completed first impl of the model

model.py

@@ -4,6 +4,9 @@ import os
 import csv
 from random import random
 from pprint import pprint as pp
+from math import log, exp
+from functools import reduce
+import numpy as np
 
 params = {"layers":[57, int(57/2), 1], "test":"TestX.csv", "train":"Train.csv"}
 
@@ -26,36 +29,72 @@ def getData(srcF, isTrain=True, addBias=True):
 
     return (X,y)
 
-def getMatrix(r, c, vec):
-    mat = []
-    for i in range(r):
-        temp = []
-        for j in range(c):
-            idx = i*(c)+j
-            temp.append(vec[j])
-        mat.append(temp)
+def sigmoid(v):
+    return 1.0/(1+exp(-v))
 
-    return mat
+def sigmoidGradient(v):
+    return [a*b for a,b in zip([sigmoid(x) for x in v], [sigmoid(x) for x in v])]
 
+def regularization(cost, w1, w2, lamb, m):
+    reg = sum(w1*w1)+sum(w2*w2)
+    reg = reg*lamb/(2*m)
+    return cost+reg
 
+def gradient(del1, del2, w1, w2, lamb, m):
+    del1,del2 = del1/m, del2/m
+    tgrad1 = del1[:,1:] + lamb*w1[:,1:]/m
+    tgrad2 = del2[:,1:] + lamb*w2[:,1:]/m
+    return np.append(tgrad1.reshape(tgrad1.size), tgrad2.reshape(tgrad2.size))
 
-def cost(li, lh, lo, weights, X, y):
+def cost(li, lh, lo, weights, X, y, lamb):
     w1t = weights[:(li+1)*lh]    #28x58
     w2t = weights[(li+1)*lh:]    #1x29
-
-    w1,w2 = getMatrix(lh, li+1, w1t), getMatrix(lo, lh+1, w2t)   
-
+    #28x58, 1x29
+    tdel1,tdel2 = np.zeros((lh,li+1),dtype=float), np.zeros((lo,lh+1),dtype=float)
+
+    #w1,w2 = getMatrix(lh, li+1, w1t), getMatrix(lo, lh+1, w2t)
+    w1 = np.array(w1t)
+    w2 = np.array(w2t)
+    #cost
+    m = len(X)
+    J = 0.0
+    for i in range(m):
+        a1 = X[i]   #1x58
+        z2 = np.dot(w1.reshape((lh,li+1)), a1) #28x58 * 58x1 = 28x1        
+        a2 = [1.0]  #bias for hidden
+        a2.extend([sigmoid(x) for x  in z2])    #29x1
+        z3 = np.dot(w2.reshape((lo,lh+1)), a2) #1x29 *29x1 = 1x1
+        z3 = z3[0]
+        h = sigmoid(z3)
+        # J += -y[i]*log(fw) - (1-y[i])*log(1-fw)
+        #backpropagation
+        del3 = h-y[i]   #1x1
+        z2 = [1]+z2.tolist()      #29x1
+        del2 = w2.reshape((lo,lh+1))*del3*np.array(sigmoidGradient(z2)).reshape((1,29))  #1x29 x 1x1 x 1x29 = 1x29
+
+        del2 = del2.reshape(lh+1,lo).tolist()
+        del(del2[0])        #28x1
+        tdel1 = np.dot(np.array(del2).reshape(lh,lo),np.array(a1).reshape(lo,li+1))+tdel1 #28x1 x 1x58
+        tdel2 = np.multiply(np.array(a2).reshape((lo,lh+1)), del3)+tdel2   #1x29 * 1x1 + 1x29 = 1x29
+        J += y[i]*log(h) + (1-y[i])*log(1-h)
+
+    J = -J/m
+
+    #regularization
+    J = regularization(J, w1, w2, lamb, m)
+
+    #calculate gradient
+    grad = gradient(tdel1, tdel2, w1.reshape((lh,li+1)), w2.reshape((lo,lh+1)), lamb, m)    
+
+    return J,grad
     
-
-
-
-
-
 if __name__ == "__main__":
     X,y = getData(params["train"])
     # tX,ty = getData(params["test"], isTrain=False)
-    print(len(X), len(X[0]), len(y), X[0])
+    # print(len(X), len(X[0]), len(y), X[0])
+    # print(len(tX), len(ty), tX[0])
     li,lh,lo = tuple(params["layers"])
     weights = [random() for _ in range(lh*(li+1)+lo*(lh+1))]
-    w1,w2 = cost(li, lh, lo, weights, X, y)
+    J,grad = cost(li, lh, lo, weights, X, y, 0.1)
+    print(J,grad)
     #print(len(w1), len(w1[0]), len(w2), len(w2[0]))
\ No newline at end of file