Unexpected behavior at torch7

Question

I was working on generating an XOR gate dataset with torch7. But when i printed the dataset i saw that the data was wrong, but i could not find the bug. There seems to be nothing wrong with the code. But I'm new to torch, so mistakes can happen.

So, here is my code

input = torch.Tensor (4,2) 
input:random(0,1)

output = torch.Tensor(1)
dataset={};
function dataset:size() return 4 end 
for i=1,dataset:size() do 
  if input[i][1]==input[i][2] then
    output[1] = 0
  else
    output[1] = 1
  end
  print("original")
  print(input[i][1].." "..input[i][2].." "..output[1]) -- the values that are going to dataset
  dataset[i] = {input[i], output}
  print("dataset")
  print(dataset[i][1][1].." "..dataset[i][1][2].." "..dataset[i][2][1])  -- for double checking i read from dataset again
end

print("Why dataset is different now?")
for i=1,4 do
    print(dataset[i][1][1].." "..dataset[i][1][2].." "..dataset[i][2][1]) -- So, why this is different?
    end

As you can see, I printed the values that are being inserted into the dataset list and for double checking i read from dataset again.

And finally i checked from dataset after full insertion. The dataset was different somehow. I ran couple of times. Every time it was different. Like it was stuck on 1 or 0.

So here is my output

original    
1 0 1   
dataset 
1 0 1   
original    
0 0 0   
dataset 
0 0 0   
original    
1 1 0   
dataset 
1 1 0   
original    
0 0 0   
dataset 
0 0 0   
Why dataset is different now?   
1 0 0   
0 0 0   
1 1 0   
0 0 0   

As you can see, the format is like this

input input output

I printed original when i read from input[i] and output.

I printed dataset when i read from dataset, after being inserted.

Also you can see that the first set of values are different when i printed. It should be 1 0 1. But it is 1 0 0.

I could not find the bug in my code. Can anyone help? If the question is not clear please let me know.

Answer 1

Problem is here: dataset[i] = {input[i], output}
You're not saving calculated result, you're saving reference to value that is changed with subsequent calculations for 'xor' function.
Naturally, when you read result, you're always getting the same number - last result written to output[1]
To fix it, either change output variable to store actual temporary value (not table), or at least read actual value from output table when saving to dataset[i], do not just save link to table, you won't get deep copy that way.