关于tf.keras的自定义层中布尔列表的问题

Question

我正在尝试为模型构造一个自定义输出层，以便可以将角度范围限制在[-90,90]中。代码如下：

class OutputLayer(Layer):
    def __init__(self):
        super(OutputLayer, self).__init__()

    def call(self, inputs, **kwargs):
        if_larger_than_90 = (inputs > 90)
        if_smaller_than_minus_90 = (inputs < -90)
        outputs = inputs - 180.0 * if_larger_than_90 + 180.0 * if_smaller_than_minus_90
        return outputs

当我尝试运行它时，它返回一个错误：

Traceback (most recent call last):
  File "E:/Studium/Thesis/Transfer Learning.py", line 78, in <module>
    main()
  File "E:/Studium/Thesis/Transfer Learning.py", line 73, in main
    metrics = a_new_model.evaluate(data_gen)
  File "C:\ProgramData\Miniconda3\envs\TF_2G\lib\site-packages\tensorflow_core\python\keras\engine\training.py", line 833, in evaluate
    use_multiprocessing=use_multiprocessing)
  File "C:\ProgramData\Miniconda3\envs\TF_2G\lib\site-packages\tensorflow_core\python\keras\engine\training_v2.py", line 456, in evaluate
    sample_weight=sample_weight, steps=steps, callbacks=callbacks, **kwargs)
  File "C:\ProgramData\Miniconda3\envs\TF_2G\lib\site-packages\tensorflow_core\python\keras\engine\training_v2.py", line 396, in _model_iteration
    distribution_strategy=strategy)
  File "C:\ProgramData\Miniconda3\envs\TF_2G\lib\site-packages\tensorflow_core\python\keras\engine\training_v2.py", line 610, in _process_inputs
    training_v2_utils._prepare_model_with_inputs(model, adapter.get_dataset())
  File "C:\ProgramData\Miniconda3\envs\TF_2G\lib\site-packages\tensorflow_core\python\keras\engine\training_v2_utils.py", line 185, in _prepare_model_with_inputs
    inputs, target, _ = model._build_model_with_inputs(dataset, targets=None)
  File "C:\ProgramData\Miniconda3\envs\TF_2G\lib\site-packages\tensorflow_core\python\keras\engine\training.py", line 2622, in _build_model_with_inputs
    self._set_inputs(cast_inputs)
  File "C:\ProgramData\Miniconda3\envs\TF_2G\lib\site-packages\tensorflow_core\python\keras\engine\training.py", line 2709, in _set_inputs
    outputs = self(inputs, **kwargs)
  File "C:\ProgramData\Miniconda3\envs\TF_2G\lib\site-packages\tensorflow_core\python\keras\engine\base_layer.py", line 842, in __call__
    outputs = call_fn(cast_inputs, *args, **kwargs)
  File "C:\ProgramData\Miniconda3\envs\TF_2G\lib\site-packages\tensorflow_core\python\keras\engine\sequential.py", line 270, in call
    outputs = layer(inputs, **kwargs)
  File "C:\ProgramData\Miniconda3\envs\TF_2G\lib\site-packages\tensorflow_core\python\keras\engine\base_layer.py", line 842, in __call__
    outputs = call_fn(cast_inputs, *args, **kwargs)
  File "C:\ProgramData\Miniconda3\envs\TF_2G\lib\site-packages\tensorflow_core\python\autograph\impl\api.py", line 237, in wrapper
    raise e.ag_error_metadata.to_exception(e)
TypeError: in converted code:

    E:/Studium/Thesis/Transfer Learning.py:19 call  *
        outputs = inputs - 180.0 * if_larger_than_90 + 180.0 * if_smaller_than_minus_90
    C:\ProgramData\Miniconda3\envs\TF_2G\lib\site-packages\tensorflow_core\python\ops\math_ops.py:924 r_binary_op_wrapper
        x = ops.convert_to_tensor(x, dtype=y.dtype.base_dtype, name="x")
    C:\ProgramData\Miniconda3\envs\TF_2G\lib\site-packages\tensorflow_core\python\framework\ops.py:1184 convert_to_tensor
        return convert_to_tensor_v2(value, dtype, preferred_dtype, name)
    C:\ProgramData\Miniconda3\envs\TF_2G\lib\site-packages\tensorflow_core\python\framework\ops.py:1242 convert_to_tensor_v2
        as_ref=False)
    C:\ProgramData\Miniconda3\envs\TF_2G\lib\site-packages\tensorflow_core\python\framework\ops.py:1296 internal_convert_to_tensor
        ret = conversion_func(value, dtype=dtype, name=name, as_ref=as_ref)
    C:\ProgramData\Miniconda3\envs\TF_2G\lib\site-packages\tensorflow_core\python\framework\tensor_conversion_registry.py:52 _default_conversion_function
        return constant_op.constant(value, dtype, name=name)
    C:\ProgramData\Miniconda3\envs\TF_2G\lib\site-packages\tensorflow_core\python\framework\constant_op.py:227 constant
        allow_broadcast=True)
    C:\ProgramData\Miniconda3\envs\TF_2G\lib\site-packages\tensorflow_core\python\framework\constant_op.py:265 _constant_impl
        allow_broadcast=allow_broadcast))
    C:\ProgramData\Miniconda3\envs\TF_2G\lib\site-packages\tensorflow_core\python\framework\tensor_util.py:449 make_tensor_proto
        _AssertCompatible(values, dtype)
    C:\ProgramData\Miniconda3\envs\TF_2G\lib\site-packages\tensorflow_core\python\framework\tensor_util.py:331 _AssertCompatible
        (dtype.name, repr(mismatch), type(mismatch).__name__))

    TypeError: Expected bool, got 180.0 of type 'float' instead.


Process finished with exit code 1

那么在Tensorflow中使用int * bool之类的命令是否合法？如果是这样，我如何使用其他方法实现相同的目标？

Answer 1

您可以将布尔值转换为浮点值：

    if_larger_than_90 = tf.keras.backend.cast(inputs > 90, "float32")

但是，尝试以这种方式限制网络似乎有点不可思议。最好构造一个将输出保持在范围内的损耗，或者将其限制在网络之外。但是，如果它对您有用-可以。