tensorflow 2.0：正在传递函数构建代码之外的操作

Question

我收到一个错误：

TypeError: An op outside of the function building code is being passed
a "Graph" tensor. It is possible to have Graph tensors
leak out of the function building context by including a
tf.init_scope in your function building code.
For example, the following function will fail:
  @tf.function
  def has_init_scope():
    my_constant = tf.constant(1.)
    with tf.init_scope():
      added = my_constant * 2

使用如下的 NVP 层：

import tensorflow_probability as tfp
tfb = tfp.bijectors
tfd = tfp.distributions
class NVPLayer(tf.keras.models.Model):

    def __init__(self, *, output_dim, num_masked, **kwargs):
        super().__init__(**kwargs)
        self.output_dim = output_dim
        self.num_masked = num_masked
        self.shift_and_log_scale_fn = tfb.real_nvp_default_template(
            hidden_layers=[2], # HERE HERE ADJUST THIS
            activation=None, # linear
            )
        self.loss = None

    def get_nvp(self):
        nvp = tfd.TransformedDistribution(
            distribution=tfd.MultivariateNormalDiag(loc=[0.] * self.output_dim),
            bijector=tfb.RealNVP(
                num_masked=self.num_masked,
                shift_and_log_scale_fn=self.shift_and_log_scale_fn)
            )
        return nvp

    def call(self, *inputs):
        nvp = self.get_nvp()
        self.loss = tf.reduce_mean(nvp.log_prob(*inputs)) # how else to do this?
        # return nvp.bijector.forward(*inputs)
        return nvp.bijector.inverse(*inputs)

我不会在任何地方打电话给 tf.init_scope。训练类似层的简单版本似乎有效。

我会尝试获得更细粒度的跟踪，但我怀疑这与非热切模式有关。

更新： 所以这肯定来自 self.loss 包含在某些渐变带层中。正确的做法是什么？

Answer 1

UPDATE: so this is definitely coming from the self.loss inclusion in some gradient tape layer. What is the correct way of doing this?

我认为正确的做法是

self.add_loss(<your loss tensor>)

（https://www.tensorflow.org/api_docs/python/tf/keras/layers/Layer#add_loss 了解更多）

（编辑抱歉，我没有注意你的 post 的日期，所以我想这已经不是很有用了哈哈）

Answer 2

几分钟前刚遇到同样的问题，在我的例子中，我想修改损失函数中的状态 class，这是我在你的例子中的解决方法。

顺便说一句 @simon 给了我如何正确评估这一点的灵感。所以支持他！

您似乎应该为训练时要更改的属性创建一个 tf.Variable。请注意，其他属性（例如 self.output_dim、self.num_masked 和其他

没有任何问题

试试这个：

import tensorflow_probability as tfp
tfb = tfp.bijectors
tfd = tfp.distributions
class NVPLayer(tf.keras.models.Model):

def __init__(self, *, output_dim, num_masked, **kwargs):
    super().__init__(**kwargs)
    self.output_dim = output_dim
    self.num_masked = num_masked
    self.shift_and_log_scale_fn = tfb.real_nvp_default_template(
        hidden_layers=[2], # HERE HERE ADJUST THIS
        activation=None, # linear
        )

    ###CHANGE HERE
    self.loss = tf.Variable(0.0)

def get_nvp(self):
    nvp = tfd.TransformedDistribution(
        distribution=tfd.MultivariateNormalDiag(loc=[0.] * self.output_dim),
        bijector=tfb.RealNVP(
            num_masked=self.num_masked,
            shift_and_log_scale_fn=self.shift_and_log_scale_fn)
        )
    return nvp

def call(self, *inputs):
    nvp = self.get_nvp()

    ### CHANGE HERE
    self.loss.assign(tf.reduce_mean(nvp.log_prob(*inputs)))
    # return nvp.bijector.forward(*inputs)
    return nvp.bijector.inverse(*inputs)

在 github 问题上也查看这个 answer，类似的问题！

tensorflow 2.0：正在传递函数构建代码之外的操作

tensorflow 2.0: An op outside of the function building code is being passed

python

tensorflow

tensorflow2.0