smoother.models.reduction._spmultivi

Classes

`SPMULTIVAE`	Add spatial loss to the latent representation in the MULTIVAE model.
`SpatialMULTIVI`	Spatially-aware Multimodal Variational Autoencoder model.

Functions

get_reconstruction_loss_protein(y, py_[, ...])

Get the reconstruction loss for protein data.

Module Contents

smoother.models.reduction._spmultivi.get_reconstruction_loss_protein(y, py_, pro_batch_mask_minibatch=None): Get the reconstruction loss for protein data.

class smoother.models.reduction._spmultivi.SPMULTIVAE(spatial_loss: smoother.SpatialLoss | None = None, lambda_spatial_loss: float = 1, sp_loss_as_kl: bool = True, **model_kwargs)

Bases: scvi.module.MULTIVAE

Add spatial loss to the latent representation in the MULTIVAE model.

Parameters:

spatial_loss – Spatial loss to apply on the latent representation. If None, no spatial loss.
lambda_spatial_loss – Weight of the spatial loss.
sp_loss_as_kl – Whether to apply the spatial loss as a KL term (i.e. replacing the original Normal(0, 1) prior) or as a separate global term.

spatial_loss = None

l_sp_loss = 1

sp_loss_as_kl = True

diag_sp_inv_cov = None

loss(tensors, inference_outputs, generative_outputs, kl_weight: float = 1.0) → scvi.module.base.LossOutput: Computes the loss function for the model.

classmethod from_vae(vae_module: scvi.module.MULTIVAE, spatial_loss: smoother.SpatialLoss | None = None, lambda_spatial_loss: float = 1, sp_loss_as_kl: bool = True)

class smoother.models.reduction._spmultivi.SpatialMULTIVI(st_adata: anndata.AnnData, spatial_loss: smoother.SpatialLoss | None = None, lambda_spatial_loss: float = 1, sp_loss_as_kl: bool = True, **model_kwargs)

Bases: scvi.model.MULTIVI

Spatially-aware Multimodal Variational Autoencoder model.

_data_splitter_cls

_module_cls

module

_model_summary_string

init_params_

dr_logs

train(max_epochs: int = 500, lr: float = 0.0001, accelerator: str = 'auto', devices='auto', weight_decay: float = 0.001, eps: float = 1e-08, n_steps_kl_warmup=None, n_epochs_kl_warmup=50, adversarial_mixing=False, datasplitter_kwargs=None, plan_kwargs=None, **kwargs): Train the model without mini-batch.

classmethod from_sc_model(st_adata: mudata.MuData, sc_model: scvi.model.MULTIVI, spatial_loss: smoother.SpatialLoss | None = None, lambda_spatial_loss: float = 1, sp_loss_as_kl: bool = True, unfrozen: bool = False, freeze_dropout: bool = False, freeze_expression: bool = True, freeze_decoder_first_layer: bool = True, freeze_batchnorm_encoder: bool = True, freeze_batchnorm_decoder: bool = False, freeze_classifier: bool = True, **spmultivae_kwargs)

Alternate constructor for exploiting a pre-trained model on non-spatial data.

Note that because of the dropout layer, even though the new instance is initialized with the same parameters as the pre-trained model, new_instance.get_latent_representation() may not return the same latent representation as the pre-trained model.

Parameters:

st_adata – registed anndata object
sc_model – pretrained MULTIVI model