View source on GitHub |
Converts any tf.IndexedSlices
to tensors.
tff.learning.optimizers.handle_indexed_slices_gradients(
gradients: _Structure[T]
) -> _Structure[Union[T, tf.IndexedSlices]]
The tf.IndexedSlices
class is used principally in the definition of
gradients for operations that have sparse gradients (e.g. tf.gather
). See
also tf.GradientTape
documentation. This method is an elementary utility
converting the slices to a tensor, which can be used to make an optimizer
immediately compatible with such gradients. All other values are left
unmodified.
Note however, this operation may be expensive in some situations. For more details, see https://github.com/tensorflow/tensorflow/blob/2b44549aca184ae0eb986a8bd46feef2b17004ab/tensorflow/python/framework/indexed_slices.py#L406
Args | |
---|---|
gradients
|
A collection of gradients to be used by an optimizer. |
Returns | |
---|---|
The same collection with tf.IndexedSlices replaced by tensors.
|