MolecularDiffusion.modules.models.difflinker.noise¶

Noise schedules, ported verbatim from DiffLinker’s src/noise.py.

Classes¶

`GammaNetwork`	Models a monotonic increasing function, construction as in the VDM paper.
`PositiveLinear`	Linear layer with weights forced to be positive.
`PredefinedNoiseSchedule`	Lookup array for predefined (non-learned) noise schedules.

Functions¶

`clip_noise_schedule`(alphas2[, clip_value])	For a noise schedule given by alpha^2, clip alpha_t / alpha_t-1 to
`cosine_beta_schedule`(timesteps[, s, raise_to_power])	Cosine schedule as proposed in https://openreview.net/forum?id=-NEXDKk8gZ
`polynomial_schedule`(timesteps[, s, power])	A noise schedule based on a simple polynomial equation: 1 - x^power.

Module Contents¶

class MolecularDiffusion.modules.models.difflinker.noise.GammaNetwork¶

Bases: torch.nn.Module

Models a monotonic increasing function, construction as in the VDM paper.

forward(t)¶

gamma_tilde(t)¶

gamma_0¶

gamma_1¶

l1¶

l2¶

l3¶

class MolecularDiffusion.modules.models.difflinker.noise.PositiveLinear(in_features: int, out_features: int, bias: bool = True, weight_init_offset: int = -2)¶

Bases: torch.nn.Module

Linear layer with weights forced to be positive.

forward(x)¶

reset_parameters() → None¶

in_features¶

out_features¶

weight¶

weight_init_offset = -2¶

class MolecularDiffusion.modules.models.difflinker.noise.PredefinedNoiseSchedule(noise_schedule, timesteps, precision)¶

Bases: torch.nn.Module

Lookup array for predefined (non-learned) noise schedules.

forward(t)¶

gamma¶

timesteps¶

MolecularDiffusion.modules.models.difflinker.noise.clip_noise_schedule(alphas2, clip_value=0.001)¶: For a noise schedule given by alpha^2, clip alpha_t / alpha_t-1 to help improve stability during sampling.

MolecularDiffusion.modules.models.difflinker.noise.cosine_beta_schedule(timesteps, s=0.008, raise_to_power: float = 1)¶: Cosine schedule as proposed in https://openreview.net/forum?id=-NEXDKk8gZ

MolecularDiffusion.modules.models.difflinker.noise.polynomial_schedule(timesteps: int, s=0.0001, power=3.0)¶: A noise schedule based on a simple polynomial equation: 1 - x^power.