preprocess.BatchReference

preprocess.BatchReference(
    features,
    grand_mean,
    beta,
    var_pooled,
    covariates,
    covariate_columns,
    g_bar,
    t2,
    a_prior,
    b_prior,
    reference_batches,
    n_reference_samples,
)

Frozen global parameters. The artefact that makes results stable.

grand_mean, beta and var_pooled are per feature; g_bar, t2, a_prior and b_prior are the empirical-Bayes hyperparameters. The hyperparameters are frozen as deliberately as the rest: leave them free and adding a batch re-estimates the prior, which moves every other batch’s shrunk estimate, which is the drift this class exists to remove.

Attributes

Name	Description
digest	Content hash, for the run manifest.