

In [1]:

    
require 'nn';
require 'torch';

Prepair data



In [54]:

    
local matio = require 'matio'
data = matio.load('ex4data1.mat')



In [55]:

    
trainset = {}
trainset.data = data.X
trainset.label = data.y[{ {}, 1}]



In [56]:

    
trainset



In [57]:

    
setmetatable(trainset,
    {__index = function(t,i)
                return {t.data[i], t.label[i]}
        end}
);
 
function trainset:size()
    return self.data:size(1)
end









    Out[57]:





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



In [58]:

    
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

define model



In [59]:

    
net = nn.Sequential()
net:add(nn.Linear(400,25))
net:add(nn.Sigmoid())
net:add(nn.Linear(25,10))
net:add(nn.Sigmoid())
net:add(nn.LogSoftMax())

define loss function



In [60]:

    
criterion = nn.ClassNLLCriterion()

let's train



In [61]:

    
trainer = nn.StochasticGradient(net, criterion)
trainer.learningRate = 0.001
trainer.maxIteration = 1e2



In [62]:

    
trainer:train(trainset)









    Out[62]:





# StochasticGradient: training	







    Out[62]:





# current error = 2.2919424613996	







    Out[62]:





# current error = 2.2623607889839	







    Out[62]:





# current error = 2.2320541713042	







    Out[62]:





# current error = 2.2006777919079	







    Out[62]:





# current error = 2.168539441064	







    Out[62]:





# current error = 2.136266976882	







    Out[62]:





# current error = 2.1045216386857	







    Out[62]:





# current error = 2.0738772603513	







    Out[62]:





# current error = 2.0447717893028	







    Out[62]:





# current error = 2.0174868472812	







    Out[62]:





# current error = 1.9921557315415	







    Out[62]:





# current error = 1.9687918809252	







    Out[62]:





# current error = 1.9473252182393	







    Out[62]:





# current error = 1.9276363789713	







    Out[62]:





# current error = 1.9095829761714	







    Out[62]:





# current error = 1.8930167082486	







    Out[62]:





# current error = 1.8777932931237	







    Out[62]:





# current error = 1.8637776287851	







    Out[62]:





# current error = 1.8508455714591	







    Out[62]:





# current error = 1.8388839048255	







    Out[62]:





# current error = 1.8277902435673	







    Out[62]:





# current error = 1.8174728210088	







    Out[62]:





# current error = 1.8078499691825	







    Out[62]:





# current error = 1.7988494845912	







    Out[62]:





# current error = 1.7904078209015	







    Out[62]:





# current error = 1.7824692704477	







    Out[62]:





# current error = 1.7749852483583	







    Out[62]:





# current error = 1.7679136182669	







    Out[62]:





# current error = 1.7612179979538	







    Out[62]:





# current error = 1.7548670215456	







    Out[62]:





# current error = 1.7488335768304	







    Out[62]:





# current error = 1.7430940597637	







    Out[62]:





# current error = 1.7376276846547	







    Out[62]:





# current error = 1.7324158913333	







    Out[62]:





# current error = 1.7274418922739	







    Out[62]:





# current error = 1.7226903699712	







    Out[62]:





# current error = 1.7181472968042	







    Out[62]:





# current error = 1.7137998219794	







    Out[62]:





# current error = 1.7096361358339	







    Out[62]:





# current error = 1.7056452713351	







    Out[62]:





# current error = 1.7018169467665	







    Out[62]:





# current error = 1.6981415234959	







    Out[62]:





# current error = 1.6946100062297	







    Out[62]:





# current error = 1.6912140229238	







    Out[62]:





# current error = 1.6879457858581	







    Out[62]:





# current error = 1.6847980478246	







    Out[62]:





# current error = 1.6817640592968	







    Out[62]:





# current error = 1.6788375278242	







    Out[62]:





# current error = 1.676012580153	







    Out[62]:





# current error = 1.6732837276552	







    Out[62]:





# current error = 1.6706458352558	







    Out[62]:





# current error = 1.6680940935506	







    Out[62]:





# current error = 1.6656239937012	







    Out[62]:





# current error = 1.6632313047934	







    Out[62]:





# current error = 1.6609120536788	







    Out[62]:





# current error = 1.6586625073457	







    Out[62]:





# current error = 1.6564791577077	







    Out[62]:





# current error = 1.6543587085247	







    Out[62]:





# current error = 1.6522980639652	







    Out[62]:





# current error = 1.6502943181067	







    Out[62]:





# current error = 1.6483447445889	







    Out[62]:





# current error = 1.6464467855753	







    Out[62]:





# current error = 1.6445980398386	







    Out[62]:





# current error = 1.6427962509004	







    Out[62]:





# current error = 1.6410392967841	







    Out[62]:





# current error = 1.6393251817682	







    Out[62]:





# current error = 1.6376520286444	







    Out[62]:





# current error = 1.6360180698685	







    Out[62]:





# current error = 1.6344216381803	







    Out[62]:





# current error = 1.6328611589585	







    Out[62]:





# current error = 1.6313351458557	







    Out[62]:





# current error = 1.6298421989898	







    Out[62]:





# current error = 1.6283810037427	







    Out[62]:





# current error = 1.6269503286812	







    Out[62]:





# current error = 1.6255490221372	







    Out[62]:





# current error = 1.6241760076194	







    Out[62]:





# current error = 1.6228302784771	







    Out[62]:





# current error = 1.6215108921748	







    Out[62]:





# current error = 1.6202169644662	







    Out[62]:





# current error = 1.618947663716	







    Out[62]:





# current error = 1.6177022055761	







    Out[62]:





# current error = 1.6164798481998	







    Out[62]:





# current error = 1.615279888089	







    Out[62]:





# current error = 1.6141016566154	







    Out[62]:





# current error = 1.6129445171429	







    Out[62]:





# current error = 1.6118078626521	







    Out[62]:





# current error = 1.6106911137369	







    Out[62]:





# current error = 1.6095937168614	







    Out[62]:





# current error = 1.6085151428036	







    Out[62]:





# current error = 1.6074548852351	







    Out[62]:





# current error = 1.6064124594164	







    Out[62]:





# current error = 1.6053874010077	







    Out[62]:





# current error = 1.6043792650534	







    Out[62]:





# current error = 1.6033876251947	







    Out[62]:





# current error = 1.6024120733069	







    Out[62]:





# current error = 1.6014522198081	







    Out[62]:





# current error = 1.6005076949361	







    Out[62]:





# current error = 1.5995781510644	







    Out[62]:





# current error = 1.5986632653485	







    Out[62]:





# current error = 1.5977627407013	
# StochasticGradient: you have reached the maximum number of iterations	
# training error = 1.5977627407013

test



In [63]:

    
correction = 0
for i=1,trainset:size() do
    local answer = trainset.label[i]
    local prediction = net:forward(trainset.data[i])
    local confidences, indices = torch.sort(prediction, true)
    if (answer == indices[1]) then
        correction = correction + 1
    end
end
print(correction, 100*correction/trainset:size() .. '%')









    Out[63]:





4615	92.3%



In [ ]: