Creating multiple Neural Networks on the fly (Lua, torch)

[aseem prashar]

Hi everyone,

For the purpose of my thesis I wanna be able to create multiple Neural Networks with fixed confuguration(number of layers) on the fly.

I have tried eveyrthing but I am stuck on the following: creating a new and unique name for each NN.

Here is the method to create Neural Networks:

function CreateNN(self)

self = nn.Sequential()

self:add(nn.Reshape(1024))

self:add(nn.Linear(inputLayer, hiddenLayer1))

self:add(nn.ReLU())

self:add(nn.Linear(hiddenLayer1, hiddenLayer2))

self:add(nn.ReLU())

self:add(nn.Linear(hiddenLayer2, outputSize))

self:add(nn.LogSoftMax())

return self

end

Here is how I use it:

baseName=mlp

name1_G=mlpname..1

print(name1_G)  -- output: mlp1

name1_G=CreateNN()

print(mlp1) --output: nil

As you can see, the neural network is created for the name1_G and not mlp1 as intended. How do i overcome this?

Thank you!