fishy.models.deep.performer

Performer model for spectral classification.

Performers use a linear approximation of the attention mechanism via Random Fourier Features (FAVOR+), allowing them to scale to very long sequences.

References: 1. Choromanski, K., et al. (2020). Rethinking Attention with Performers. arXiv:2009.14794.

Classes

class fishy.models.deep.performer.FastAttention(dim, heads=8, nb_features=64)

Bases: Module

FAVOR+ Fast Attention implementation.

__init__(dim, heads=8, nb_features=64)

Initialize internal Module state, shared by both nn.Module and ScriptModule.

forward(q, k, v)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class fishy.models.deep.performer.Performer(input_dim: int, output_dim: int, hidden_dim: int = 128, num_layers: int = 4, dropout: float = 0.1, num_heads: int = 8, nb_features: int = 64, **kwargs)

Bases: Module

Performer model for spectral classification.

__init__(input_dim: int, output_dim: int, hidden_dim: int = 128, num_layers: int = 4, dropout: float = 0.1, num_heads: int = 8, nb_features: int = 64, **kwargs) → None

Initialize internal Module state, shared by both nn.Module and ScriptModule.

forward(x: Tensor) → Tensor

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class fishy.models.deep.performer.PerformerLayer(dim, heads=8, nb_features=64, dropout=0.1)

Bases: Module

__init__(dim, heads=8, nb_features=64, dropout=0.1)

Initialize internal Module state, shared by both nn.Module and ScriptModule.

forward(x: Tensor)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

Functions

fishy.models.deep.performer.softmax_kernel(data, projection_matrix, is_query, epsilon=1e-06)

Softmax kernel approximation for Performer.

s