UNAGI.UNAGI_tool.UNAGI.setup_training

UNAGI.setup_training(task, dist, device=None, epoch_iter=10, epoch_initial=20, lr=1e-4, lr_dis=5e-4, beta=1, hidden_dim=256, latent_dim=64, graph_dim=1024, BATCHSIZE=512, max_iter=10, GPU=False, adversarial=True, GCN=True)[source]

Set up the training parameters and the model parameters.

Parameters:

  • task (str) – the name of this task. It is used to name the output folder.

  • dist (str) – the distribution of the single-cell data. Chosen from ‘ziln’ (zero-inflated log normal), ‘zinb’ (zero-inflated negative binomial), ‘zig’ (zero-inflated gamma), and ‘nb’ (negative binomial).

  • device (str) – the device to run the model. If GPU is enabled, the device should be specified. Default is None.

  • epoch_iter (int) – the number of epochs for the iterative training process. Default is 10.

  • epoch_initial (int) – the number of epochs for the inital iteration. Default is 20.

  • lr (float) – the learning rate of the VAE model. Default is 1e-4.

  • lr_dis (float) – the learning rate of the discriminator. Default is 5e-4.

  • beta (float) – the beta parameter of the beta-VAE. Default is 1.

  • hiddem_dim (int) – the hidden dimension of the VAE model. Default is 256.

  • latent_dim (int) – the latent dimension of the VAE model. Default is 64.

  • graph_dim (int) – the dimension of the GCN layer. Default is 1024.

  • BATCHSIZE (int) – the batch size for the model training. Default is 512.

  • max_iter (int) – the maximum number of iterations for the model training. Default is 10.

  • GPU (bool) – whether to use GPU for the model training. Default is False.