ディープラーニング勉強会 AutoEncoder ref: http://qiita.com/piyo7/items/60576759430910ffe5be

file0.scala

import scala.math
import scala.util.Random
import MatrixImplicits._ // 自作のimplicit classによるSeqラッパ

class dA(train_N: Int, n_visible: Int, n_hidden: Int, seed: Int) {
  val rng = new Random(seed)

  var W = Seq.fill(n_hidden, n_visible)(uniform(-1.0 / n_visible, 1.0 / n_visible))
  var hbias = Seq.fill(n_hidden)(0.0)
  var vbias = Seq.fill(n_visible)(0.0)

  def uniform(min: Double, max: Double): Double = rng.nextDouble() * (max - min) + min
  def sigmoid(x: Double): Double = 1.0 / (1.0 + math.exp(-x))
  def corrupted(x: Seq[Double], p: Double): Seq[Double] = x.map(_ * (if (rng.nextDouble() < p) 0.0 else 1.0))

  def encode(x: Seq[Double]): Seq[Double] = ((W mXc x) + hbias).map(sigmoid) 
  def decode(y: Seq[Double]): Seq[Double] = ((W.T mXc y) + vbias).map(sigmoid)

  def train(x: Seq[Double], learning_rate: Double, corruption_level: Double) {
    val tilde_x = corrupted(x, corruption_level)
    val y = encode(tilde_x)
    val z = decode(y)

    val L_vbias = x - z
    val L_hbias = (W mXc L_vbias) * y * y.map(1.0 - _)

    vbias = vbias + L_vbias.map(_ * learning_rate / train_N)
    hbias = hbias + L_hbias.map(_ * learning_rate / train_N)
    W = W + ((L_hbias cXr tilde_x) + (y cXr L_vbias)).map2(_ * learning_rate / train_N)
  }
}