Heimdall.fe.Fe

Contents

Heimdall.fe.Fe#

class Heimdall.fe.Fe(data, vocab_size, embedding_parameters, d_embedding, pad_value=None, mask_value=None, drop_zeros=True, rng=0)[source]#

Bases: ABC

Abstraction for expression-based embedding.

Parameters:

  • adata – input AnnData-formatted dataset, with gene names in the .var dataframe.

  • vocab_size (int) – total number of possible expression tokens and special tokens

  • d_embedding (int) – dimensionality of embedding for each expression entity

  • data (CellRepresentation)

  • embedding_parameters (DictConfig)

  • pad_value (int)

  • mask_value (int)

  • drop_zeros (bool)

  • rng (int | Generator)

Attributes

adata

gene_panel_idx

Methods

load_from_cache(expression_embeddings)

Load processed values from cache.

prepare_embedding_parameters()

Replace config placeholders with values after preprocessing.

preprocess_embeddings([float_dtype])

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