multiple_inference.utils#

Conditional inference utilities.

Functions

expected_wasserstein_distance(mean, cov, ...)

Compute the expected Wasserstein distance.

holm_bonferroni_correction([filename, ...])

Get significant coefficients by performing a Holm-Bonferroni correction.

weighted_quantile(values, quantiles[, ...])

Compute weighted quantiles.
multiple_inference.utils.expected_wasserstein_distance(mean: Sequence[float], cov: ndarray, estimated_means: ndarray, sample_weight: ndarray | None = None, **kwargs: Any) → float[source]#

Compute the expected Wasserstein distance.

This loss function computes the Wasserstein distance between the observed means mean and the distribution of means you would expect to observe given the estimated population means estimated_means.

Parameters:

  • mean (Numeric1DArray) – (n,) array of conventional point estimates.

  • cov (np.ndarray) – (n, n) covariance matrix of conventional estimates.

  • estimated_means (np.ndarray) – (# samples, n) matrix of draws from a distribution of population means.

  • sample_weight (np.ndarray, optional) – (# samples,) array of sample weights for estimated_means. Defaults to None.

  • **kwargs (Any) – Keyword arguments for scipy.stats.wasserstein_distance.

Returns:

Loss.

Return type:

float

multiple_inference.utils.holm_bonferroni_correction(filename: str | None = None, results: LikelihoodModelResults | None = None, alpha: float = 0.05) → Series[source]#

Get significant coefficients by performing a Holm-Bonferroni correction.

Parameters:

  • filename (str, optional) – Name of the csv file with conventional estimates. Defaults to None.

  • results (LikelihoodModelResults, optional) – Results. Defaults to None.

  • alpha (float, optional) – Significance level. Defaults to .05.

Raises:

ValueError – You must specify either filename or results but not both.

Returns:

Dataframe indicating which coefficients are significant.

Return type:

pd.DataFrame

multiple_inference.utils.weighted_quantile(values: ndarray, quantiles: float | Sequence[float], sample_weight: ndarray | None = None, values_sorted: bool = False) → ndarray[source]#

Compute weighted quantiles.

Parameters:

  • values (np.ndarray) – (n,) array over which to compute quantiles.

  • quantiles (Union[float, Numeric1DArray]) – (k,) array of quantiles of interest.

  • sample_weight (np.ndarray, optional) – (n,) array of sample weights. Defaults to None.

  • values_sorted (bool, optional) – Indicates that values have been pre-sorted. Defaults to False.

Returns:

(k,) array of weighted quantiles.

Return type:

np.array

Acknowledgements:

Credit to Stackoverflow.