Fix loss summation (#16)

Commit

7 years ago

Fix loss summation (#16) The loss summation relies on data being 1-d. Once Tensor and Variable are merged the loss.data will be 0-d.