Weighted euclidean loss

[AlexanderSavochkin]

Hi,

I propose weighted euclidean loss layer. I think it might be very useful, for example it can naturally handle missing labeling data in training of regression models.

This layer has additional input ("bottom") which should correspond 1 by 1 with labels. Values in this additional input represents label "weight"

[shaibagon]

Thank you for this contribution. There are a few things you can do to improve it, though

1. Please squash your commits into a single one.
2. Please fix lint errors (type make lint to see these error).
3. Make sure your PR passes the Travis automatic tests.

Thanks,

[AlexanderSavochkin]

Thanks @shaibagon . I am working on the fixes