tf.register_tensor_conversion_function( base_type, conversion_func, priority=100 )
See the guide: Building Graphs > For libraries building on TensorFlow
Registers a function for converting objects of
The conversion function must have the following signature:
def conversion_func(value, dtype=None, name=None, as_ref=False): # ...
It must return a
Tensor with the given
dtype if specified. If the conversion function creates a new
Tensor, it should use the given
name if specified. All exceptions will be propagated to the caller.
The conversion function may return
NotImplemented for some inputs. In this case, the conversion process will continue to try subsequent conversion functions.
as_ref is true, the function must return a
Tensor reference, such as a
NOTE: The conversion functions will execute in order of priority, followed by order of registration. To ensure that a conversion function
F runs before another conversion function
G, ensure that
F is registered with a smaller priority than
base_type: The base type or tuple of base types for all objects that
conversion_func: A function that converts instances of
priority: Optional integer that indicates the priority for applying this conversion function. Conversion functions with smaller priority values run earlier than conversion functions with larger priority values. Defaults to 100.
TypeError: If the arguments do not have the appropriate type.
© 2018 The TensorFlow Authors. All rights reserved.
Licensed under the Creative Commons Attribution License 3.0.
Code samples licensed under the Apache 2.0 License.