stackedembeddings

Documentation

Overview

Package stackedembeddings provides convenient types to stack multiple word embedding representations by concatenating them. The concatenation is then followed by a linear layer. The latter has the double utility of being able to project the concatenated embeddings in a smaller dimension, and to further train the final words representation.

Index

• type Model
  • func (m *Model) Encode(words []string) []ag.Node
• type WordsEncoderProcessor

Types

type Model

type Model struct {
	nn.BaseModel
	WordsEncoders   []WordsEncoderProcessor
	ProjectionLayer *linear.Model
}

Model implements a stacked embeddings model. TODO: optional use of the projection layer? TODO: include an optional layer normalization?

func (*Model) Encode

func (m *Model) Encode(words []string) []ag.Node

Encode transforms a string sequence into an encoded representation.

type WordsEncoderProcessor

type WordsEncoderProcessor interface {
	nn.Model
	// Encode transforms a string sequence into an encoded representation.
	Encode([]string) []ag.Node
}

WordsEncoderProcessor extends an nn.Processor providing the Encode method to transform a string sequence into an encoded representation.