Xor.lua

require "nn" require 'optim' mlp = nn.Sequential(); -- make a multi-layer perceptron inputs = 2; outputs = 1; HUs = 20; -- parameters mlp:add(nn.Linear(inputs, HUs)) mlp:add(nn.Sigmoid()) mlp:add(nn.Linear(HUs, outputs)) mlp:add(nn.Sigmoid()) criterion = nn.BCECriterion() print(mlp) batchSize = 128 batchInputs = torch.DoubleTensor(batchSize, inputs) -- or CudaTensor for GPU training batchLabels = torch.DoubleTensor(batchSize) -- or CudaTensor for GPU training for i = 1, batchSize do local input = torch.randn(2) -- normally distributed example in 2d local label for i = 1, 2 do input[i] = (input[i] > 0 and 1 or 0) end label = bit.bxor(input[1], input[2]) batchInputs[i]:copy(input) batchLabels[i] = label end params, gradParams = mlp:getParameters() local optimState = {learningRate = 0.01} for epoch = 1, 500 do function feval(params) gradParams:zero() local outputs = mlp:forward(batchInputs) local loss = criterion:forward(outputs, batchLabels) local dloss_doutputs = criterion:backward(outputs, batchLabels) mlp:backward(batchInputs, dloss_doutputs) return loss, gradParams end optim.adam(feval, params, optimState) end print("\n---Tests---\n") x = torch.Tensor(2) x[1] = 1; x[2] = 1; print("Input: ", x[1], ", ", x[2]); print("Output: ", mlp:forward(x)) x[1] = 1; x[2] = 0; print("Input: ", x[1], ", ", x[2]); print("Output: ", mlp:forward(x)) x[1] = 0; x[2] = 1; print("Input: ", x[1], ", ", x[2]); print("Output: ", mlp:forward(x)) x[1] = 0; x[2] = 0; print("Input: ", x[1], ", ", x[2]); print("Output: ", mlp:forward(x))