cpa.CPAModule#
- class cpa.CPAModule(n_genes, n_perts, covars_encoder, drug_embeddings=None, n_latent=128, recon_loss='nb', doser_type='logsigm', n_hidden_encoder=256, n_layers_encoder=3, n_hidden_decoder=256, n_layers_decoder=3, n_hidden_doser=128, n_layers_doser=2, use_batch_norm_encoder=True, use_layer_norm_encoder=False, use_batch_norm_decoder=True, use_layer_norm_decoder=False, dropout_rate_encoder=0.0, dropout_rate_decoder=0.0, variational=False, seed=0)[source]#
CPA module using Gaussian/NegativeBinomial/Zero-InflatedNegativeBinomial Likelihood
- Parameters:
- n_genes int
Number of input genes
- n_perts int,
Number of total unique perturbations
- covars_encoder dict
- Dictionary of covariates with keys as each covariate name and values as
unique values of the corresponding covariate
- n_latent int
dimensionality of the latent space
- recon_loss str
Autoencoder loss (either “gauss”, “nb” or “zinb”)
- doser_type str
Type of dosage network (either “logsigm”, “sigm”, or “linear”)
- n_hidden_encoder int
Number of hidden units in encoder
- n_layers_encoder int
Number of layers in encoder
- n_hidden_decoder int
Number of hidden units in decoder
- n_layers_decoder int
Number of layers in decoder
- n_hidden_doser int
Number of hidden units in dosage network
- n_layers_doser int
Number of layers in dosage network
- use_batch_norm_encoder bool
Whether to use batch norm in encoder
- use_layer_norm_encoder bool
Whether to use layer norm in encoder
- use_batch_norm_decoder bool
Whether to use batch norm in decoder
- use_layer_norm_decoder bool
Whether to use layer norm in decoder
- dropout_rate_encoder float
Dropout rate in encoder
- dropout_rate_decoder float
Dropout rate in decoder
- variational bool
Whether to use variational inference
- seed int
Random seed
Methods
disentanglement(tensors, inference_outputs, ...)generative(z[, library])Run the generative model.
get_expression(tensors[, n_samples, ...])Computes gene expression means and std.
get_pert_embeddings(tensors, **inference_kwargs)inference(x, perts, perts_doses, covars_dict)Run the recognition model.
loss(tensors, inference_outputs, ...)Computes the reconstruction loss (AE) or the ELBO (VAE)
mixup_data(tensors[, alpha, opt])Returns mixed inputs, pairs of targets, and lambda
r2_metric(tensors, inference_outputs, ...[, ...])