我在自定义损失函数中使用K.map_fn（），在该函数中我同时传递了形状（无，无）的y_true和y_pred作为此函数的elems参数。但是，当调用map_fn中指定的函数时，在该函数中获得的元素具有不同的形状。就是这个问题。
这里是例子：
我的自定义损失函数：
def negative_avg_log_error(y_true, y_pred):
def sum_of_log_probabilities(true_and_pred):
y_true, y_pred = true_and_pred
print(K.int_shape(y_true))
print(K.int_shape(y_pred))
start_index = int(y_true[0])
end_index = int(y_true[1])
start_probability = y_pred[start_index]
end_probability = y_pred[end_index]
return K.log(start_probability) + K.log(end_probability)
print(K.int_shape(y_true))
print(K.int_shape(y_pred))
batch_probability_sum = K.map_fn(lambda x: sum_of_log_probabilities(x), elems=[y_true, y_pred], dtype='float32')
return -K.mean(batch_probability_sum, axis=1)
调用此损失函数的代码：
model.compile(loss=negative_avg_log_error, optimizer='adadelta', metrics='loss')
错误日志：
(None, None)
(None, None)
(None,)
(None, 1)
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
<ipython-input-15-807b6d06a435> in <module>()
----> 1 model.compile(loss=negative_avg_log_error, optimizer='adadelta', metrics='loss')
C:\Python36\lib\site-packages\keras\engine\training.py in compile(self, optimizer, loss, metrics, loss_weights, sample_weight_mode, weighted_metrics, target_tensors, **kwargs)
331 with K.name_scope(self.output_names[i] + '_loss'):
332 output_loss = weighted_loss(y_true, y_pred,
--> 333 sample_weight, mask)
334 if len(self.outputs) > 1:
335 self.metrics_tensors.append(output_loss)
C:\Python36\lib\site-packages\keras\engine\training_utils.py in weighted(y_true, y_pred, weights, mask)
401 """
402 # score_array has ndim >= 2
--> 403 score_array = fn(y_true, y_pred)
404 if mask is not None:
405 # Cast the mask to floatX to avoid float64 upcasting in Theano
E:\Deep Learning Material\IMP for Project\model\scripts\loss_function.py in negative_avg_log_error(y_true, y_pred)
16 print(K.int_shape(y_true))
17 print(K.int_shape(y_pred))
---> 18 batch_probability_sum = K.map_fn(lambda x: sum_of_log_probabilities(x), elems=[y_true, y_pred], dtype='float32')
19 return -K.mean(batch_probability_sum, axis=1)
C:\Python36\lib\site-packages\keras\backend\tensorflow_backend.py in map_fn(fn, elems, name, dtype)
4229 Tensor with dtype `dtype`.
4230 """
-> 4231 return tf.map_fn(fn, elems, name=name, dtype=dtype)
4232
4233
C:\Python36\lib\site-packages\tensorflow\python\ops\functional_ops.py in map_fn(fn, elems, dtype, parallel_iterations, back_prop, swap_memory, infer_shape, name)
457 back_prop=back_prop,
458 swap_memory=swap_memory,
--> 459 maximum_iterations=n)
460 results_flat = [r.stack() for r in r_a]
461
C:\Python36\lib\site-packages\tensorflow\python\ops\control_flow_ops.py in while_loop(cond, body, loop_vars, shape_invariants, parallel_iterations, back_prop, swap_memory, name, maximum_iterations, return_same_structure)
3230 ops.add_to_collection(ops.GraphKeys.WHILE_CONTEXT, loop_context)
3231 result = loop_context.BuildLoop(cond, body, loop_vars, shape_invariants,
-> 3232 return_same_structure)
3233 if maximum_iterations is not None:
3234 return result[1]
C:\Python36\lib\site-packages\tensorflow\python\ops\control_flow_ops.py in BuildLoop(self, pred, body, loop_vars, shape_invariants, return_same_structure)
2950 with ops.get_default_graph()._mutation_lock(): # pylint: disable=protected-access
2951 original_body_result, exit_vars = self._BuildLoop(
-> 2952 pred, body, original_loop_vars, loop_vars, shape_invariants)
2953 finally:
2954 self.Exit()
C:\Python36\lib\site-packages\tensorflow\python\ops\control_flow_ops.py in _BuildLoop(self, pred, body, original_loop_vars, loop_vars, shape_invariants)
2885 flat_sequence=vars_for_body_with_tensor_arrays)
2886 pre_summaries = ops.get_collection(ops.GraphKeys._SUMMARY_COLLECTION) # pylint: disable=protected-access
-> 2887 body_result = body(*packed_vars_for_body)
2888 post_summaries = ops.get_collection(ops.GraphKeys._SUMMARY_COLLECTION) # pylint: disable=protected-access
2889 if not nest.is_sequence(body_result):
C:\Python36\lib\site-packages\tensorflow\python\ops\control_flow_ops.py in <lambda>(i, lv)
3199 cond = lambda i, lv: ( # pylint: disable=g-long-lambda
3200 math_ops.logical_and(i < maximum_iterations, orig_cond(*lv)))
-> 3201 body = lambda i, lv: (i + 1, orig_body(*lv))
3202
3203 if context.executing_eagerly():
C:\Python36\lib\site-packages\tensorflow\python\ops\functional_ops.py in compute(i, tas)
446 """
447 packed_values = input_pack([elem_ta.read(i) for elem_ta in elems_ta])
--> 448 packed_fn_values = fn(packed_values)
449 nest.assert_same_structure(dtype or elems, packed_fn_values)
450 flat_fn_values = output_flatten(packed_fn_values)
E:\Deep Learning Material\IMP for Project\model\scripts\loss_function.py in <lambda>(x)
16 print(K.int_shape(y_true))
17 print(K.int_shape(y_pred))
---> 18 batch_probability_sum = K.map_fn(lambda x: sum_of_log_probabilities(x), elems=[y_true, y_pred], dtype='float32')
19 return -K.mean(batch_probability_sum, axis=1)
E:\Deep Learning Material\IMP for Project\model\scripts\loss_function.py in sum_of_log_probabilities(true_and_pred)
8 print(K.int_shape(y_true))
9 print(K.int_shape(y_pred))
---> 10 start_index = int(y_true[0])
11 end_index = int(y_true[1])
12 start_probability = y_pred[start_index]
TypeError: int() argument must be a string, a bytes-like object or a number, not 'Tensor'
我目前专注于： 在negative_avg_log_error（）=> K.int_shape（y_true）=（无，无）和K.int_shape（y_pred）=（无，无） 在sum_of_log_probabilities（）=> K.int_shape（y_true）= {None，）和K.int_shape（y_pred）= {None，1）
但是我认为sum_of_log_probabilities（）中这两个张量的形状应该为（None，）。
需要帮助，因为这将帮助我解决此错误。 我还尝试过将y_true和y_pred作为元组而不是具有相同结果的列表传递。
在Keras / Tensorflow打开了一个问题：https://github.com/keras-team/keras/issues/11918