简单来说，目的是在MyClass中，根据self.inputs选择合适的membership function，要么是rul，要么是2-rul，在def __init__中定义self.inputs的shape =（None, n_inputs）,所以rul.shape[0]也是未知的。由于张量的值不能修改，计划先用tf.where()提取indices，随后用tf.gather_nd提取对应的值，然后用tf.SparseTensor构建稀疏张量，最后用tf.assign更新self.rul。具体代码如下。现在问题来了，到tf.SparseTensor这里报错“ValueError: Cannot convert a partially known TensorShape to a Tensor: (?, 24)”，我觉得是dense_shape应该要明确指定的缘故。诚心求教，怎么解决这个问题？不胜感激！
import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
class MyClass:
def __init__(self, n_inputs, n_rules, learning_rate=1e-2):
self.n = n_inputs
self.m = n_rules
self.inputs = tf.placeholder(tf.float32, shape=(None, n_inputs)) # Input
self.targets = tf.placeholder(tf.float32, shape=None) # Desired output
mu = tf.get_variable("mu", [n_rules * n_inputs],
initializer=tf.random_normal_initializer(0, 1)) # Means of Gaussian MFS
sigma = tf.get_variable("sigma", [n_rules * n_inputs],
initializer=tf.random_normal_initializer(0, 1)) # Standard deviations of Gaussian MFS
y = tf.get_variable("y", [1, n_rules], initializer=tf.random_normal_initializer(0, 1)) # Sequent centers
self.params = tf.trainable_variables()
# rules activation.
Rul = tf.exp(-0.5 * tf.square(tf.subtract(tf.tile(self.inputs,(1,n_rules)),mu)) / tf.square(sigma))
Rul_diff = 2. - 2 * Rul
indices = tf.where(tf.less(tf.tile(tf.reduce_mean(self.inputs, axis = 0),[n_rules]),tf.tile(self.inputs,(1,n_rules))))
gather_nd = tf.gather_nd(Rul, indices)
Rul1 = tf.SparseTensor(indices, gather_nd, tf.shape(Rul))
Rul_add = tf.sparse_tensor_to_dense(tf.SparseTensor(indices, gather_nd, tf.shape(Rul)))
Ruls = tf.assign_add(Rul, Rul_add)
........
~\ in __init__(self, n_inputs, n_rules, learning_rate)
36 indices = tf.where(tf.less(tf.tile(tf.reduce_mean(self.inputs, axis = 0),[n_rules]),tf.tile(self.inputs,(1,n_rules))))
37 gather_nd = tf.gather_nd(Rul, indices)
---> 38 Rul1 = tf.SparseTensor(indices, gather_nd, tf.shape(Rul))
39 #Rul_add = tf.sparse_tensor_to_dense(tf.SparseTensor(indices, gather_nd, tf.shape(Rul)))
40 Rul_add = tf.sparses_tensor_to_dense(Rul1)
D:\Anaconda3\envs\tf1.5\lib\site-packages\tensorflow\python\framework\sparse_tensor.py in __init__(self, indices, values, dense_shape)
126 values, name="values", as_ref=True)
127 dense_shape = ops.convert_to_tensor(
--> 128 dense_shape, name="dense_shape", dtype=dtypes.int64)
129 self._indices = indices
130 self._values = values
D:\Anaconda3\envs\tf1.5\lib\site-packages\tensorflow\python\framework\ops.py in convert_to_tensor(value, dtype, name, preferred_dtype)
944 name=name,
945 preferred_dtype=preferred_dtype,
--> 946 as_ref=False)
947
948
D:\Anaconda3\envs\tf1.5\lib\site-packages\tensorflow\python\framework\ops.py in internal_convert_to_tensor(value, dtype, name, as_ref, preferred_dtype, ctx)
1034
1035 if ret is None:
-> 1036 ret = conversion_func(value, dtype=dtype, name=name, as_ref=as_ref)
1037
1038 if ret is NotImplemented:
D:\Anaconda3\envs\tf1.5\lib\site-packages\tensorflow\python\framework\constant_op.py in _tensor_shape_tensor_conversion_function(s, dtype, name, as_ref)
254 if not s.is_fully_defined():
255 raise ValueError(
--> 256 "Cannot convert a partially known TensorShape to a Tensor: %s" % s)
257 s_list = s.as_list()
258 int64_value = 0
ValueError: Cannot convert a partially known TensorShape to a Tensor: (?, 24)
你好，我是有问必答小助手，非常抱歉，本次您提出的有问必答问题，技术专家团超时未为您做出解答
本次提问扣除的有问必答次数，将会以问答VIP体验卡（1次有问必答机会、商城购买实体图书享受95折优惠）的形式为您补发到账户。
因为有问必答VIP体验卡有效期仅有1天，您在需要使用的时候【私信】联系我，我会为您补发。
几天过去，最后还是自己解决这个问题。不使用
tf.SparseTensor()
而是直接使用
tf.where(cond, x, y)
就可以实现目的。
通过这次案例，深刻理解了
tf.where(cond, x, y)
x与y为None和不为None的区别。
推及到其他函数，提醒自己要认真理解函数的说明文档。特别在此记录一下。