allowing validation data to volatile

Author: bmccann

OpenNMT/onmt/Dataset.py

@@ -6,7 +6,7 @@
 
 class Dataset(object):
 
- def __init__(self, srcData, tgtData, batchSize, cuda):
+ def __init__(self, srcData, tgtData, batchSize, cuda, volatile=False):
  self.src = srcData
  if tgtData:
  self.tgt = tgtData
@@ -16,7 +16,8 @@ def __init__(self, srcData, tgtData, batchSize, cuda):
  self.cuda = cuda
 
  self.batchSize = batchSize
- self.numBatches = (len(self.src) + batchSize - 1) // batchSize
+ self.numBatches = len(self.src) // batchSize
+ self.volatile = volatile
 
  def _batchify(self, data, align_right=False):
  max_length = max(x.size(0) for x in data)
@@ -30,7 +31,7 @@ def _batchify(self, data, align_right=False):
  if self.cuda:
  out = out.cuda()
 
- v = Variable(out)
+ v = Variable(out, volatile=self.volatile)
  return v
 
  def __getitem__(self, index):

OpenNMT/train.py

@@ -269,7 +269,8 @@ def main():
  trainData = onmt.Dataset(dataset['train']['src'],
  dataset['train']['tgt'], opt.batch_size, opt.gpus)
  validData = onmt.Dataset(dataset['valid']['src'],
- dataset['valid']['tgt'], opt.batch_size, opt.gpus)
+ dataset['valid']['tgt'], opt.batch_size, opt.gpus,
+ volatile=True)
 
  dicts = dataset['dicts']
  print(' * vocabulary size. source = %d; target = %d' %