nnpd.nre.models

Interchangeable classifiers returning logits, not sigmoid probabilities.

Attributes

NETWORKS

Classes

MLP
ResidualBlock
ResidualMLP

Functions

activation(→ torch.nn.Module)
make_mlp(inputs, definition)
make_residual(inputs, definition)
build_network(→ torch.nn.Module)

Module Contents

nnpd.nre.models.activation(name: str) → torch.nn.Module

class nnpd.nre.models.MLP(inputs: int, hidden: list[int], nonlinearity: str = 'relu', dropout: float = 0)

Bases: torch.nn.Module

layers

forward(inputs: torch.Tensor) → torch.Tensor

class nnpd.nre.models.ResidualBlock(width: int, nonlinearity: str, dropout: float)

Bases: torch.nn.Module

body

activation

forward(x)

class nnpd.nre.models.ResidualMLP(inputs: int, width: int, blocks: int, nonlinearity: str, dropout: float)

Bases: torch.nn.Module

layers

forward(inputs)

nnpd.nre.models.make_mlp(inputs, definition)

nnpd.nre.models.make_residual(inputs, definition)

nnpd.nre.models.NETWORKS

nnpd.nre.models.build_network(inputs: int, definition: dict) → torch.nn.Module