Class Embedding

Inheritance Relationships

Base Types

Class Documentation

class Embedding : public marian::LayerBase, public marian::IEmbeddingLayer

A regular embedding layer.

Note that this also applies dropout if the option is passed (pass 0 when in inference mode). It is best to not use Embedding directly, but rather via getEmbeddingLayer() in EncoderDecoderLayerBase, which knows to pass on all required parameters from options.

Public Functions

Embedding(Ptr<ExpressionGraph> graph, Ptr<Options> options)

Construct a regular embedding layer in the graph.

Parameters

  • graph: The expression graph.

  • options: The options used for this embedding layer.

std::tuple<Expr, Expr> apply(Ptr<data::SubBatch> subBatch) const

Apply/Link this embedding layer (with the given batch of sentences) to the expression graph.

Return

The expression tuple holding the embedding layer and the masking layer

Parameters

  • subBatch: The batch of sentences

Expr apply(const Words &words, const Shape &shape) const

Apply/Link this embedding layer (with the given words and shape) to the expression graph.

Return

The expression holding the embedding layer

Parameters

  • words: Sequence of vocabulary items

  • shape: Shape of the words

Expr applyIndices(const std::vector<WordIndex> &embIdx, const Shape &shape) const

Apply/Link this embedding layer (with the given WordIndex vector and shape) to the expression graph.

Return

The expression holding the embedding layer

Parameters

  • embIdx: The vector of WordIndex objects

  • shape: Shape of the WordIndex vector