hongshaorou/Tianshou
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

minor fixed

Browse Source
This commit is contained in:
Tongzheng Ren 2017-11-10 14:40:23 +08:00
parent 1d7d4e14ef
commit ab1b3775e7
2 changed files with 32 additions and 23 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
AlphaGo/Network.py
View File

@ -12,7 +12,7 @@ os.environ["CUDA_VISIBLE_DEVICES"] = "1"
def residual_block(input, is_training): def residual_block(input, is_training):
normalizer_params = {'is_training': is_training, normalizer_params = {'is_training': is_training,
'updates_collections': None} 'updates_collections': tf.GraphKeys.UPDATE_OPS}
h = layers.conv2d(input, 256, kernel_size=3, stride=1, activation_fn=tf.nn.relu, h = layers.conv2d(input, 256, kernel_size=3, stride=1, activation_fn=tf.nn.relu,
normalizer_fn=layers.batch_norm, normalizer_params=normalizer_params, normalizer_fn=layers.batch_norm, normalizer_params=normalizer_params,
weights_regularizer=layers.l2_regularizer(1e-4)) weights_regularizer=layers.l2_regularizer(1e-4))
@ -25,7 +25,7 @@ def residual_block(input, is_training):
def policy_heads(input, is_training): def policy_heads(input, is_training):
normalizer_params = {'is_training': is_training, normalizer_params = {'is_training': is_training,
'updates_collections': None} 'updates_collections': tf.GraphKeys.UPDATE_OPS}
h = layers.conv2d(input, 2, kernel_size=1, stride=1, activation_fn=tf.nn.relu, h = layers.conv2d(input, 2, kernel_size=1, stride=1, activation_fn=tf.nn.relu,
normalizer_fn=layers.batch_norm, normalizer_params=normalizer_params, normalizer_fn=layers.batch_norm, normalizer_params=normalizer_params,
weights_regularizer=layers.l2_regularizer(1e-4)) weights_regularizer=layers.l2_regularizer(1e-4))
@ -36,7 +36,7 @@ def policy_heads(input, is_training):
def value_heads(input, is_training): def value_heads(input, is_training):
normalizer_params = {'is_training': is_training, normalizer_params = {'is_training': is_training,
'updates_collections': None} 'updates_collections': tf.GraphKeys.UPDATE_OPS}
h = layers.conv2d(input, 2, kernel_size=1, stride=1, activation_fn=tf.nn.relu, h = layers.conv2d(input, 2, kernel_size=1, stride=1, activation_fn=tf.nn.relu,
normalizer_fn=layers.batch_norm, normalizer_params=normalizer_params, normalizer_fn=layers.batch_norm, normalizer_params=normalizer_params,
weights_regularizer=layers.l2_regularizer(1e-4)) weights_regularizer=layers.l2_regularizer(1e-4))
@ -52,7 +52,7 @@ z = tf.placeholder(tf.float32, shape=[None, 1])
pi = tf.placeholder(tf.float32, shape=[None, 362]) pi = tf.placeholder(tf.float32, shape=[None, 362])
h = layers.conv2d(x, 256, kernel_size=3, stride=1, activation_fn=tf.nn.relu, normalizer_fn=layers.batch_norm, h = layers.conv2d(x, 256, kernel_size=3, stride=1, activation_fn=tf.nn.relu, normalizer_fn=layers.batch_norm,
normalizer_params={'is_training': is_training, 'updates_collections': None}, normalizer_params={'is_training': is_training, 'updates_collections': tf.GraphKeys.UPDATE_OPS},
weights_regularizer=layers.l2_regularizer(1e-4)) weights_regularizer=layers.l2_regularizer(1e-4))
for i in range(19): for i in range(19):
h = residual_block(h, is_training) h = residual_block(h, is_training)

47
utils/text2data.py
View File

@ -1,5 +1,6 @@
import numpy as np import numpy as np
import os import os
import time
def hex2board(hex): def hex2board(hex):
scale = 16 scale = 16
@ -20,7 +21,7 @@ def str2prob(str):
if np.sum(np.isnan(prob))==0: if np.sum(np.isnan(prob))==0:
return prob, True return prob, True
else: else:
return 0, False return 0, False
dir = "/home/yama/leela-zero/data/sgf-txt-files/" dir = "/home/yama/leela-zero/data/sgf-txt-files/"
name = os.listdir(dir) name = os.listdir(dir)
@ -29,37 +30,45 @@ for n in name:
if n[-4:]==".txt": if n[-4:]==".txt":
text.append(n) text.append(n)
print(text) print(text)
total_start = -time.time()
for t in text: for t in text:
start = -time.time()
num = 0 num = 0
boards = np.zeros([0, 19, 19, 17], dtype='int8') boards = []
board = np.zeros([1, 19, 19, 0], dtype='int8') board = []
win = np.zeros([0, 1], dtype='int8') win = []
p = np.zeros([0, 362]) p = []
flag = False flag = False
for line in open(dir + t): for line in open(dir + t):
if num % 19 == 0: if num % 19 == 0:
flag = False flag = False
if num % 19 < 16: if num % 19 < 16:
new_board = hex2board(line) new_board = hex2board(line)
board = np.concatenate([board, new_board], axis=3) board.append(new_board)
if num % 19 == 16: if num % 19 == 16:
if int(line) == 0: if int(line) == 0:
new_board = np.ones([1, 19 ,19 ,1], dtype='int8') new_board = np.ones([1, 19 ,19 ,1], dtype='int8')
if int(line) == 1: if int(line) == 1:
new_board = np.zeros([1, 19, 19, 1], dtype='int8') new_board = np.zeros([1, 19, 19, 1], dtype='int8')
board = np.concatenate([board, new_board], axis=3) board.append(new_board)
boards = np.concatenate([boards, board], axis=0) board = np.concatenate(board, axis=3)
board = np.zeros([1, 19, 19, 0], dtype='int8') boards.append(board)
board = []
if num % 19 == 17: if num % 19 == 17:
if str2prob(line)[1] == True: if str2prob(line)[1] == True:
p = np.concatenate([p,str2prob(line)[0]], axis=0) p.append(str2prob(line)[0])
else: else:
flag = True flag = True
boards = boards[:-1] boards = boards[:-1]
if num % 19 == 18: if num % 19 == 18:
if flag == False: if flag == False:
win = np.concatenate([win, np.array(float(line), dtype='int8').reshape(1,1)], axis=0) win.append(np.array(int(line), dtype='int8').reshape(1,1))
num=num+1 num=num+1
boards = np.concatenate(boards, axis=0)
win = np.concatenate(win, axis=0)
p = np.concatenate(p, axis=0)
print("Boards Shape: {}, Wins Shape: {}, Probs Shape : {}".format(boards.shape[0], win.shape[0], p.shape[0])) print("Boards Shape: {}, Wins Shape: {}, Probs Shape : {}".format(boards.shape[0], win.shape[0], p.shape[0]))
print "Finished " + t print("Finished {} Time {}".format(t, time.time()+start))
np.savez("/home/tongzheng/meta-data/"+t[:-4], boards=boards, win=win, p=p) np.savez("/home/tongzheng/meta-data/"+t[:-4], boards=boards, win=win, p=p)
del boards, board, win, p
print("All finished! Time {}".format(time.time()+total_start))