使用TensorFlow在张量上映射函数的问题

Question

我必须执行一个简单的函数，减去两个列表的元素，并使用TensorFlowfn_map函数将结果放入新列表中。没有这个函数的代码工作得很好，但是当我使用fn_map修改它时，会发生以下错误:

Exception has occurred: ValueError
in user code:
/home/imene/ESS/ess.py:359 sup  *
for el1 in inputs[0]:
/home/imene/anaconda3/envs/master/lib/python3.6/site-packages/tensorflow/python/autograph/operators/control_flow.py:420 for_stmt
iter_, extra_test, body, get_state, set_state, symbol_names, opts)
/home/imene/anaconda3/envs/master/lib/python3.6/site-packages/tensorflow/python/autograph/operators/control_flow.py:485 _known_len_tf_for_stmt
n = py_builtins.len_(iter_)
/home/imene/anaconda3/envs/master/lib/python3.6/site-packages/tensorflow/python/autograph/operators/py_builtins.py:249 len_
return _tf_tensor_len(s)
/home/imene/anaconda3/envs/master/lib/python3.6/site-packages/tensorflow/python/autograph/operators/py_builtins.py:277 _tf_tensor_len
'len requires a non-scalar tensor, got one of shape {}'.format(shape))
ValueError: len requires a non-scalar tensor, got one of shape []
File "/tmp/tmpgps8mh9_.py", line 33, in tf__sup
ag__.for_stmt(ag__.ld(inputs)[0], None, loop_body_1, get_state_1, set_state_1, (), {'iterate_names': 'el1'})
During handling of the above exception, another exception occurred:
File "/home/imene/ESS/ess.py", line 366, in supp
ss =tf.map_fn(sup, inputs, dtype = tf.float32)
File "/home/imene/ESS/ess.py", line 373, in <module>
ss  = supp(ab)

代码如下:

import tensorflow as tf

a = [1, 3, 5, 8] 
b = [1, 5, 3, 60]

def supp(inputs):
def sup(inputs):
ss = []
for el1 in inputs[0]:
for el2 in inputs[1]:
ss.append(el1 - el2)
return ss  

ss = tf.map_fn(sup, inputs, dtype=tf.float32)
return ss
a = tf.convert_to_tensor(a)      
b = tf.convert_to_tensor(b)  
ab = tf.convert_to_tensor([a, b])
ss  = supp(ab)
print(ss)

有什么问题吗?

Answer 1

也许只是试着在你的张量a和b上做元素减法，而不做任何循环:

import tensorflow as tf
a = [1, 3, 5, 8] 
b = [1, 5, 3, 60]
def supp(inputs):
def sup(inputs):
return tf.cast(inputs[0] - inputs[1], dtype=tf.float32)  

ss = tf.map_fn(sup, inputs, dtype=tf.float32)
return ss
a = tf.convert_to_tensor(a)      
b = tf.convert_to_tensor(b)  
ss  = supp([a, b])
print(ss)
# tf.Tensor([  0.  -2.   2. -52.], shape=(4,), dtype=float32)
print(ss.numpy())
# [  0.  -2.   2. -52.]