`edsnlp.pipes.trainable.embeddings.doc_pooler.factory`

`create_component = registry.factory.register('eds.doc_pooler', assigns=[], deprecated=[])(DocPooler)` `module-attribute` [source]

Pools word embeddings over the entire document to produce a single embedding per doc.

PARAMETER	DESCRIPTION
`nlp`	The pipeline object TYPE: `Optional[Pipeline]` DEFAULT: `None`
`name`	Name of the component TYPE: `str` DEFAULT: `'document_pooler'`
`embedding`	The word embedding component TYPE: `WordEmbeddingComponent`
`pooling_mode`	How word embeddings are aggregated into a single embedding per document. TYPE: `Literal['max', 'sum', 'mean', 'cls', 'attention']` DEFAULT: `mean`
`hidden_size`	The size of the hidden layer. If None, no projection is done. TYPE: `Optional[int]`