Config SettingsΒΆ

  • load_split_dataset (bool) : Whether to load split dataset. Defaults to False.

  • feature_type (str) : Generating Type of node features. Defaults to all_one. Range in ['all_one', 'all_zero', 'random', 'node_type'].

  • task (str) : Task name. Defaults to satisfiability. Range in ['satisfiability', 'maxsat', 'unsat_core'].

  • task_type (str) : Construction Type of graph. Defaults to lcg. If model in ['gin', 'gcn', 'gat'], task_type in ['lcg', 'vcg', 'vig', 'lig']. Otherwise task_type in ['lcg', 'aig'].

  • task_level (str) : Task level. Defaults to graph. Range in ['e2e_graph', 'e2e_node', 'e2e_link', 'lr_graph', 'lr_node', 'lr_link', 'graph'].

  • dataset_path (str) : Path to the dataset. Defaults ./dataset/satisfiability.

  • model_settings (dict) : Settings for the model.

    • model (str) : Model name. Defaults neurosat. Range in ['neurosat', 'neurocore', 'nlocalsat', 'gms', 'deepsat', 'querysat', 'satformer', 'gin', 'gcn', 'gat'].

    • input_size (int) : Input dimension of the feature. Defaults to 1.

    • hidden_size (int) : Hidden dimension of the embedding. Defaults to 128.

    • output_size (int) : Output dimension of the embedding. Defaults to 1.

    • loss (str) : Loss function. Defaults to binary_cross_entropy. Range in ['cross_entropy', 'binary_cross_entropy', 'mse', 'mae'].

    • num_fc (int) : Number of fully connected layers. Defaults to 3.

    • num_round (int) : Number of message-passing rounds. Defaults to 32.

    • dropout_ratio (float) : Dropout ratio. Defaults to 0.

    • sigmoid (bool) : Whether to use sigmoid. Defaults to True.

    • pooling (str) : Pooling type. Defaults to mean. Range in ['mean', 'max'].

  • optimizer (str) : Optimizer. Defaults adam. Range in ['adam', 'sgd', 'adagrad', 'adadelta', 'rmsprop'].

  • scheduler_settings (dict) : Settings for the scheduler(Optional).

    • scheduler (str) : Scheduler name. Defaults to ReduceLROnPlateau. Range in ['ReduceLROnPlateau', 'StepLR'].

    • patience (int) : Patience for the scheduler. Defaults to 10.

    • factor (float) : Factor for the scheduler. Defaults to 0.5.

    • mode (str) : Mode for the scheduler. Defaults to min.

  • valid_metric (str) : Validation metric. Defaults to acc. Range in ['acc', 'mse', 'mae', 'rmse'].

  • eval_metric (str) : Evaluation metric. Defaults to acc.

  • eval_step (int) : Evaluation step. Defaults to 1.

  • epochs (int) : The number of training epochs. Defaults to 100.

  • early_stop (bool) : Whether to early stop. Defaults to False.

  • lr (float) : Learning rate. Defaults to 1e-4.

  • weight_decay (float) : Weight decay. Defaults to 1e-10.

  • device (str) : Device for training. Defaults to cuda:6.

  • split_ratio (list) : Split ratio for train, validation, test. Defaults to [0.6, 0.2, 0.2].

  • batch_size (int) : Batch size. Defaults to 64.

  • save_model (str) : Path to save the model. Defaults to ./save_model/neurosat.pt.

  • tensorboard_dir (str) : Directory for TensorBoard logs. Defaults to ./tensorboard_run.

  • log settings (dict) : Log settings.

    • log_file (str) : Path to the log file. Defaults to ./log/neurosat.log.