now running on mini-batches

Diag.lua

@@ -14,21 +14,29 @@ end
 
 function Diag:updateOutput(input)
    self.output:resizeAs(input):copy(input)
-   if input:dim() > 1 then
-      for i=1,input:size(1) do
-	 self.output[{{i}}]:mul(self.weight[i])
+   if input:dim() == 4 then -- batch mode
+      for i=1,input:size(2) do
+	 self.output[{{},{i}}]:mul(self.weight[i])
       end
+   elseif input:dim() == 3 then -- no batch
+    for i=1,input:size(1) do
+      self.output[{{i}}]:mul(self.weight[i])
+    end
    else
-      self.output:cmul(self.weight)
+     self.output:cmul(self.weight)
    end
    return self.output
 end
 
 function Diag:updateGradInput(input, gradOutput)
    self.gradInput:resizeAs(gradOutput):copy(gradOutput)
-   if input:dim() > 1 then
+   if input:dim() == 4 then -- batch mode
+      for i=1,input:size(2) do
+	 self.gradInput[{{},{i}}]:mul(self.weight[i])
+      end
+   elseif input:dim() == 3 then
       for i=1,input:size(1) do
-	 self.gradInput[{{i}}]:mul(self.weight[i])
+         self.gradInput[{{i}}]:mul(self.weight[i])
       end
    else
       self.gradInput:cmul(self.weight)
@@ -37,8 +45,14 @@ function Diag:updateGradInput(input, gradOutput)
 end
 
 function Diag:accGradParameters(input, gradOutput, scale)
+  if input:dim() == 4 then -- batch mode
+   for i=1,input:size(2) do
+      self.gradWeight[i] = self.gradWeight[i] + scale*gradOutput[{{},{i}}]:dot(input[{{},{i}}])
+   end
+  elseif input:dim() == 3 then
    for i=1,input:size(1) do
       self.gradWeight[i] = self.gradWeight[i] + scale*gradOutput[{{i}}]:dot(input[{{i}}])
    end
+  end
 end
 

SpatialConvFistaL1.lua

@@ -8,7 +8,7 @@ local SpatialConvFistaL1, parent = torch.class('unsupgpu.SpatialConvFistaL1','un
 -- padh            : zero padding vertical
 -- lambda          : sparsity coefficient
 -- params          : optim.FistaLS parameters
-function SpatialConvFistaL1:__init(conntable, kw, kh, iw, ih, padw, padh, lambda, params)
+function SpatialConvFistaL1:__init(conntable, kw, kh, iw, ih, padw, padh, lambda, batchSize)
 
    -- parent.__init(self)
 
@@ -22,18 +22,29 @@ function SpatialConvFistaL1:__init(conntable, kw, kh, iw, ih, padw, padh, lambda
    -----------------------------------------
    -- L2 reconstruction cost with weighting
    -----------------------------------------
-   local tt = torch.Tensor(inputFeatures,ih,iw)
-   local utt= tt:unfold(2,kh,1):unfold(3,kw,1)
+   local batchSize = batchSize or 1
+   local tt, utt
+   if batchSize > 1 then 
+     tt = torch.Tensor(batchSize,inputFeatures,ih,iw)
+     utt= tt:unfold(3,kh,1):unfold(4,kw,1)
+   else
+     tt = torch.Tensor(inputFeatures,ih,iw)
+     utt= tt:unfold(2,kh,1):unfold(3,kw,1)
+   end
    tt:zero()
    utt:add(1)
    tt:div(tt:max())
    local Fcost = nn.WeightedMSECriterion(tt)
    Fcost.sizeAverage = false;
 
-   parent.__init(self,D,Fcost,lambda,params)
+   parent.__init(self,D,Fcost,lambda)
 
    -- this is going to be passed to optim.FistaLS
-   self.code:resize(outputFeatures, utt:size(2)+2*padw,utt:size(3)+2*padh)
+   if batchSize > 1 then
+     self.code:resize(batchSize, outputFeatures, utt:size(3)+2*padw,utt:size(4)+2*padh)
+   else
+     self.code:resize(outputFeatures, utt:size(2)+2*padw,utt:size(3)+2*padh)
+   end
    self.code:fill(0)
 end