NG

local PReLU, parent = torch.class('nn.NG','nn.Module')

function PReLU:__init(nOutputPlane)
   parent.__init(self)
   -- if no argument provided, use shared model (weight is scalar)
   self.nOutputPlane = nOutputPlane or 0
   self.weight = torch.Tensor(nOutputPlane or 1):fill(0)
   self.gradWeight = torch.Tensor(nOutputPlane or 1)
   self.buf = torch.Tensor()
   self.buf2 = torch.Tensor()
end

function PReLU:updateOutput(input)
   self.buf:resizeAs(input):copy(self.weight:view(1,self.nOutputPlane,1,1):expandAs(input))
   self.output:resizeAs(input):copy(input)
   self.output:cmax(self.buf)
   return self.output
end

function PReLU:updateGradInput(input, gradOutput)
   if self.gradInput then
    self.gradInput:resizeAs(input):copy(gradOutput)
    self.gradInput[torch.lt(input,self.buf)]=0
    return self.gradInput
   end
end

function PReLU:accGradParameters(input, gradOutput, scale)
   self.buf2:resizeAs(input):copy(gradOutput)
   self.buf2[torch.gt(input,self.buf)]=0
   self.gradWeight:add(torch.mul(torch.sum(torch.sum(torch.sum(self.buf2,1),3),4):view(self.nOutputPlane),scale or 1))
   return self.gradWeight
end