MolecularDiffusion.modules.models.diffspectra.specformer_layers¶

Building blocks for specformer – ported verbatim from others/DiffSpectra/models/specformer_layers.py (itself PatchTST’s building blocks, https://github.com/yuqinie98/PatchTST). No behaviour changes; only the unused __all__ re-export list and a couple of docstrings were trimmed.

moving_avg/series_decomp/Coord1dPosEncoding/Coord2dPosEncoding are dead in DMT’s own use of SpecFormer (it always constructs with pe='zeros'), but are kept because positional_encoding() dispatches to them for other pe values and the file is small enough that a partial port would only be a trap for whoever changes pe later.

Attributes¶

SinCosPosEncoding

Classes¶

`Transpose`
`moving_avg`	Moving average block to highlight the trend of a time series.
`series_decomp`	Series decomposition block.

Functions¶

`Coord1dPosEncoding`(q_len[, exponential, normalize])
`Coord2dPosEncoding`(q_len, d_model[, exponential, ...])
`PositionalEncoding`(q_len, d_model[, normalize])
`get_activation_fn`(activation)
`positional_encoding`(pe, learn_pe, q_len, d_model)

Module Contents¶

class MolecularDiffusion.modules.models.diffspectra.specformer_layers.Transpose(*dims, contiguous=False)¶

Bases: torch.nn.Module

forward(x)¶

class MolecularDiffusion.modules.models.diffspectra.specformer_layers.moving_avg(kernel_size, stride)¶

Bases: torch.nn.Module

Moving average block to highlight the trend of a time series.

forward(x)¶

avg¶

kernel_size¶

class MolecularDiffusion.modules.models.diffspectra.specformer_layers.series_decomp(kernel_size)¶

Bases: torch.nn.Module

Series decomposition block.

forward(x)¶

moving_avg¶

MolecularDiffusion.modules.models.diffspectra.specformer_layers.Coord1dPosEncoding(q_len, exponential=False, normalize=True)¶

MolecularDiffusion.modules.models.diffspectra.specformer_layers.Coord2dPosEncoding(q_len, d_model, exponential=False, normalize=True, eps=0.001, verbose=False)¶

MolecularDiffusion.modules.models.diffspectra.specformer_layers.PositionalEncoding(q_len, d_model, normalize=True)¶

MolecularDiffusion.modules.models.diffspectra.specformer_layers.get_activation_fn(activation)¶

MolecularDiffusion.modules.models.diffspectra.specformer_layers.positional_encoding(pe, learn_pe, q_len, d_model)¶

MolecularDiffusion.modules.models.diffspectra.specformer_layers.SinCosPosEncoding¶