sc_reconstruction.models.ReconAE#

class sc_reconstruction.models.ReconAE(input_dim, n_hidden, n_latent, distribution='normal', library_size_mode='none', learning_rate=0.001, reduce_lr_on_plateau=False, lr_factor=0.6, lr_patience=5, lr_threshold=0.001, lr_min=0.0, decoder_output_activation=None)[source]#

Bases: BaseReconstructionModel

Parameters:

  • input_dim (int)

  • n_hidden (list)

  • n_latent (int)

  • distribution (str)

  • library_size_mode (str)

  • learning_rate (float)

  • reduce_lr_on_plateau (bool)

  • lr_factor (float)

  • lr_patience (int)

  • lr_threshold (float)

  • lr_min (float)

  • decoder_output_activation (str | nn.Module | None)

__init__(input_dim, n_hidden, n_latent, distribution='normal', library_size_mode='none', learning_rate=0.001, reduce_lr_on_plateau=False, lr_factor=0.6, lr_patience=5, lr_threshold=0.001, lr_min=0.0, decoder_output_activation=None)[source]#
Parameters:

  • input_dim (int)

  • n_hidden (list)

  • n_latent (int)

  • distribution (str)

  • library_size_mode (str)

  • learning_rate (float)

  • reduce_lr_on_plateau (bool)

  • lr_factor (float)

  • lr_patience (int)

  • lr_threshold (float)

  • lr_min (float)

  • decoder_output_activation (str | nn.Module | None)

Methods

__init__(input_dim, n_hidden, n_latent[, ...])

get_latent_representation(X)

load(path[, map_location])

Load the model

predict(X)

Run inference in identifier level (e.g., cell line X drug X dose), and return a dictionary mapping each identifier to its predicted result.

predict_relu(X)

prepare([adata])

Perform any setup steps needed before training, such as data preprocessing or model initialization.

save(path)

Save the trained model.

train([datamodule])

Train the model.