我想出一个相当令人困惑的问题，我要么非常盲目又错过了某些东西，要么我的重用变量同时设置为False和True。
这些是我的模型定义：
def discriminator(data, reuse=False):
if reuse:
tf.get_variable_scope().reuse_variables()
# Fully Connected 1
d_w1 = tf.get_variable('d_w1', [41, 1024], initializer=tf.truncated_normal_initializer(stddev=0.02))
d_b1 = tf.get_variable('d_b1', [1024], initializer=tf.constant_initializer(0))
d1 = tf.nn.relu(tf.matmul(data, d_w1) + d_b1)
# Fully Connected 2 Wide
d_w2 = tf.get_variable('d_w2', [1024, 6144], initializer=tf.truncated_normal_initializer(stddev=0.02))
d_b2 = tf.get_variable('d_b2', [6144], initializer=tf.constant_initializer(0))
d2 = tf.nn.relu(tf.matmul(d1, d_w2) + d_b2)
# Fully Connected 3 Choking
d_w3 = tf.get_variable('d_w3', [6144, 1024], initializer=tf.truncated_normal_initializer(stddev=0.02))
d_b3 = tf.get_variable('d_b3', [1024], initializer=tf.constant_initializer(0))
d3 = tf.nn.relu(tf.matmul(d2, d_w3) + d_b3)
d_w4 = tf.get_variable('d_w4', [1024, 1], initializer=tf.truncated_normal_initializer(stddev=0.02))
d_b4 = tf.get_variable('d_b4', [1], initializer=tf.constant_initializer(0))
output = tf.nn.sigmoid(tf.matmul(d3, d_w4) + d_b4)
return output
def generator(z, batch_size, z_dim):
# Input layer
g_w1 = tf.get_variable('g_w1', [z_dim, 41], dtype=tf.float32,
initializer=tf.truncated_normal_initializer(stddev=0.02))
g_b1 = tf.get_variable('g_b1', [41], initializer=tf.truncated_normal_initializer(stddev=0.02))
g1 = tf.matmul(z, g_w1) + g_b1
g1 = tf.reshape(g1, [-1, 41])
g1 = tf.contrib.layers.batch_norm(g1, epsilon=1e-5, scope='bn1')
g1 = tf.nn.relu(g1)
g_w2 = tf.get_variable('g_w2', [41, 1024], dtype=tf.float32,
initializer=tf.truncated_normal_initializer(stddev=0.02))
g_b2 = tf.get_variable('g_b2', [1024], initializer=tf.truncated_normal_initializer(stddev=0.02))
g2 = tf.matmul(g1, g_w2) + g_b2
g2 = tf.contrib.layers.batch_norm(g2, epsilon=1e-5, scope='bn2')
g2 = tf.nn.relu(g2)
g_w3 = tf.get_variable('g_w3', [1024, 5120], dtype=tf.float32,
initializer=tf.truncated_normal_initializer(stddev=0.02))
g_b3 = tf.get_variable('g_b3', [5120], initializer=tf.truncated_normal_initializer(stddev=0.02))
g3 = tf.matmul(g2, g_w3) + g_b3
g3 = tf.contrib.layers.batch_norm(g3, epsilon=1e-5, scope='bn3')
g3 = tf.nn.relu(g3)
g_w4 = tf.get_variable('g_w4', [5120, 41], dtype=tf.float32,
initializer=tf.truncated_normal_initializer(stddev=0.02))
g_b4 = tf.get_variable('g_b4', [41], initializer=tf.truncated_normal_initializer(stddev=0.02))
g4 = tf.matmul(g3, g_w4) + g_b4
g4 = tf.sigmoid(g4)
return g4
这是我对优化器/培训师的定义：
batch_size = 50
tf.reset_default_graph()
sess = tf.Session()
z_placeholder = tf.placeholder(tf.float32, [None, z_dimensions], name='z_placeholder') # Hold my noise
x_placeholder = tf.placeholder(tf.float32, shape=[None, 41], name='x_placeholder') # Hold my data
Gz = generator(z_placeholder, batch_size, z_dimensions) # Hold my counterfeits
Dx = discriminator(x_placeholder) # Hold Predictions on the real data
Dg = discriminator(Gz, reuse=True) # Hold Predictions on the fake data
# Loss
d_loss_real = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=Dx, labels=tf.ones_like(Dx)))
d_loss_fake = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=Dg, labels=tf.ones_like(Dg)))
g_loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=Dg, labels=tf.ones_like(Dg)))
# Trainable Vars
# Separate weights and biases via a name prefix basis, Thanks Jon Bruner and Adit Deshpande.
tvars = tf.trainable_variables()
d_vars = [var for var in tvars if 'd_' in var.name]
g_vars = [var for var in tvars if 'g_' in var.name]
print([v.name for v in d_vars])
print([v.name for v in g_vars])
# Optimizers!
with tf.variable_scope(tf.get_variable_scope(), reuse=False):
print("reuse or not: {}".format(tf.get_variable_scope().reuse))
assert tf.get_variable_scope().reuse == False, "Problems!"
