Least Square Discriminant

Author: Lewuathe

.cache

@@ -10,10 +10,29 @@ object App {
     var classData = ClassReader.trainRead("./data/classes.csv")
     var testData = ClassReader.testRead("./data/test_cat.csv")
 
+/*
+    val ld = new LeastSquareDiscriminant(2, 2)
+    val W = ld.train(classData)
+    println(W)
+    val a = - (W(0,1) - W(1,1)) / (W(0,2) - W(1,2))
+    val b = - (W(0,0) - W(1,0)) / (W(0,2) - W(1,2))
+    var estimate_cat = new PrintWriter("./data/estimate_cat.csv")
+    testData.foreach {
+      x => {
+    	  val y = a * x + b
+    	  estimate_cat.println(x + "," + y)
+      }
+    }
+    estimate_cat.flush()
+    estimate_cat.close()
+*/
+    
+    
+
+     //* For Perceptron discriminant
     val ld = new LinearDiscriminant(2)
     val w = ld.train(classData)
     println(w)
-    
     val a = -w(1)/w(2)
     val b = -w(0)/w(2)
     var estimate_cat = new PrintWriter("./data/estimate_cat.csv")

src/main/scala/App.scala

@@ -0,0 +1,35 @@
+package com.PhysicsEngine.cpistats
+import breeze.linalg._
+
+class LeastSquareDiscriminant(D: Int, K: Int) {
+	private var _w = DenseVector.ones[Double](D+1)
+
+	def train(plist: List[DenseVector[Double]]): DenseMatrix[Double] = {
+	  val n = plist.length
+	  val T = DenseMatrix.zeros[Double](n, K)
+	  val X = DenseMatrix.zeros[Double](n, D+1)
+	  
+	  for (i <- 0 until n) {
+	    val x = plist(i)
+	    X(i, 0) = 1.0
+	    for (j <- 1 until D+1) {
+	    	X(i, j) = x(j-1)
+	    }
+	  }
+	  
+	  for (i <- 0 until n) {
+	    val t = plist(i)
+	    if (t(2) > 0.0) {
+	      T(i, 0) = 1
+	      T(i, 1) = 0
+	    } else {
+	      T(i, 0) = 0
+	      T(i, 1) = 1
+	    }
+	  }
+
+	  //println(inv(X.t * X) * X.t * T)
+	  val W = inv(X.t * X) * X.t * T
+	  W.t
+	}
+}

src/main/scala/LeastSquareDiscriminant.scala

@@ -4,7 +4,6 @@ import java.io._
 
 class LinearDiscriminant(val D: Int) {
 	private var _w = DenseVector.ones[Double](D+1)
-	_w = _w * -5.0
 	private var _eta = 0.5
 	def train(plist: List[DenseVector[Double]]): DenseVector[Double] = {
 	  var correct = 0
@@ -13,17 +12,6 @@ class LinearDiscriminant(val D: Int) {
 	    step += 1
 	    correct = 0
 
-	    /*
-	    val index = Math.floor(Math.random()*plist.length).toInt
-	    val point = plist(index)
-	    val phi = DenseVector(1.0, point(0), point(1))
-        if ((_w.t * phi).apply(0) * point(2) < 0) {
-     	   _w += phi * point(2) * _eta
-	    }
-	    * 
-	    */
-
-
 	    plist.foreach {
 	      point => {
 	       val phi = DenseVector(1.0, point(0), point(1))
@@ -41,22 +29,8 @@ class LinearDiscriminant(val D: Int) {
 	       }
 	      }
 	    }
-	    /*
-	    plist.foreach {
-	      point => {
-	        val phi = DenseVector(1, point(0), point(1))
-	          if ((_w.t * phi).apply(0) * point(2) < 0) {
-	        	_w += phi * point(2) * _eta
-	          }
-	          else {
-	        	correct += 1
-	          }
-	      }
-	    }
-	    * 
-	    */
 	  }
-	  println(step)
+	  println("Step: " + step)
 	  _w
 	}
 }

src/main/scala/LinearDiscriminant.scala

@@ -6,18 +6,21 @@
 from pylab import *
 
 K = 2
-N = 200
+N = 50
 
 if __name__ == "__main__":
     cls1 = []
     cls2 = []
 
-    mean1 = [-0.5, 0.7]  
-    mean2 = [0.8, -0.8]  
-    cov = [[0.2,0.1], [0.1,0.2]]  
+    
+    mean1 = [-1.2, 1.4]  
+    mean2 = [-0.5, 0.5]  
+    mean3 = [1.6, -1.4]
+    cov = [[0.2,0.1], [0.1,0.1]]  
 
     cls1.extend(np.random.multivariate_normal(mean1, cov, N/2))
     cls2.extend(np.random.multivariate_normal(mean2, cov, N/2))
+    cls2.extend(np.random.multivariate_normal(mean3, cov, N/2))
 
 
     f = open("/Users/sasakiumi/MyWorks/cpi-stats/data/classes.csv", "w")