div_ only supports scalar multiplication

[rawatenator]

Going to look into this in detail now, but got the following error while running the imagenet ipynb

RuntimeErrorTraceback (most recent call last)
<ipython-input-17-c4f537643c1f> in <module>()
      8 prob_outputs_dog = Variable(torch.zeros(1,1000)) ; prob_outputs_dog.data[:,dog_id] += 1
      9 
---> 10 prob_inputs_cat = eb.excitation_backprop(model, inputs, prob_outputs_cat, contrastive=False)
     11 prob_inputs_dog = eb.excitation_backprop(model, inputs, prob_outputs_dog, contrastive=False)

/home/rrawat/coderepo/excitationbp/excitationbp/utils.py in excitation_backprop(model, inputs, prob_outputs, contrastive, target_layer)
     45     if not contrastive:
     46         outputs = model(inputs)
---> 47         return torch.autograd.grad(top_h_, target_h_, grad_outputs=prob_outputs)[0]
     48 
     49     pos_evidence = torch.autograd.grad(top_h_, contr_h_, grad_outputs=prob_outputs.clone())[0]

/home/rrawat/anaconda2/lib/python2.7/site-packages/torch/autograd/__init__.pyc in grad(outputs, inputs, grad_outputs, retain_graph, create_graph, only_inputs)
    151     return Variable._execution_engine.run_backward(
    152         outputs, grad_outputs, retain_graph,
--> 153         inputs, only_inputs)
    154 
    155 if not torch._C._autograd_init():

/home/rrawat/anaconda2/lib/python2.7/site-packages/torch/autograd/function.pyc in apply(self, *args)
     89 
     90     def apply(self, *args):
---> 91         return self._forward_cls.backward(self, *args)
     92 
     93 

/home/rrawat/coderepo/excitationbp/excitationbp/functions/eb_linear.pyc in backward(ctx, grad_output)
     28 
     29         input.data = input.data - input.data.min() if input.data.min() < 0 else input.data
---> 30         grad_output /= input.mm(weight.t()).abs() + 1e-10 # normalize
     31         ### stop EB-SPECIFIC CODE  ###
     32 

/home/rrawat/anaconda2/lib/python2.7/site-packages/torch/autograd/variable.pyc in __idiv__(self, other)
    847 
    848     def __idiv__(self, other):
--> 849         return self.div_(other)
    850 
    851     def __pow__(self, other):

/home/rrawat/anaconda2/lib/python2.7/site-packages/torch/autograd/variable.pyc in div_(self, other)
    357         if not isinstance(other, Variable) and not torch.is_tensor(other):
    358             return DivConstant.apply(self, other, True)
--> 359         raise RuntimeError("div_ only supports scalar multiplication")
    360 
    361     def pow(self, other):

RuntimeError: div_ only supports scalar multiplication

[greydanus]

Interesting. I just tried to reproduce your error with PyTorch 0.3 and the latest commit of excitationbp. I wasn't able to reproduce your error.

I did push a small two-line update to this repo about an hour ago, but I'm not sure how that could have caused the error you are getting.