In [1]:

    

require 'nn'
require 'optim'


    

In [2]:

    

local matio = require 'matio'
data = matio.load('ex3data1.mat')


    

In [97]:

    

trainset = {}


    

In [98]:

    

trainset.data = data.X


    

In [99]:

    

mean = {}
stdv = {}
for i=1,400 do
    mean[i] = trainset.data[{ {},{i} }]:mean()
    stdv[i] = trainset.data[{ {}, {i} }]:std()
    --print(i .. 'th mean: ' .. mean[i])
    --print(i .. 'th std dev: ' .. stdv[i])
    trainset.data[{ {},{i} }]:add(-mean[i])
    if stdv[i] ~= 0 then
        trainset.data[{ {},{i} }]:div(stdv[i])
    end
end


    

In [100]:

    

trainset.label = torch.Tensor(data.y):resize(5000)


    

In [101]:

    

trainset


    

In [102]:

    

numOutput = torch.max(trainset.label) - torch.min(trainset.label) + 1


    

    
Out[102]:





{
  data : DoubleTensor - size: 5000x400
  label : DoubleTensor - size: 5000
}

In [103]:

    

-- ignore setmetatable for now, it is a feature beyond the scope of this tutorial. It sets the index operator.
setmetatable(trainset,
    {__index = function(t,i)
                return {t.data[i], t.label[i]}
        end}
);

function trainset:size()
    return self.data:size(1)
end


    

In [104]:

    

model = nn.Sequential()
model:add(nn.Linear(400,10))
model:add(nn.Sigmoid())
model:add(nn.LogSoftMax())


    

In [105]:

    

criterion = nn.ClassNLLCriterion()


    

In [106]:

    

trainer = nn.StochasticGradient(model, criterion)
trainer.learningRate = 0.001
trainer.maxIteration = 2e1


    

In [107]:

    

trainset


    

In [108]:

    

trainer:train(trainset)


    

    
Out[108]:





{
  data : DoubleTensor - size: 5000x400
  size : function: 0x40a53cc8
  label : DoubleTensor - size: 5000
}
# StochasticGradient: training	







    
Out[108]:





# current error = 2.0314825492176	







    
Out[108]:





# current error = 1.892785270708	







    
Out[108]:





# current error = 1.857515794643	







    
Out[108]:





# current error = 1.8362176983487	







    
Out[108]:





# current error = 1.8201991297292	







    
Out[108]:





# current error = 1.8070407034207	







    
Out[108]:





# current error = 1.7957306828976	







    
Out[108]:





# current error = 1.7857439502779	







    
Out[108]:





# current error = 1.7767694930645	







    
Out[108]:





# current error = 1.7686057142731	







    
Out[108]:





# current error = 1.7611128587086	







    
Out[108]:





# current error = 1.7541888513453	







    
Out[108]:





# current error = 1.7477559231712	







    
Out[108]:





# current error = 1.7417525512366	







    
Out[108]:





# current error = 1.7361285707694	







    
Out[108]:





# current error = 1.7308424704224	







    
Out[108]:





# current error = 1.725859821649	







    
Out[108]:





# current error = 1.7211520244222	







    
Out[108]:





# current error = 1.7166950702014	

In [109]:

    

acc = 0.0
for i = 1, trainset:size() do
    local groundtruth = trainset.label[i]
    local prediction = model:forward(trainset.data[i])
    local confidences, indices = torch.sort(prediction, true) -- true means descending order
    if groundtruth == indices[1] then
        acc = acc + 1
    end
end
print('accuracy: ' .. 100*acc/trainset:size() .. '%')


    

    
Out[109]:





accuracy: 84.48%	

In [ ]: