How this happen?

[ursb2017]

I use the example code，but…

[ursb2017]

`import edward as ed
import numpy as np
import tensorflow as tf
from edward.models.random_variables import *
import sys

x_train = np.linspace(-3, 3, num=10)
y_train = np.cos(x_train) + np.random.normal(0, 0.1, size=10)
x_train = x_train.astype(np.float32).reshape((10, 1))
y_train = y_train.astype(np.float32).reshape((10, 1))

W_0 = Normal(loc=tf.zeros([1, 2]), scale=tf.ones([1, 2]))
W_1 = Normal(loc=tf.zeros([2, 1]), scale=tf.ones([2, 1]))
b_0 = Normal(loc=tf.zeros(2), scale=tf.ones(2))
b_1 = Normal(loc=tf.zeros(1), scale=tf.ones(1))

x = x_train
y = Normal(loc=tf.matmul(tf.tanh(tf.matmul(x, W_0) + b_0), W_1) + b_1,
scale=0.1)

qW_0 = Normal(loc=tf.get_variable("qW_0/loc", [1, 2]),
scale=tf.nn.softplus(tf.get_variable("qW_0/scale", [1, 2])))
qW_1 = Normal(loc=tf.get_variable("qW_1/loc", [2, 1]),
scale=tf.nn.softplus(tf.get_variable("qW_1/scale", [2, 1])))
qb_0 = Normal(loc=tf.get_variable("qb_0/loc", [2]),
scale=tf.nn.softplus(tf.get_variable("qb_0/scale", [2])))
qb_1 = Normal(loc=tf.get_variable("qb_1/loc", [1]),
scale=tf.nn.softplus(tf.get_variable("qb_1/scale", [1])))

inference = ed.KLqp({W_0: qW_0, b_0: qb_0, W_1: qW_1, b_1: qb_1}, data={y: y_train})
inference.run(n_iter=1)

`

[ursb2017]

Traceback (most recent call last):
File "D:/学习和科研/科研/研究项目/强化学习ing/RL/Bayesian_Inference/Test.py", line 33, in
inference.run(n_iter=1)
File "D:\Anaconda\lib\site-packages\edward\inferences\inference.py", line 125, in run
self.initialize(*args, **kwargs)
File "D:\Anaconda\lib\site-packages\edward\inferences\klqp.py", line 110, in initialize
return super(KLqp, self).initialize(*args, **kwargs)
File "D:\Anaconda\lib\site-packages\edward\inferences\variational_inference.py", line 68, in initialize
self.loss, grads_and_vars = self.build_loss_and_gradients(var_list)
File "D:\Anaconda\lib\site-packages\edward\inferences\klqp.py", line 145, in build_loss_and_gradients
return build_reparam_kl_loss_and_gradients(self, var_list)
File "D:\Anaconda\lib\site-packages\edward\inferences\klqp.py", line 717, in build_reparam_kl_loss_and_gradients
qz_copy = copy(qz, scope=scope)
File "D:\Anaconda\lib\site-packages\edward\util\random_variables.py", line 229, in copy
copy(v, dict_swap, scope, True, copy_q, True)
File "D:\Anaconda\lib\site-packages\edward\util\random_variables.py", line 229, in copy
copy(v, dict_swap, scope, True, copy_q, True)
File "D:\Anaconda\lib\site-packages\edward\util\random_variables.py", line 229, in copy
copy(v, dict_swap, scope, True, copy_q, True)
[Previous line repeated 983 more times]
File "D:\Anaconda\lib\site-packages\edward\util\random_variables.py", line 228, in copy
for v in get_parents(org_instance):
File "D:\Anaconda\lib\site-packages\edward\util\random_variables.py", line 608, in get_parents
collection = random_variables()
File "D:\Anaconda\lib\site-packages\edward\util\graphs.py", line 54, in random_variables
graph = tf.get_default_graph()
File "D:\Anaconda\lib\site-packages\tensorflow_core\python\framework\ops.py", line 5874, in get_default_graph
return _default_graph_stack.get_default()
File "D:\Anaconda\lib\site-packages\tensorflow_core\python\framework\ops.py", line 5454, in get_default
ret = super(_DefaultGraphStack, self).get_default()
File "D:\Anaconda\lib\site-packages\tensorflow_core\python\framework\ops.py", line 5266, in get_default
return self.stack[-1] if len(self.stack) >= 1 else None
RecursionError: maximum recursion depth exceeded in comparison

[raquelaoki]

I'm having the same problem when using the ed.copy()