15번째 수업(0502)

멀티 레이어 퍼셉트론

import numpy as np
import math as m


def sigmoid(x):
    return (1.0 / (1 + m.exp(-x)))


lrate = 0.02
INDIM = 7 #(바이어스까지 해서 7개)
H1DIM = 5
OUTDIM = 3

PTTN_NUM = 3 #(학습시키는 개수)

x = np.array([[1.0, 1.0, 1.0, 1.0, 0.1, 0.1, 0.1],  # T-1    111000
              [1.0, 0.1, 0.1, 1.0, 1.0, 1.0, 0.1],  # T-2    000110
              [1.0, 0.1, 0.1, 0.1, 1.0, 1.0, 1.0],  # T-3    000111
            ])

t = np.array([[1.0, 0.0, 0.0],  #100
              [0.0, 1.0, 0.0],  #010
              [0.0, 0.0, 1.0],  #001
            ])

#test_pattern
xt = np.array([[1.0, 0.1, 1.0, 0.1, 1.0, 0.1, 1.0],  # T-1  010101
              [1.0, 1.0, 0.1, 1.0, 0.1, 1.0, 0.1],  # T-2   101010
            ])

w1 = np.zeros([H1DIM, INDIM])
for i in range(H1DIM):
    for j in range(INDIM):
        w1[i][j] = np.random.rand() / 10.0

w2 = np.zeros([OUTDIM, H1DIM])
for i in range(OUTDIM):
    for j in range(H1DIM):
        w2[i][j] = np.random.rand() / 10.0

print(w1)
print(w2)

y1 = np.zeros(H1DIM)
y2 = np.zeros(OUTDIM)

d1 = np.zeros(H1DIM)
d2 = np.zeros(OUTDIM)

for epoch in range(100000):
    if (epoch % 1000) == 0:
        print("epoch", epoch)
    for p in range(PTTN_NUM):
        ###########################################
        # feed forwarding
        ###########################################
        # LAYER-1 (Hidden Layer)
        for i in range(H1DIM):
            out = 0.0
            for j in range(INDIM):
                out += w1[i][j] * x[p][j]
            y1[i] = sigmoid(out)
        # LAYER-2 (Output Layer)
        # y1[0] = 1.0  ### bias

        for i in range(OUTDIM):
            out = 0.0
            for j in range(H1DIM):
                out += w2[i][j] * y1[j]
            y2[i] = sigmoid(out)

        ###########################################
        # Back Propagation
        ###########################################
        # delta(error)  for Layer-2 (Output Layer)
        for i in range(OUTDIM):
            d2[i] = (t[p][i] - y2[i]) * (y2[i] * (1 - y2[i]))
        # delta(error)  for Layer-1 (Hidden Layer)
        for i in range(H1DIM):
            for j in range(OUTDIM):
                d1[i] = d2[j] * w2[j][i] * (y1[i] * (1 - y1[i]))

        ###########################################
        # Weight Adjustment
        ###########################################
        # for Layer-2 (Output Layer)
        for i in range(OUTDIM):
            for j in range(H1DIM):
                w2[i][j] += lrate * d2[i] * y1[j]

        # for Layer-1 (Hidden Layer)
        for i in range(H1DIM):
            for j in range(INDIM):
                w1[i][j] += lrate * d1[i] * x[p][j]

        if (epoch % 1000) == 0:
            print("pattern number[%d] : " % p)
            print("target : ", "%7.2f" % t[p][0], "%7.2f" % t[p][1], "%7.2f" % t[p][2]) #타겟
            print("output : ", "%7.2f" % y2[0], "%7.2f" % y2[1], "%7.2f" % y2[2])       #출력

#########################################################
# Evaluation

PLOT_SIZE = 21
px = np.zeros(PLOT_SIZE)
py = np.zeros(PLOT_SIZE)
for k in range(PLOT_SIZE):
    # feed forwarding
    # LAYER-1 (Hidden Layer)
    x[0][1] = k / (PLOT_SIZE - 1)
    for i in range(H1DIM):
        out = 0.0
        for j in range(INDIM):
            out += w1[i][j] * x[0][j]
            y1[i] = sigmoid(out)
    # LAYER-2 (Output Layer)
    out = 0.0
    for i in range(H1DIM):
        out += w2[i] * y1[i]
    y2 = sigmoid(out)

    print("%7.2f" % x[0][1], "%7.2f" % y2)
    px[k] = x[0][1]
    py[k] = y2

 

다음주까지 중간고사 5번 문제 t, c, e 로 해오기

 

                        target

T 25개 비트    1 0 0

C 25개 비트    0 1 0

E 25개 비트    0 0 1