(tf2.keras) InternalError: Recorded operation 'GradientReversalOperator' returned too few gradients. Expected 3 but received 2

Question

我的代码在 github 上可用。

我写了一个自定义渐变图层如下：

@tf.custom_gradient
def GradientReversalOperator(x, lambdal):
    def grad(dy):
        return lambdal * tf.negative(dy)
    return x, grad


class GradientReversalLayer(tf.keras.layers.Layer):
    def __init__(self, lambdal):
        super(GradientReversalLayer, self).__init__()
        self.lambdal = lambdal

    def call(self, inputs):
        return GradientReversalOperator(inputs, self.lambdal)

如果我删除 lambdal，一切正常。但是当我把它加回去时，我得到了错误：

InternalError: Recorded operation 'GradientReversalOperator' returned too few gradients. Expected 3 but received 2

一些答案报告我应该再制作一个假的 return 值，但错误变成了“梯度太多”。回溯如下：

File "D:\Users\xiqxi\Anaconda3\envs\tf2\lib\contextlib.py", line 130, in exit self.gen.throw(type, value, traceback) File "D:\Users\xiqxi\Anaconda3\envs\tf2\lib\site-packages\tensorflow\python\ops\variable_scope.py", line 2804, in variable_creator_scope yield File "D:\Users\xiqxi\Anaconda3\envs\tf2\lib\site-packages\tensorflow\python\keras\engine\training.py", line 1695, in train_on_batch logs = train_function(iterator) File "D:\Users\xiqxi\Anaconda3\envs\tf2\lib\site-packages\tensorflow\python\eager\def_function.py", line 780, in call result = self._call(*args, **kwds) File "D:\Users\xiqxi\Anaconda3\envs\tf2\lib\site-packages\tensorflow\python\eager\def_function.py", line 823, in _call self._initialize(args, kwds, add_initializers_to=initializers) File "D:\Users\xiqxi\Anaconda3\envs\tf2\lib\site-packages\tensorflow\python\eager\def_function.py", line 697, in _initialize *args, **kwds)) File "D:\Users\xiqxi\Anaconda3\envs\tf2\lib\site-packages\tensorflow\python\eager\function.py", line 2855, in _get_concrete_function_internal_garbage_collected graph_function, _, _ = self._maybe_define_function(args, kwargs) File "D:\Users\xiqxi\Anaconda3\envs\tf2\lib\site-packages\tensorflow\python\eager\function.py", line 3213, in _maybe_define_function graph_function = self._create_graph_function(args, kwargs) File "D:\Users\xiqxi\Anaconda3\envs\tf2\lib\site-packages\tensorflow\python\eager\function.py", line 3075, in _create_graph_function capture_by_value=self._capture_by_value), File "D:\Users\xiqxi\Anaconda3\envs\tf2\lib\site-packages\tensorflow\python\framework\func_graph.py", line 986, in func_graph_from_py_func func_outputs = python_func(*func_args, **func_kwargs) File "D:\Users\xiqxi\Anaconda3\envs\tf2\lib\site-packages\tensorflow\python\eager\def_function.py", line 600, in wrapped_fn return weak_wrapped_fn().wrapped(*args, **kwds) File "D:\Users\xiqxi\Anaconda3\envs\tf2\lib\site-packages\tensorflow\python\framework\func_graph.py", line 973, in wrapper raise e.ag_error_metadata.to_exception(e) tensorflow.python.framework.errors_impl.InternalError: in user code:
D:\Users\xiqxi\Anaconda3\envs\tf2\lib\site-packages\tensorflow\python\keras\engine\training.py:806 train_function * return step_function(self, iterator) D:\Users\xiqxi\Anaconda3\envs\tf2\lib\site-packages\tensorflow\python\keras\engine\training.py:796 step_function ** outputs = model.distribute_strategy.run(run_step, args=(data,)) D:\Users\xiqxi\Anaconda3\envs\tf2\lib\site-packages\tensorflow\python\distribute\distribute_lib.py:1211 run return self._extended.call_for_each_replica(fn, args=args, kwargs=kwargs) D:\Users\xiqxi\Anaconda3\envs\tf2\lib\site-packages\tensorflow\python\distribute\distribute_lib.py:2585 call_for_each_replica return self._call_for_each_replica(fn, args, kwargs) D:\Users\xiqxi\Anaconda3\envs\tf2\lib\site-packages\tensorflow\python\distribute\distribute_lib.py:2945 _call_for_each_replica return fn(*args, **kwargs) D:\Users\xiqxi\Anaconda3\envs\tf2\lib\site-packages\tensorflow\python\keras\engine\training.py:789 run_step ** outputs = model.train_step(data) D:\Users\xiqxi\Anaconda3\envs\tf2\lib\site-packages\tensorflow\python\keras\engine\training.py:757 train_step self.trainable_variables) D:\Users\xiqxi\Anaconda3\envs\tf2\lib\site-packages\tensorflow\python\keras\engine\training.py:2722 _minimize gradients = tape.gradient(loss, trainable_variables) D:\Users\xiqxi\Anaconda3\envs\tf2\lib\site-packages\tensorflow\python\eager\backprop.py:1073 gradient unconnected_gradients=unconnected_gradients) D:\Users\xiqxi\Anaconda3\envs\tf2\lib\site-packages\tensorflow\python\eager\imperative_grad.py:77 imperative_grad compat.as_str(unconnected_gradients.value)) InternalError: Recorded operation 'GradientReversalOperator' returned too few gradients. Expected 3 but received 2

Answer 1

我遇到了同样的问题，试试这个：

class GradientReversal(Layer):
    '''Flip the sign of gradient during training.'''

    @tf.custom_gradient
    def grad_reverse(self, x):
        y = tf.identity(x)
        def custom_grad(dy):
            return -self.hp_lambda * dy
        return y, custom_grad

    # --------------------------------------
    def __init__(self, hp_lambda, **kwargs):
        super(GradientReversal, self).__init__(**kwargs)
        self.hp_lambda = K.variable(hp_lambda, dtype='float32',     name='hp_lambda')


    # --------------------------------------
    def call(self, x, mask=None):
        return self.grad_reverse(x)

    # --------------------------------------
    def set_hp_lambda(self,hp_lambda):
        #self.hp_lambda = hp_lambda
        K.set_value(self.hp_lambda, hp_lambda)

    # --------------------------------------
    def increment_hp_lambda_by(self,increment):
        new_value = float(K.get_value(self.hp_lambda)) +  increment
        K.set_value(self.hp_lambda, new_value)

# --------------------------------------
    def get_hp_lambda(self):
        return float(K.get_value(self.hp_lambda))

(tf2.keras) InternalError: Recorded operation 'GradientReversalOperator' returned too few gradients. Expected 3 but received 2

(tf2.keras) InternalError: Recorded operation 'GradientReversalOperator' returned too few gradients. Expected 3 but received 2

keras

tensorflow

tf.keras

tensorflow2.0