updates for torchtext and loading from snapshot

Author: bmccann

snli/train.py

@@ -15,12 +15,12 @@
 
 args = get_args()
 
-inputs = data.Field()
+inputs = data.Field(lower=args.lower)
 answers = data.Field(sequential=False)
 
 train, dev, test = datasets.SNLI.splits(inputs, answers)
 
-inputs.build_vocab(train, dev, test, lower=args.lower)
+inputs.build_vocab(train, dev, test)
 if args.word_vectors:
  if os.path.isfile(args.vector_cache):
  inputs.vocab.vectors = torch.load(args.vector_cache)
@@ -40,10 +40,14 @@
 if config.birnn:
  config.n_cells *= 2
 
-model = SNLIClassifier(config)
-if args.word_vectors:
- model.embed.weight.data = inputs.vocab.vectors
-model.cuda()
+if args.resume_snapshot:
+ model = torch.load(args.resume_snapshot, map_location=lambda storage, locatoin: storage.cuda(args.gpu))
+else:
+ model = SNLIClassifier(config)
+ if args.word_vectors:
+ model.embed.weight.data = inputs.vocab.vectors
+ model.cuda(args.gpu)
+
 criterion = nn.CrossEntropyLoss()
 opt = O.Adam(model.parameters(), lr=args.lr)
 

snli/util.py

@@ -23,5 +23,6 @@ def get_args():
  parser.add_argument('--data_cache', type=str, default=os.path.join(os.getcwd(), '.data_cache'))
  parser.add_argument('--vector_cache', type=str, default=os.path.join(os.getcwd(), '.vector_cache/input_vectors.pt'))
  parser.add_argument('--word_vectors', type=str, default='glove.42B')
+ parser.add_argument('--resume_snapshot', type=str, default='results/snapshot.pt')
  args = parser.parse_args()
  return args