probinet.models.acd

probinet.models.acd#

Class definition of ACD, the algorithm to perform inference in networks with anomaly. The latent variables are related to community memberships and anomaly parameters [SDB22] .

Classes

AnomalyDetection([convergence_tol, ...])

Class definition of AnomalyDetection, the algorithm to perform inference and anomaly detection on networks with reciprocity.
class probinet.models.acd.AnomalyDetection(convergence_tol: float = 0.1, decision: int = 2, err: float = 0.01, err_max: float = 1e-08, num_realizations: int = 1, max_iter: int = 500, **kwargs)[source]#

Class definition of AnomalyDetection, the algorithm to perform inference and anomaly detection on networks with reciprocity.

compute_likelihood()[source]#

Compute the log-likelihood of the data.

This is an abstract method that must be implemented in each derived class.

fit(gdata: GraphData, ag: float = 1.5, bg: float = 10.0, pibr0: float | None = None, mupr0: float | None = None, flag_anomaly: bool = True, fix_pibr: bool = False, fix_mupr: bool = False, K: int = 3, undirected: bool = False, initialization: int = 0, assortative: bool = True, constrained: bool = False, fix_w: bool = False, fix_communities: bool = False, mask: ndarray | None = None, out_inference: bool = True, out_folder: Path = PosixPath('outputs'), end_file: str | None = None, files: PathLike | None = None, rng: Generator | None = None, **_AnomalyDetection__kwargs: Any) → Tuple[ndarray, ndarray, ndarray, float, float, float][source]#

Fit the AnomalyDetection models to the provided data.

Parameters:

  • gdata – Graph adjacency tensor.

  • ag – Shape of gamma prior, by default 1.5.

  • bg – Rate of gamma prior, by default 10.0.

  • pibr0 – Initial value for the anomaly parameter pi, by default None.

  • mupr0 – Initial value for the prior mu parameter, by default None.

  • flag_anomaly – If True, the anomaly detection is enabled, by default True.

  • fix_pibr – If True, the anomaly parameter pi is fixed, by default False.

  • fix_mupr – If True, the prior mu parameter is fixed, by default False.

  • K – Number of communities, by default 3.

  • undirected – If True, the graph is considered undirected, by default False.

  • initialization – Indicator for choosing how to initialize u, v, and w, by default 0.

  • assortative – If True, the network is considered assortative, by default True.

  • constrained – If True, constraints are applied on the updates, by default False.

  • fix_w – If True, the affinity tensor w is fixed, by default False.

  • fix_communities – If True, the community memberships are fixed, by default False.

  • mask – Mask for selecting the held-out set in the adjacency tensor in case of cross-validation, by default None.

  • out_inference – If True, evaluation inference results, by default True.

  • out_folder – Output folder for inference results, by default “outputs/”.

  • end_file – Suffix for the evaluation file, by default None.

  • files – Path to the file for initialization, by default None.

  • rng – Random number generator, by default None.

  • **kwargs – Additional parameters for the model.

Returns:

  • u_f – Final out-going membership matrix.

  • v_f – Final in-coming membership matrix.

  • w_f – Final affinity tensor.

  • pibr_f – Final anomaly parameter pi.

  • mupr_f – Final prior mu parameter.

  • maxL – Maximum log-likelihood.