fix.lua

--[[

	- here's a better way to train RNNLM with large dataset, 
		- persist the weights from LookupTable into file 
		- and load it back later to use in sentiment analysis tasks

	- detailed discussion here: https://github.com/torch/nn/issues/747

	- thanks @fmassa for pointing out the direct direction

	- one usually simply copy the weights from a pre-trained model. For example,

		old_lookup = torch.load(...) -- can be a full model
		lookup = nn.LookupTable(...)
		lookup.weight:copy(old_lookup.weight)
		-- add lookup in your network

	- Another way (if you only want to finetune the last layer for example) is to load the full network,
	- removing the last layer and adding a new randomly-initialized layer on top of the network.

		net = torch.load(...)
		net:remove() -- remove last layer
		net:add(nn.Linear(4096,21)) -- add new layer replacing the old one

]]--

require 'nn'
require 'rnn'


criterion = nn.ClassNLLCriterion()

lookup = nn.LookupTable(3, 4)

mlp = nn.Sequential()
mlp:add( lookup)
mlp:add( nn.SplitTable(1, 2))
mlp:add( nn.Sequencer(nn.FastLSTM(4, 4)) )
mlp:add( nn.SelectTable(-1) )
mlp:add( nn.Linear( 4, 3))
mlp:add( nn.LogSoftMax() )

-- mock questions and answers
inputs = torch.Tensor{{1, 2, 3}, {2, 3, 1}}
targets = torch.Tensor{3, 2}

-- training
outputs = mlp:forward(inputs)
err = criterion:forward(outputs, targets)
gradOutputs = criterion:backward(outputs, targets)
gradInputs = mlp:backward(inputs, gradOutputs)

print('write out')
print(mlp:get(1).weight)

torch.save('z1.t7', mlp:get(1).weight )

old = torch.load('z1.t7')
newLookup = nn.LookupTable(3, 4)
newLookup.weight:copy(old)
print('after copy', newLookup.weight)