train.lua

require 'rnn'

batchSize = 1000
rho = 999999
hiddenSize = 7
nIndex = 10
iters = 10
learningRate = 0.1
inputSize = 5
dataSize = 18563460

nn.FastLSTM.bn = true
rnn = nn.FastLSTM(inputSize, inputSize)
rnn = nn.Sequencer(rnn)

sequence = torch.Tensor(dataSize, inputSize)

criterion = nn.MSECriterion()
criterion = nn.SequencerCriterion(criterion)

print("Model :")
print(rnn)

offsets = torch.LongTensor(batchSize):random(1, dataSize)

-- training
minErr = inputSize -- report min error
minK = 0
avgErrs = torch.Tensor(iters):fill(0)
for k = 1, iters do
    -- 1. create a sequence of rho time-steps
    local inputs, targets = {}, {}
    for step=1,rho do
        -- batch of inputs
        inputs[step] = inputs[step] or sequence.new()
        inputs[step]:index(sequence, 1, offsets)
        -- batch of targets
        offsets:add(1) -- increase indices by 1
        offsets[offsets:gt(opt.dataSize)] = 1
        targets[step] = targets[step] or sequence.new()
        targets[step]:index(sequence, 1, offsets)
    end

    -- 2. forward sequence through rnn

    rnn:zeroGradParameters()

    local outputs = rnn:forward(inputs)
    local err = criterion:forward(outputs, targets)

    print('Iter: ' .. k .. '   Err: ' .. err)
    avgErrs[k] = err
    if avgErrs[k] < minErr then
        minErr = avgErrs[k]
        minK = k
    end

    -- 3. backward sequence through rnn (i.e. backprop through time)

    local gradOutputs = criterion:backward(outputs, targets)
    local gradInputs = rnn:backward(inputs, gradOutputs)

    -- 4. update

    rnn:updateParameters(learningRate)
end


print('min err: ' .. minErr .. ' on iteration ' .. minK)