对于我的损失函数，我给出了错误： ValueError: None values not supported

Question

我在下面写了损失函数：

def custom_loss(q_k):
  def loss(y_true,y_pred):
    loss= y_true * y_true /np.log(y_pred + q_k)

  return loss

我给出了错误：

Cannot convert a symbolic Tensor (2nd_target:0) to a NumPy array

然后根据@Dr.snoopy 和@Kaveh，我编辑了我的损失函数但我得到了一个新错误，我的新损失函数：

def custom_loss(q_k):
  def loss(y_true,y_pred):
     return y_true * y_true /tf.math.log(y_pred + q_k)
 
  # Return a function
  return loss

我应该说y_true 和q_k 是张量。

但我得到了错误：

ValueError: in user code:

    /usr/local/lib/python3.7/dist-packages/keras/engine/training.py:853 train_function  *
        return step_function(self, iterator)
    /usr/local/lib/python3.7/dist-packages/keras/engine/training.py:842 step_function  **
        outputs = model.distribute_strategy.run(run_step, args=(data,))
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/distribute/distribute_lib.py:1286 run
        return self._extended.call_for_each_replica(fn, args=args, kwargs=kwargs)
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/distribute/distribute_lib.py:2849 call_for_each_replica
        return self._call_for_each_replica(fn, args, kwargs)
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/distribute/distribute_lib.py:3632 _call_for_each_replica
        return fn(*args, **kwargs)
    /usr/local/lib/python3.7/dist-packages/keras/engine/training.py:835 run_step  **
        outputs = model.train_step(data)
    /usr/local/lib/python3.7/dist-packages/keras/engine/training.py:789 train_step
        y, y_pred, sample_weight, regularization_losses=self.losses)
    /usr/local/lib/python3.7/dist-packages/keras/engine/compile_utils.py:201 __call__
        loss_value = loss_obj(y_t, y_p, sample_weight=sw)
    /usr/local/lib/python3.7/dist-packages/keras/losses.py:143 __call__
        losses, sample_weight, reduction=self._get_reduction())
    /usr/local/lib/python3.7/dist-packages/keras/utils/losses_utils.py:308 compute_weighted_loss
        losses = tf.convert_to_tensor(losses)
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/util/dispatch.py:206 wrapper
        return target(*args, **kwargs)
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/framework/ops.py:1431 convert_to_tensor_v2_with_dispatch
        value, dtype=dtype, dtype_hint=dtype_hint, name=name)
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/framework/ops.py:1441 convert_to_tensor_v2
        as_ref=False)
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/profiler/trace.py:163 wrapped
        return func(*args, **kwargs)
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/framework/ops.py:1566 convert_to_tensor
        ret = conversion_func(value, dtype=dtype, name=name, as_ref=as_ref)
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/framework/constant_op.py:346 _constant_tensor_conversion_function
        return constant(v, dtype=dtype, name=name)
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/framework/constant_op.py:272 constant
        allow_broadcast=True)
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/framework/constant_op.py:290 _constant_impl
        allow_broadcast=allow_broadcast))
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/framework/tensor_util.py:445 make_tensor_proto
        raise ValueError("None values not supported.")

    ValueError: None values not supported.

代码

更新：

感谢@Matthias Fripp，我将损失编辑为：

    def custom_loss(q_k):
       def loss_func(y_true,y_pred):
           return y_true * tf.math.log(y_pred + q_k)               
       return loss_func

但我收到新错误： InvalidArgumentError：需要可广播的形状 [[node gradient_tape/loss/mul/Mul（定义于：1）]] [Op:__inference_train_function_2026]

函数调用栈： train_function

Answer 1

您不能在损失函数中使用 numpy 数组，因为它将在图形模式下执行。

改为使用这样的 tensorflow 方法：

def custom_loss(q_k):

  def loss(y_true,y_pred):
    return y_true * tf.math.log(y_pred + q_k) #tf.math.log instead of np.log
 
  # Return a function
  return loss