modified the network

AlphaGo/Network.py

@@ -4,13 +4,6 @@ import time
 import multi_gpu
 import tensorflow.contrib.layers as layers
 
-data = np.load("data.npz")
-boards = data["boards"]
-wins = data["wins"]
-ps = data["ps"]
-print (boards.shape)
-print (wins.shape)
-print (ps.shape)
 def residual_block(input, is_training):
 	normalizer_params = {'is_training': is_training,
 						 'updates_collections': None}
@@ -44,7 +37,6 @@ def value_heads(input, is_training):
 	h = layers.fully_connected(h, 1, activation_fn=tf.nn.tanh, weights_regularizer=layers.l2_regularizer(1e-4))
 	return h
 
-
 x = tf.placeholder(tf.float32,shape=[None,19,19,17])
 is_training = tf.placeholder(tf.bool, shape=[])
 z = tf.placeholder(tf.float32, shape=[None, 1])
@@ -62,6 +54,14 @@ train_op = tf.train.RMSPropOptimizer(1e-2).minimize(total_loss)
 
 var_list = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES)
 saver = tf.train.Saver(max_to_keep=10, var_list=var_list)
+def train():
+	data = np.load("data.npz")
+	boards = data["boards"]
+	wins = data["wins"]
+	ps = data["ps"]
+	print (boards.shape)
+	print (wins.shape)
+	print (ps.shape)
 	epochs = 100
 	batch_size = 32
 	batch_num = boards.shape[0] // batch_size
@@ -93,3 +93,15 @@ with multi_gpu.create_session() as sess:
 				if iter % 20 == 0:
 					save_path = "Epoch{}.Iteration{}.ckpt".format(epoch, iter)
 					saver.save(sess, result_path + save_path)
+
+def forward(board):
+	result_path = "./results/"
+	with multi_gpu.create_session() as sess:
+		sess.run(tf.global_variables_initializer())
+		ckpt_file = tf.train.latest_checkpoint(result_path)
+		if ckpt_file is not None:
+			print('Restoring model from {}...'.format(ckpt_file))
+			saver.restore(sess, ckpt_file)
+		else:
+			raise ValueError("No model loaded")
+		return sess.run([p,v], feed_dict={x:board})