sfaira.estimators.EstimatorKerasEmbedding¶

class sfaira.estimators.EstimatorKerasEmbedding(data: typing.Union[anndata._core.anndata.AnnData, numpy.ndarray, sfaira.data.store.stores.single.StoreSingleFeatureSpace], model_dir: typing.Optional[str], model_id: typing.Optional[str], model_topology: sfaira.versions.topologies.class_interface.TopologyContainer, weights_md5: typing.Optional[str] = None, cache_path: str = 'cache/', adata_ids: sfaira.consts.adata_fields.AdataIds = <sfaira.consts.adata_fields.AdataIdsSfaira object>)¶

Estimator class for the embedding model.

Attributes

`model_type`
`obs`
`obs_eval`
`obs_test`
`obs_train`
`organism`
`using_store`
`model`

Methods

`compute_gradients_input`([batch_size, ...])
`evaluate`([batch_size, max_steps])	Evaluate the custom model on test data.
`evaluate_any`(idx[, batch_size, max_steps])	Evaluate the custom model on any local data.
`get_one_time_tf_dataset`(idx, mode[, ...])
`init_model`([clear_weight_cache, ...])	instantiate the model :return:
`load_pretrained_weights`()	Loads model weights from local directory or zenodo.
`load_weights_from_cache`()
`predict`([batch_size])	return the prediction of the model
`predict_embedding`([batch_size, variational])	return the prediction in the latent space (z_mean for variational models)
`save_weights_to_cache`()
`split_train_val_test`(val_split, test_split)	Split indices in store into train, valiation and test split.
`train`(optimizer, lr[, epochs, ...])	Train model.