- neural_tangents.stax.AvgPool(window_shape, strides=None, padding='VALID', normalize_edges=False, batch_axis=0, channel_axis=-1)[source]
int]) – The number of pixels over which pooling is to be performed.
int]]) – The stride of the pooling window.
Nonecorresponds to a stride of
str) – Can be
CIRCULARuses periodic boundary conditions on the image.
Trueto normalize output by the effective receptive field,
Falseto normalize by the window size. Only has effect at the edges when
SAMEpadding is used. Set to
Trueto retain correspondence to
int) – Specifies the batch dimension. Defaults to
0, the leading axis.
int) – Specifies the channel / feature dimension. Defaults to
-1, the trailing axis. For
kernel_fn, channel size is considered to be infinite.
- Return type
(init_fn, apply_fn, kernel_fn).