Add files via upload

XOR.txt

@@ -0,0 +1,4 @@
+0 0 0
+0 1 1
+1 0 1
+1 1 0

annxor.py

@@ -0,0 +1,139 @@
+# the structure of neural network: 
+#    input layer with 2 inputs
+#    1 hidden layer with 2 units, tanh()
+#    output layer with 1 unit, sigmoid()
+
+import numpy as np
+import scipy
+from scipy.special import expit
+import math
+def run():
+  X = loadData("XOR.txt")
+  W = paraIni()
+  #print(X.shape)
+  intermRslt = feedforward(X,W)
+  #print(intermRslt[2])
+  Y = X[:, len(X[0])-1:len(X[0])]
+  #print(Y)
+  Yhat = intermRslt[2]
+  #print(Yhat)
+  #print(errCompute(Y, Yhat))
+  B=backpropagate(X, W, intermRslt, 0.5)
+  #print(B)
+
+  numIter = [100, 1000, 5000, 10000]
+  alp = [0.01, 0.5]
+  for i in range(len(numIter)):
+    for j in range(len(alp)):
+      R=FFMain("XOR.txt", numIter[i], alp[j])
+      np.savetxt('Error(numIter=' + repr(numIter[i]) + ',alp='+ repr(alp[j])+ ')' + '.txt', R[0], fmt="%.8f")
+      np.savetxt('Output(numIter=' + repr(numIter[i]) + ',alp='+ repr(alp[j])+ ')' + '.txt', R[1], fmt="%.8f")
+      np.savetxt('NewHidden(numIter=' + repr(numIter[i]) + ',alp='+ repr(alp[j])+ ')' + '.txt', R[2][0], fmt="%.8f")
+      np.savetxt('NewOutput(numIter=' + repr(numIter[i]) + ',alp='+ repr(alp[j])+ ')' + '.txt', R[2][1], fmt="%.8f")
+  #R=FFMain("XOR.txt", 10000, 0.5)
+  #print("R",R[1])
+
+def loadData(Filename):
+  X=[]
+  count = 0
+
+  text_file = open(Filename, "r")
+  lines = text_file.readlines()
+
+  for line in lines:
+    X.append([])
+    words = line.split(' ')
+    #print(words)
+    # convert value of first attribute into float  
+    for word in words:
+      #print(word)
+      X[count].append(float(word))
+    count += 1
+  X = np.asarray(X)
+  n,m = X.shape # for generality
+  X0 = np.ones((n,1))
+  X = np.hstack((X0,X))
+  return X
+
+def paraIni():
+  #code for fixed network and initial values
+  # parameters for hidden layer, 3 by 3 
+  #wh=np.array([[0.1859,-0.7706,0.6257],[-0.7984,0.5607,0.2109]])
+  #wh=np.random.random_sample((2,3))
+  wh=np.random.uniform(low = -1.0001, high=1.0001, size=(2,3))
+  wh[wh>1.0] = 1.0
+  wh[wh<-1.0] = -1.0
+
+  # parameter for output layer 1 by 3
+  #wo=np.array([[0.1328,0.5951,0.3433]])
+  #wo=np.random.random_sample((1,3))
+  wo=np.random.uniform(low = -1.0001, high=1.0001, size=(1,3))
+  wo[wo>1.0] = 1.0
+  wo[wo<-1.0] = -1.0
+
+  return [wh,wo]
+
+def feedforward(X,paras):
+  tempX = X[:, 0:len(X[0])-1] #x,y
+  tempY = X[:, len(X[0])-1:len(X[0])]
+  oh = np.tanh(np.dot(paras[0], tempX.transpose()))
+  n,m = oh.shape # for generality
+  X0 = np.ones((m,1))
+  ino = np.vstack((X0.transpose(),oh))
+  oo = expit(np.dot(paras[1],ino))
+  return [oh,ino,oo]
+
+def errCompute(Y,Yhat):
+  #this will not have the output from the original array
+  sum = 0
+  Yo = Yhat[0]
+
+  for k in range(len(Y)):
+    sum += pow((Y[k] - Yo[k]),2)
+  #sum all values & find error value
+  J = sum / (2 * len(Yo))
+
+  return J
+
+def backpropagate(X,paras,intermRslt,alpha):
+  #Initializing
+  Y = X[:, len(X[0])-1:len(X[0])]
+  tempX = X[:, 0:len(X[0])-1]
+  oo = intermRslt[2][0]
+  ino = intermRslt[1]
+  oh = intermRslt[0]
+  wh = paras[0]
+  wo = paras[1]
+
+  delta = np.multiply(np.multiply((Y.transpose() - oo), oo), (1.0-oo))
+  wo = wo + (alpha * np.dot(delta,ino.transpose()))/4.0
+  wop = wo[:, 1:len(wo[0])]
+  dot =np.dot(wop.transpose(), delta)
+
+  deltah = np.multiply(dot, (1.0-oh*oh))
+  wh = wh + alpha * np.dot(deltah, tempX) / 4.0
+
+  return [wh,wo]
+
+def FFMain(filename,numIteration, alpha):
+  #data load
+  X = loadData(filename)
+  #
+  W = paraIni()
+
+  #number of features
+  n = X.shape[1]
+
+  #error
+  errHistory = np.zeros((numIteration,1))
+
+  for i in range(numIteration):
+    #feedforward
+    intermRslt=feedforward(X,W)
+    #Cost function
+    errHistory[i,0]=errCompute(X[:,n-1:n],intermRslt[2])
+    #backpropagate
+    W=backpropagate(X,W,intermRslt,alpha)
+
+  Yhat=np.around(intermRslt[2]) 
+  return [errHistory,intermRslt[2],W]