Heimdall.fg.PretrainedFg

Contents

Heimdall.fg.PretrainedFg#

class Heimdall.fg.PretrainedFg(data, embedding_parameters, embedding_filepath=None, **fg_kwargs)[source]#

Bases: Fg, ABC

Abstraction for pretrained `Fg`s that can be loaded from disk.

Parameters:

  • embedding_filepath (str | PathLike | None) – filepath from which to load pretrained embeddings

  • data (CellRepresentation)

  • embedding_parameters (OmegaConf)

Raises:

ValueError – if config.d_embedding is larger than embedding dimensionality given in filepath.

Attributes

Methods

load_embeddings()

Load the embeddings from disk and process into map.

preprocess_embeddings([float_dtype])

Preprocess gene embeddings and store them for use during model inference.
Contents