batch-represent: Remove unused functions.

batch-represent/dataset.lua

@@ -365,129 +365,6 @@ local function tableToOutput(self, dataTable, scalarTable)
    return data, scalarLabels, labels
 end
 
--- sampler, samples from the training set.
-function dataset:sample(quantity)
-   if self.split == 0 then
-      error('No training mode when split is set to 0')
-   end
-   quantity = quantity or 1
-   local dataTable = {}
-   local scalarTable = {}
-   for _=1,quantity do
-      local class = torch.random(1, #self.classes)
-      local out = self:getByClass(class)
-      table.insert(dataTable, out)
-      table.insert(scalarTable, class)
-   end
-   local data, scalarLabels, labels = tableToOutput(self, dataTable, scalarTable)
-   return data, scalarLabels, labels
-end
-
--- TODO: Triplet selection.
---       This naively randomly samples for triplets.
-function dataset:sampleTriplet(quantity)
-   if self.split == 0 then
-      error('No training mode when split is set to 0')
-   end
-   quantity = quantity or 1
-   local dataTable = {}
-   local scalarTable = {}
-
-   -- Anchors
-   for _=1,quantity do
-      local anchorClass = torch.random(1, #self.classes)
-      table.insert(dataTable, self:getByClass(anchorClass))
-      table.insert(scalarTable, anchorClass)
-   end
-
-   -- Positives
-   for i=1,quantity do
-      local posClass = scalarTable[i]
-      table.insert(dataTable, self:getByClass(posClass))
-      table.insert(scalarTable, posClass)
-   end
-
-   -- Negatives
-   for i=1,quantity do
-      local posClass = scalarTable[i]
-      local negClass = posClass
-      while negClass == posClass do
-         negClass = torch.random(1, #self.classes)
-      end
-      table.insert(dataTable, self:getByClass(negClass))
-      table.insert(scalarTable, negClass)
-   end
-   local data, scalarLabels, labels = tableToOutput(self, dataTable, scalarTable)
-   return data, scalarLabels, labels
-end
-
-
-function dataset:samplePeople(peoplePerBatch, imagesPerPerson)
-   if self.split == 0 then
-      error('No training mode when split is set to 0')
-   end
-
-   local classes = torch.randperm(#trainLoader.classes)[{{1,peoplePerBatch}}]:int()
-   local numPerClass = torch.Tensor(peoplePerBatch)
-   for i=1,peoplePerBatch do
-      local n = math.min(self.classListSample[classes[i]]:nElement(), imagesPerPerson)
-      numPerClass[i] = n
-   end
-
-   local data = torch.Tensor(numPerClass:sum(),
-                             self.sampleSize[1], self.sampleSize[2], self.sampleSize[3])
-
-   local dataIdx = 1
-   for i=1,peoplePerBatch do
-      local cls = classes[i]
-      local n = numPerClass[i]
-      local shuffle = torch.randperm(n)
-      for j=1,n do
-         imgNum = self.classListSample[cls][shuffle[j]]
-         imgPath = ffi.string(torch.data(self.imagePath[imgNum]))
-         data[dataIdx] = self:sampleHookTrain(imgPath)
-         dataIdx = dataIdx + 1
-      end
-   end
-   assert(dataIdx - 1 == numPerClass:sum())
-
-   return data, numPerClass
-end
-
-function dataset:sampleAllOfClass(quantity, posClass)
-   if self.split == 0 then
-      error('No training mode when split is set to 0')
-   end
-   quantity = quantity or 1
-
-   local n = math.min(self.classListSample[posClass]:nElement(), quantity)
-   local data = torch.Tensor(n, self.sampleSize[1],
-                             self.sampleSize[2], self.sampleSize[3])
-   for i=1,n do
-      imgpath = ffi.string(torch.data(self.imagePath[self.classListSample[posClass][i]]))
-      data[i] = self:sampleHookTrain(imgpath)
-   end
-
-   return data
-end
-
-function dataset:sampleNoneOfClass(quantity, posClass)
-   if self.split == 0 then
-      error('No training mode when split is set to 0')
-   end
-   quantity = quantity or 1
-   local data = torch.Tensor(quantity, self.sampleSize[1],
-                             self.sampleSize[2], self.sampleSize[3])
-   for i=1,quantity do
-      local negClass = posClass
-      while negClass == posClass do
-         negClass = torch.random(1, #self.classes)
-      end
-      data[i] = self:getByClass(negClass)
-   end
-   return data
-end
-
 function dataset:get(i1, i2)
    local indices, quantity
    if type(i1) == 'number' then