• Index

API reference

nn.attention

nn.DotProductAttention(qdim[, output_dim, ...])	DotProductAttention.
nn.SelfAttention(qdim[, dropout, transform, ...])	Self Attention module using DotProductAttention
nn.SelfMultiheadAttention(embed_dim, num_heads)	Self Attention module using torch.nn.MultiheadAttention

nn.loss

nn.JSDivLoss([size_average, reduce, reduction])
nn.KLDivLoss([size_average, reduce, reduction])

nn.pe

nn.PositionalEncoding(d_model[, dropout, ...])

nn.pyramid

nn.LaplacianPyramidLayer()
nn.PyramidDown()
nn.PyramidUp()

nn.residual

nn.ResidualBlock(input_dim[, bottleneck, ...])

Residual Block If bottleneck=None, this is plain Residual Block with 2 FC layers (input_dim=>input_dim=>input_dim) and 2 activation layers.

nn.sum

nn.SumLayer(dim)

experimental.variance_decomposition

experimental.variance_decomposition.variance_decomposition(...)	param inputs inputs data. B (batch size) x D (Dimension) or B (batch size) x L (input data length) x D (Dimension)
experimental.variance_decomposition.VarianceDecomposition(...)	Variance decomposition module

previous

pytorch_extra_mhirano

next

pytorch_extra_mhirano.nn.DotProductAttention