3 layer perceptron for iris data by torch7. iris.csv can be obtained from https://gist.github.com/chiral/5b3609e884e3af8b412a

iris.lua

-- loarocks install csv
csv = require "csv"
nn = require "nn"

-- prepare data set

dataset={}
size=0

ans={setosa=1,versicolor=2,virginica=3}

local f = csv.open("iris.csv")
for cols in f:lines() do
  local input=torch.Tensor(4)
  local output=torch.Tensor(3)
  for i,v in ipairs(cols) do
    if i>=2 and i<=5 then
      input[i-1] = v
    end
    if i==6 then
      for j=1,3 do
        output[j] = j==ans[v] and 1 or 0
      end
    end
    print (input,output)
  end
  size=size+1
  dataset[size]={input,output}
end

function dataset:size() return size end


-- make model

nhidden = 5

model = nn.Sequential();
model:add(nn.Linear(4,nhidden))
model:add(nn.Tanh())
model:add(nn.Linear(nhidden,3))
model:add(nn.SoftMax())

print (model)

criterion = nn.MSECriterion()
trainer = nn.StochasticGradient(model, criterion)
trainer.learningRate = 0.01
trainer.maxIteration = 1000
trainer:train(dataset)

test1_in = torch.Tensor(4)
test1_in[1] = 6.7
test1_in[2] = 2.5
test1_in[3] = 5.8
test1_in[4] = 1.8
-- answer = 3

test1_out = model:forward(test1_in)
print (test1_out) -- OK