아래는 이 집구석 주인장이 "딥러닝 실패사례 : 말 해 줄거라 생각했다" 의 제목으로 작성한 글에 실린 예제 코드입니다.
"밑바닥부터 시작하는 딥러닝" 서적에 실린 코드를 기반으로 작성하였습니다.
# coding: utf-8
import numpy as np
import matplotlib.pyplot as plt
from two_layer_net import TwoLayerNet
import re
def deepLearn(gun_scores, naver_scores, train_size, batch_size):
x_train = naver_scores[:train_size]
t_train = gun_scores[:train_size]
x_test = naver_scores[train_size:]
t_test = gun_scores[train_size:]
network = TwoLayerNet(input_size=2, hidden_size=10, output_size=5)
# 하이퍼파라미터
iters_num = 10000 # 반복 횟수를 적절히 설정한다.
train_size = x_train.shape[0]
learning_rate = 0.1
train_loss_list = []
train_acc_list = []
test_acc_list = []
# 1에폭당 반복 수
iter_per_epoch = max(train_size / batch_size, 1)
for i in range(iters_num):
# 미니배치 획득
batch_mask = np.random.choice(train_size, batch_size)
x_batch = x_train[batch_mask]
t_batch = t_train[batch_mask]
# 기울기 계산
#grad = network.numerical_gradient(x_batch, t_batch)
grad = network.gradient(x_batch, t_batch)
# 매개변수 갱신
for key in ('W1', 'b1', 'W2', 'b2'):
network.params[key] -= learning_rate * grad[key]
# 학습 경과 기록
loss = network.loss(x_batch, t_batch)
train_loss_list.append(loss)
# 1에폭당 정확도 계산
if i % iter_per_epoch == 0:
train_acc = network.accuracy(x_train, t_train)
test_acc = network.accuracy(x_test, t_test)
train_acc_list.append(train_acc)
test_acc_list.append(test_acc)
print("train acc, test acc | " + str(train_acc) + ", " + str(test_acc))
x = np.arange(len(train_acc_list))
plt.plot(x, train_acc_list, label='train acc')
plt.plot(x, test_acc_list, label='test acc', linestyle='--')
plt.xlabel("epochs")
plt.ylabel("accuracy")
plt.ylim(0, 1.0)
plt.legend(loc='lower right')
plt.show()
#
def readScore():
gun_score = []
naver_score = []
f = open("gunman_scores.txt", 'r')
while True:
line = f.readline()
if not line: break
scores = re.split("::", line)
myscore = int(scores[1])
netiscore = float(scores[2])
spscore = float(scores[3])
naver_score.append([netiscore, spscore])
scoreboard = [0, 0, 0, 0, 0]
scoreboard[myscore - 1] = 1
gun_score.append(scoreboard)
f.close()
naver_score = np.array(naver_score)
gun_score = np.array(gun_score)
return naver_score, gun_score
naver_score, gun_score = readScore()
train_size = 600
batch_size = 100
deepLearn(gun_score, naver_score, train_size, batch_size)
그리고 아래 코드는 상기 코드를 텐서 플로우를 사용하여 구현한 것입니다.
"텐서 플로 첫걸음" 서적에 실린 코드를 참고하여 작성했습니다.
# -*- coding: utf-8 -*-
import tensorflow as tf
import numpy as np
import re
import matplotlib.pyplot as plt
def deepLearn(gun_scores, naver_scores, train_size, batch_size):
x_train = naver_scores[:train_size]
t_train = gun_scores[:train_size]
x_test = naver_scores[train_size:]
t_test = gun_scores[train_size:]
W1 = tf.Variable(tf.random_uniform([2, 10], -1.0, 1.0))
#W2 = tf.Variable(tf.random_uniform([10, 10], -1.0, 1.0))
#W3 = tf.Variable(tf.random_uniform([10, 10], -1.0, 1.0))
#W4 = tf.Variable(tf.random_uniform([10, 10], -1.0, 1.0))
#W5 = tf.Variable(tf.random_uniform([10, 10], -1.0, 1.0))
W6 = tf.Variable(tf.random_uniform([10, 5], -1.0, 1.0))
b1 = tf.Variable(tf.zeros([10]))
#b2 = tf.Variable(tf.zeros([10]))
#b3 = tf.Variable(tf.zeros([10]))
#b4 = tf.Variable(tf.zeros([10]))
#b5 = tf.Variable(tf.zeros([4]))
b6 = tf.Variable(tf.zeros([5]))
x = tf.placeholder("float", [None, 2])
L2 = tf.sigmoid(tf.matmul(x, W1) + b1)
#L3 = tf.sigmoid(tf.matmul(L2, W2) + b2)
#L4 = tf.sigmoid(tf.matmul(L3, W3) + b3)
#L5 = tf.sigmoid(tf.matmul(L4, W4) + b4)
#L6 = tf.sigmoid(tf.matmul(L5, W5) + b5)
hypothesis = tf.nn.softmax(tf.matmul(L2, W6) + b6)
y_ = tf.placeholder("float", [None, 5])
cross_entropy = -tf.reduce_sum(y_ * tf.log(hypothesis))
train_step = tf.train.GradientDescentOptimizer(0.01).minimize(cross_entropy)
sess = tf.Session()
sess.run(tf.global_variables_initializer())
train_acc_list = []
test_acc_list = []
for i in range(1000):
batch_mask = np.random.choice(train_size, batch_size)
x_batch = x_train[batch_mask]
y_batch = t_train[batch_mask]
sess.run(train_step, feed_dict={x: x_batch, y_: y_batch})
correct_prediction = tf.equal(tf.argmax(hypothesis,1), tf.argmax(y_,1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
#if i % 100 == 0:
acc_train = sess.run(accuracy, feed_dict={x: x_batch, y_: y_batch})
acc_test = sess.run(accuracy, feed_dict={x: x_test, y_: t_test})
train_acc_list.append(acc_train)
test_acc_list.append(acc_test)
x = np.arange(len(train_acc_list))
plt.plot(x, train_acc_list, label='train acc')
plt.plot(x, test_acc_list, label='test acc', linestyle='--')
plt.xlabel("epochs")
plt.ylabel("accuracy")
plt.ylim(0, 1.0)
plt.legend(loc='lower right')
plt.show()
#
def readScore():
gun_score = []
naver_score = []
f = open("gunman_scores.txt", 'r')
while True:
line = f.readline()
if not line: break
scores = re.split("::", line)
myscore = int(scores[1])
netiscore = float(scores[2])
spscore = float(scores[3])
naver_score.append([netiscore, spscore])
scoreboard = [0, 0, 0, 0, 0]
scoreboard[myscore-1] = 1
gun_score.append(scoreboard)
f.close()
naver_score = np.array(naver_score)
gun_score = np.array(gun_score)
return naver_score, gun_score
train_size = 600
batch_size = 100
naver_score, gun_score = readScore()
deepLearn(gun_score, naver_score, train_size, batch_size)