dotime.DoTime

class dotime.DoTime(config=None, seed=42, chain_prob=0.15, regime_switching_prob=0.15)

Bases: object

Prior distribution over temporal SCMs with interventions.

Main interface for generating synthetic causal time series data.

Parameters:

• config (dict[str, Any] | None)

• seed (int)

• chain_prob (float)

• regime_switching_prob (float)

__init__(config=None, seed=42, chain_prob=0.15, regime_switching_prob=0.15)

Parameters:

• config (Dict[str, Any], optional) – Configuration dictionary. If None, uses DEFAULT_CONFIG.

• seed (int) – Random seed for reproducibility.

• chain_prob (float) – Probability of generating a chain SCM (default 0.15).

• regime_switching_prob (float) – Probability of generating a regime-switching SCM (default 0.15).

Methods

__init__([config, seed, chain_prob, ...])
generate_dataset(n_scms[, T])	Generate a dataset of paired observational/interventional time series.
generate_pair([T])	Generate a pair of observational and interventional time series.
generate_regime_pair([T, num_regimes])	Generate a paired (obs, int) trajectory from a regime-switching SCM.
generate_training_tuples(n_scms[, T])	Generate training tuples for PFN training.
sample_scm()	Sample a temporal SCM from the prior.

sample_scm()

Sample a temporal SCM from the prior.

Distribution: - chain_prob: chain SCMs - regime_switching_prob: regime-switching SCMs - remaining: diverse nonlinear SCMs

Return type:

TemporalSCM

Returns:

TemporalSCM – Sampled temporal SCM (or compatible regime-switching SCM).

generate_pair(T=None)

Generate a pair of observational and interventional time series.

Parameters:

T (int, optional) – Length of time series. If None, uses config default.

Return type:

tuple[Tensor, Tensor, InterventionSpec, TemporalSCM]

Returns:

Tuple[torch.Tensor, torch.Tensor, InterventionSpec, TemporalSCM] – (X_obs, X_int, intervention_spec, scm)

generate_regime_pair(T=None, num_regimes=2)

Generate a paired (obs, int) trajectory from a regime-switching SCM.

Like generate_pair() but forces a regime-switching SCM with a fixed number of regimes (for the regime-density benchmark tiers).

Return type:

tuple[Tensor, Tensor, InterventionSpec, RegimeSwitchingTemporalSCM]

Parameters:

• T (int | None)

• num_regimes (int)

generate_dataset(n_scms, T=None)

Generate a dataset of paired observational/interventional time series.

Parameters:

• n_scms (int) – Number of SCMs to sample.

• T (int, optional) – Length of time series. If None, uses config default.

Return type:

list[tuple[Tensor, Tensor, InterventionSpec]]

Returns:

List[Tuple[torch.Tensor, torch.Tensor, InterventionSpec]] – List of (X_obs, X_int, intervention_spec) tuples.

generate_training_tuples(n_scms, T=None)

Generate training tuples for PFN training.

Format: (X_obs, targets, times, values, Y_int_tau)

Parameters:

• n_scms (int) – Number of SCMs to sample.

• T (int, optional) – Length of time series. If None, uses config default.

Return type:

list[tuple[Tensor, list[int], list[int], Any, Tensor]]

Returns:

List[Tuple[torch.Tensor, List[int], List[int], Any, torch.Tensor]] – Training tuples suitable for PFN training.