Main.hs

module Main where

import qualified Numeric.LinearAlgebra as LA
import qualified MachineLearning as ML
import qualified MachineLearning.Optimization as Opt
import qualified MachineLearning.NeuralNetwork as NN
import qualified MachineLearning.NeuralNetwork.TopologyMaker as TM
import qualified MachineLearning.TerminalProgress as TP


main = do
  putStrLn "\n== Neural Networks (Digits Recognition) ==\n"

  -- Step 1. Data loading.
  -- Step 1.1 Training Data loading.
  (x, y) <- pure ML.splitToXY <*> LA.loadMatrix "digits_classification/optdigits.tra"
  -- Step 1.1 Testing Data loading.
  (xTest, yTest) <- pure ML.splitToXY <*> LA.loadMatrix "digits_classification/optdigits.tes"

  -- Step 2. Initialize Neural Network.
  let nnt = TM.makeTopology TM.ARelu TM.LSoftmax (LA.cols x) 10 [100, 100]
      model = NN.NeuralNetwork nnt
  -- Step 3. Initialize theta with randon values.
      initTheta = NN.initializeTheta 5191711 nnt

      lambda = NN.L2 $ 5 / (fromIntegral $ LA.rows x)

  -- Step 4. Train the Neural Network.
  (thetaNN, optPath) <- TP.learnWithProgressBar (Opt.minimize (Opt.BFGS2 0.03 0.7) model 1e-7 5 lambda x y) initTheta 20

  -- Step 5. Making predictions and checking accuracy on training and test sets.
  let accuracyTrain = NN.calcAccuracy y (NN.hypothesis model x thetaNN)
      accuracyTest = NN.calcAccuracy yTest (NN.hypothesis model xTest thetaNN)

  -- Step 6. Printing results.

  putStrLn $ "\nNumber of iterations to learn the Neural Network: " ++ show (LA.rows optPath)

  putStrLn $ "\nAccuracy on train set (%): " ++ show (accuracyTrain*100)
  putStrLn $ "Accuracy on test set (%): " ++ show (accuracyTest*100)