COSIE.configure.get_default_config

get_default_config()

Returns the hyperparameters configuration dictionary used to initialize and train the COSIE model.

Returns

configdict

A dictionary containing the following sections:

• GraphAutoencoder:

  • hidden_dim (list of int): Hidden layer dimensions for the graph autoencoder.

  • activations (str): Activation function used (default: ‘relu’).

• Prediction:

  • hidden_dim (list of int): List of hidden layer dimensions used in the dual-prediction module.

• training:

  • seed (int): Random seed for reproducibility.

  • start_dual_prediction (int): Epoch to start dual-prediction loss.

  • start_cross_section_integration (int): Epoch to start cross-section integration.

  • epoch (int): Total number of training epochs.

  • lr (float): Learning rate for optimizer.

  • gamma (float): Weight for entropy regularization in contrastive loss.

  • lambda1 (float): Weight for contrastive loss.

  • lambda2 (float): Weight for prediction loss.

  • lambda3 (float): Weight for triplet loss.

  • knn_neighbors_spatial (int): Number of neighbors in spatial graph construction.

  • knn_neighbors_feature (int): Number of neighbors in feature graph construction.

  • print_num (int): Interval (in epochs) to print training progress.