic_coords[idx][:batch_sizes[idx]], use_gpu))
- return torch.nn.utils.rnn.pad_packed_sequence(
- torch.nn.utils.rnn.pack_sequence(angles))
+ # this was problematic:
+ packed_angles = torch.nn.utils.rnn.pack_sequence(angles)
+ if use_gpu:
+ packed_angles=packed_angles.cuda()
+ angles_padded = torch.nn.utils.rnn.pad_packed_sequence(packed_angles)
 
+ return angles_padded
 
 def protein_id_to_str(protein_id_list):
  _aa_dict_inverse = {v: k for k, v in AA_ID_DICT.items()}