该楼层疑似违规已被系统折叠 隐藏此楼查看此楼
训练时长已经7个小时了,开始时还能明显观察到小鸟运动的变化是越来越好的,后来好像就停留在某个阶段了,过了几个小时还是那样,总是撞到上面的柱子,代码在下面,求大神点播,小弟感激不尽。
import tensorflow as tf
import cv2
import sys
sys.path.append("d:/study/Keras_FlappyBird-master/game")
import wrapped_flappy_bird as game
import numpy as np
t = 0
i = 1
def weight_variable(shape):
initial = tf.truncated_normal(shape, stddev=0.01)
return tf.Variable(initial)
def bias_variable(shape):
initial = tf.constant(0.01, shape=shape)
return tf.Variable(initial)
def conv2d(x, W, stride):
return tf.nn.conv2d(x, W, strides=[1, stride, stride, 1], padding="SAME")
def max_pool_2x2(x):
return tf.nn.max_pool(x, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding="SAME")
W_conv1 = weight_variable([8, 8, 4, 32])
b_conv1 = bias_variable([32])
W_conv2 = weight_variable([4, 4, 32, 64])
b_conv2 = bias_variable([64])
W_conv3 = weight_variable([3, 3, 64, 64])
b_conv3 = bias_variable([64])
W_fc1 = weight_variable([1600, 512])
b_fc1 = bias_variable([512])
W_fc2 = weight_variable([512, 2])
b_fc2 = bias_variable([2])
s = tf.placeholder("float", [1, 80, 80, 4])
h_conv1 = tf.nn.relu(conv2d(s, W_conv1, 4) + b_conv1)#出32,80*80
h_pool1 = max_pool_2x2(h_conv1)#出32,40*40
h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2, 2) + b_conv2)#出64,40*40
h_conv3 = tf.nn.relu(conv2d(h_conv2, W_conv3, 1) + b_conv3)#出64,40*40
h_conv3_flat = tf.reshape(h_conv3, [-1, 1600])
h_fc1 = tf.nn.relu(tf.matmul(h_conv3_flat, W_fc1) + b_fc1)#出512维,64个
readout = tf.matmul(h_fc1,W_fc2) + b_fc2#出2维,64个
a = tf.placeholder("float", [1, 2])#占位输入2维,1行,2列
y = tf.placeholder("float", [1])#输入一维,赋值r_t
readout_action = tf.reduce_sum(tf.multiply(readout, a))
#sum,得到数字一个
cost = tf.reduce_mean(tf.square(y - readout_action))
train_step = tf.train.AdamOptimizer(1e-6).minimize(cost)
game_state = game.GameState()
x_t, r_0, terminal = game_state.frame_step([1,0])
x_t = cv2.cvtColor(cv2.resize(x_t, (80, 80)), cv2.COLOR_BGR2GRAY)
ret, x_t = cv2.threshold(x_t, 1, 255, cv2.THRESH_BINARY)
s_t = np.stack((x_t, x_t, x_t, x_t), axis=2)
sess = tf.InteractiveSession()
sess.run(tf.global_variables_initializer())
while "flappy bird" != "angry bird":
readout_t = sess.run(readout,feed_dict={s:[s_t]})[0]#1行,2列
a_t = np.zeros([2])
action_index = np.argmax(readout_t)
a_t[action_index] = 1
x_t1_colored, r_t, terminal = game_state.frame_step(a_t)#输入a_t
#return image_data, reward, terminal
x_t1 = cv2.cvtColor(cv2.resize(x_t1_colored, (80, 80)), cv2.COLOR_BGR2GRAY)
ret, x_t1 = cv2.threshold(x_t1, 1, 255, cv2.THRESH_BINARY)
x_t1 = np.reshape(x_t1, (80, 80, 1))
s_t1 = np.append(x_t1, s_t[:, :, :3], axis=2)
train_step.run(feed_dict={y:[r_t],a:[a_t],s:[s_t]})
s_t = s_t1
t += 1
if t > 8000*i:
saver = tf.train.Saver()
saver.save(sess,'d:/study/dataset/flappy.ckpt', global_step=t)
i += 1
print('saved')
if t == 1:
saver = tf.train.Saver()
ckpt = tf.train.get_checkpoint_state('d:/study/dataset')
saver.restore(sess, ckpt.model_checkpoint_path)
print('loaded')