bartz.PredictKind

class bartz.PredictKind(*values)

Kind of output of Bart.predict.

mean = 'mean'

The posterior mean of the conditional mean, shape (m,) (or (k, m) for multivariate regression).

mean_samples = 'mean_samples'

Per-sample conditional mean, shape (num_chains * n_save, m) (or (num_chains * n_save, k, m)). For binary regression, this is the probit-transformed sum-of-trees.

outcome_samples = 'outcome_samples'

Samples of the outcome variable, shape (num_chains * n_save, m) (or (num_chains * n_save, k, m)). For binary regression, these are Bernoulli draws. For continuous regression, these are Gaussian draws with the posterior noise variance.

latent_samples = 'latent_samples'

Raw sum-of-trees values, shape (num_chains * n_save, m) (or (num_chains * n_save, k, m)).