iscan.utils.DataGenerator¶

class iscan.utils.DataGenerator(d: int, s0: int, graph_type: str, noise_std: float | numpy.ndarray | list[float] = 0.5, noise_type: str = 'Gaussian')¶

Generate synthetic data to evaluate shifted nodes and shifted edges

Defines the class of graphs and data to be generated.

Parameters:¶

d : int¶

Number of variables

s0 : int¶

Expected number of edges in the random graphs

graph_type : str¶

One of ["ER", "SF"]. ER and SF refer to Erdos-Renyi and Scale-free graphs, respectively.

noise_std : Union[float, np.ndarray, List[float]], optional¶

Sets the scale (variance) of the noise distributions. Should be either a scalar, or a list (array) of dim d. By default 1.

noise_type : str, optional¶

One of ["Gaussian", "Laplace", "Gumbel"], by default “Gaussian”.

Methods¶

_simulate_dag(→ numpy.ndarray)	Simulate random DAG with some expected number of edges.
_sample_GP(X[, lengthscale])
_choose_shifted_nodes(→ numpy.ndarray)	Randomly choose shifted nodes
_delete_edges(→ numpy.ndarray)	Generates a new adjacency matrix where some incoming edges (parents) are randomly deleted for the shifted nodes.
_sample_from_same_structs(→ Tuple[numpy.ndarray, ...)	Sample data where both DAGs have the same structure
_sample_from_diff_structs(→ Tuple[numpy.ndarray, ...)	Sample data where both DAGs have the different structures
sample(→ Tuple[numpy.ndarray, numpy.ndarray])	Samples two datasets from randomly generated DAGs