get-mnist.lua

local util = require 'autograd.util'

-- Get/create dataset
local function setupData(opt)
   if not path.exists(sys.fpath()..'/mnist') then
      os.execute[[
      curl https://s3.amazonaws.com/torch.data/mnist.tgz -o mnist.tgz
      tar xvf mnist.tgz
      rm mnist.tgz
      ]]
   end

   local classes = {'0','1','2','3','4','5','6','7','8','9'}
   local trainData = torch.load(sys.fpath()..'/mnist/train.t7')
   local testData = torch.load(sys.fpath()..'/mnist/test.t7')
   trainData.y = util.oneHot(trainData.y)
   testData.y = util.oneHot(testData.y)
   return trainData, testData, classes
end

return setupData