d_trainer_fake = tf.train.AdamOptimizer(0.0003).minimize(d_loss_fake, var_list=d_vars)
d_trainer_real = tf.train.AdamOptimizer(0.0003).minimize(d_loss_real, var_list=d_vars)
g_trainer = tf.train.AdamOptimizer(0.0001).minimize(g_loss, var_list=g_vars)
运行我的代码后，我得到了这个奇妙的错误：
Traceback (most recent call last): File "C:/Users/FW/PycharmProjects/GAN IDS/GAN 2.py", line 151, in <module> ['d_w1:0', 'd_b1:0', 'd_w2:0', 'd_b2:0', 'd_w3:0', 'd_b3:0', 'd_w4:0', 'd_b4:0'] ['g_w1:0', 'g_b1:0', 'g_w2:0', 'g_b2:0', 'g_w3:0', 'g_b3:0', 'g_w4:0', 'g_b4:0'] reuse or not: True
assert tf.get_variable_scope().reuse == False, "Problems!" AssertionError: Problems!
如果没有捕获物就会变成这样：
Traceback (most recent call last):
File "C:/Users/FW/PycharmProjects/GAN IDS/GAN 2.py", line 152, in <module>
d_trainer_fake = tf.train.AdamOptimizer(0.0003).minimize(d_loss_fake, var_list=d_vars)
File "C:\Users\FW\Anaconda3\lib\site-packages\tensorflow\python\training\optimizer.py", line 325, in minimize
name=name)
File "C:\Users\FW\Anaconda3\lib\site-packages\tensorflow\python\training\optimizer.py", line 446, in apply_gradients
self._create_slots([_get_variable_for(v) for v in var_list])
File "C:\Users\FW\Anaconda3\lib\site-packages\tensorflow\python\training\adam.py", line 128, in _create_slots
self._zeros_slot(v, "m", self._name)
File "C:\Users\FW\Anaconda3\lib\site-packages\tensorflow\python\training\optimizer.py", line 766, in _zeros_slot
named_slots[_var_key(var)] = slot_creator.create_zeros_slot(var, op_name)
File "C:\Users\FW\Anaconda3\lib\site-packages\tensorflow\python\training\slot_creator.py", line 174, in create_zeros_slot
colocate_with_primary=colocate_with_primary)
File "C:\Users\FW\Anaconda3\lib\site-packages\tensorflow\python\training\slot_creator.py", line 146, in create_slot_with_initializer
dtype)
File "C:\Users\FW\Anaconda3\lib\site-packages\tensorflow\python\training\slot_creator.py", line 66, in _create_slot_var
validate_shape=validate_shape)
File "C:\Users\FW\Anaconda3\lib\site-packages\tensorflow\python\ops\variable_scope.py", line 1065, in get_variable
use_resource=use_resource, custom_getter=custom_getter)
File "C:\Users\FW\Anaconda3\lib\site-packages\tensorflow\python\ops\variable_scope.py", line 962, in get_variable
use_resource=use_resource, custom_getter=custom_getter)
File "C:\Users\FW\Anaconda3\lib\site-packages\tensorflow\python\ops\variable_scope.py", line 367, in get_variable
validate_shape=validate_shape, use_resource=use_resource)
File "C:\Users\FW\Anaconda3\lib\site-packages\tensorflow\python\ops\variable_scope.py", line 352, in _true_getter
use_resource=use_resource)
File "C:\Users\FW\Anaconda3\lib\site-packages\tensorflow\python\ops\variable_scope.py", line 682, in _get_single_variable
"VarScope?" % name)
ValueError: Variable d_w1/Adam/ does not exist, or was not created with tf.get_variable(). Did you mean to set reuse=None in VarScope?
我已经找到了这个问题最常见的线程，到目前为止他们的解决方案都没有，所以我不知道出了什么问题。
我认为在我的模型或代码的优化器部分中出现了可怕的错误。
答案 0 :(得分：2)
因此，您正在创建三个AdamOptimizer个实例。如果要将它们分开，最好的解决方案是在自己的范围内创建每个：
with tf.variable_scope('fake-optimizer'):
d_trainer_fake = tf.train.AdamOptimizer(0.0003).minimize(d_loss_fake, var_list=d_vars)
with tf.variable_scope('real-optimizer'):
d_trainer_real = tf.train.AdamOptimizer(0.0003).minimize(d_loss_real, var_list=d_vars)
with tf.variable_scope('optimizer'):
g_trainer = tf.train.AdamOptimizer(0.0001).minimize(g_loss, var_list=g_vars)
如果出于某种原因，您希望共享其内部参数，则应使用reuse=None创建第一个优化器，然后使用reuse=True创建第二个优化器。或者全部三个reuse=tf.AUTO_REUSE（在tensorflow 1.4中支持），它们也是如此。