run

run(name='Cora', data_root='/tmp', no_features=False, model='VGAE', num_epochs=10000, patience=20, runs=10, cl_runs=5, dims: List[int] | None = None, hidden_multiplier=2, target_patch_degree=4.0, min_overlap: int | None = None, target_overlap: int | None = None, gamma=0.0, sparsify='resistance', train_directed=False, cluster='metis', num_clusters=10, beta=0.1, num_iters: int | None = None, lr=0.001, cl_model='logistic', cl_train_args={}, cl_model_args={}, dist=False, output='.', device: str | None = None, verbose_train=False, verbose_l2g=False, levels=1, resparsify=0, run_baseline=True, normalise=False, restrict_lcc=False, scale=False, rotate=True, translate=True, mmap_edges=False, mmap_features=False, random_split=True, use_tmp=False, cluster_init=False, use_gpu_frac=1.0, grid_search_params=True, progress_bars=True)[source]

Run training example.

By default this function writes results to the current working directory. To override this use the output argument.

Parameters:

  • name – Name of data set to load (one of {'Cora', 'PubMed', 'AMZ_computers', 'AMZ_photo'})

  • data_root – Directory to use for downloaded data

  • no_features – If True, discard features and use node identity.

  • model – embedding model type (one of {‘VGAE’, ‘GAE’, ‘DGI’})

  • num_epochs – Number of training epochs

  • patience – Patience for early stopping

  • runs – Number of training runs (keep best result)

  • dims – list of embedding dimensions (default: [2])

  • hidden_multiplier – Hidden dimension is hidden_multiplier * dim

  • target_patch_degree – Target patch degree for resistance sparsification.

  • min_overlap – Minimum target patch overlap (default: max(dims) + 1)

  • target_overlap – Target patch overlap (default: 2 * max(dims))

  • gamma – Value of ‘gamma’ for RMST sparsification

  • sparsify – Sparsification method to use (one of {'resistance', 'none', 'rmst'})

  • train_directed – Use the orignal directed network (only relevant for some loaders) (default: False)

  • cluster – Clustering method to use (one of {'louvain', 'fennel' , 'distributed', 'metis'})

  • num_clusters – Target number of clusters for distributed, fennel, or metis.

  • beta – Parameter for the distributed clustering algorithm

  • num_iters – Maximum iterations for distributed or fennel

  • lr – Learning rate

  • cl_model – the classification model to use (one of “logistic” or “mlp”) (default: “logistic”)

  • cl_train_args – extra arguments to pass down to the classification training (default: {})

  • cl_model_args – extra arguments to pass to the classfication model constructor (default: {})

  • dist – If True, use distance decoder instead of inner product decoder

  • verbose_l2g – Verbose output for the alignment step (default: False)

  • normalise – If True, normalise the dataset features (default: False)

  • restrict_lcc – If True, restrict the dataset to only consider largest connected component (default: False)

  • output – output folder

  • levels – number of hierarchical patch levels (default: 1)

  • resparsify – if > 0, use resistance sparsification for all levels of th hierarchy (default: 0)

  • scale – apply scaling transformations during alignment (default: False)

  • rotate – apply rotations during alignment (default: True)

  • translate – apply translations during alignment (default: True)

  • mmap_edges – use memory mapping for edges (only supported by some loaders) (default: False)

  • mmap_features – use memory mapping for features (only supported by some loaders) (default: False)

  • random_split – use random train-test splits for evaluation (default: True)

  • use_tmp – copy data to tmp dir during load (default: False)

  • cluster_init – run the cluster initialisation script (default: False)

  • use_gpu_frac – fraction of gpu to use by each worker (default: 1.0)

  • grid_search_params – use grid search for classification parameters (only for cl_model=’mlp’)

  • device – Device used for training e.g., ‘cpu’, ‘cuda’ (defaults to ‘cuda’ if available else ‘cpu’)

  • verbose_train – If True, show progress info

  • run_baseline – if True, run baseline full model

  • progress_bars – show progress bars (default: True)

This function is also exposed as a command-line interface.

References