bootstrap_funcs

Contains functions for bootstrap calculations for use in Spark queries.

Functions

bootstrap_group_key	Return a hashable key that identifies identical bootstrap configs.
create_circularblock_func	Create the CircularBlock bootstrap function.
create_gumboot_func	Create the Gumboot bootstrap function.
create_shared_bootstrap_func	Create a single bootstrap UDF that evaluates multiple metrics per draw.
create_stationary_func	Create the Stationary bootstrap function.
partition_metrics_by_bootstrap	Split metrics into non-bootstrap and bootstrap-sharing groups.

bootstrap_group_key(metric: MetricsBasemodel) → tuple | None

Return a hashable key that identifies identical bootstrap configs.

Two metrics with the same key can share a single set of bootstrap samples. Returns None for metrics without a bootstrap configuration.

create_circularblock_func(model: MetricsBasemodel) → Callable

Create the CircularBlock bootstrap function.

If model.bootstrap.block_size is None, the block size is estimated using arch.bootstrap.optimal_block_length (b_cb column).

Not used by the aggregation pipeline. format.py routes every bootstrap group – including groups of one – through create_shared_bootstrap_func, so this is retained as the public Bootstrappers.*.func field default (frozen, see models/bootstrap.py), for the autodoc page, and as an independent reference implementation the equivalence tests compare against. It has no sample-size/mean/variance quality guards, unlike the shared path.

create_gumboot_func(model: MetricsBasemodel) → Callable

Create the Gumboot bootstrap function.

Not used by the aggregation pipeline. format.py routes every bootstrap group – including groups of one – through create_shared_bootstrap_func, so this is retained as the public Bootstrappers.*.func field default (frozen, see models/bootstrap.py), for the autodoc page, and as an independent reference implementation the equivalence tests compare against. It has no sample-size/mean/variance quality guards, unlike the shared path.

create_shared_bootstrap_func(metrics: List[MetricsBasemodel]) → Callable

Create a single bootstrap UDF that evaluates multiple metrics per draw.

All metrics in metrics must share the same bootstrap configuration (same class, reps, seed, block_size, quantiles, guards, and input fields).

The caller must pass the bootstrap’s sort_by field(s) as the last positional args of the UDF, which is what format.py does. How many there are is read off the same sort_by it read to build the arg list, rather than being handed over separately – one fact, one place.

Parameters:

metrics (List[MetricsBasemodel]) – Metrics sharing the same bootstrap config.

Returns:

Callable – UDF returning dict with per-metric quantiles or raw bootstrap arrays.

create_stationary_func(model: MetricsBasemodel) → Callable

Create the Stationary bootstrap function.

If model.bootstrap.block_size is None, the block size is estimated using arch.bootstrap.optimal_block_length (b_sb column).

Not used by the aggregation pipeline. format.py routes every bootstrap group – including groups of one – through create_shared_bootstrap_func, so this is retained as the public Bootstrappers.*.func field default (frozen, see models/bootstrap.py), for the autodoc page, and as an independent reference implementation the equivalence tests compare against. It has no sample-size/mean/variance quality guards, unlike the shared path.

partition_metrics_by_bootstrap(metrics: List[MetricsBasemodel]) → Tuple[List[MetricsBasemodel], Dict[tuple, List[MetricsBasemodel]]]

Split metrics into non-bootstrap and bootstrap-sharing groups.

Returns:

• no_boot (list) – Metrics without a bootstrap config.

• boot_groups (dict) – Mapping of group key → list of metrics that can share samples. Singleton groups (len==1) are included so callers can treat all bootstrap metrics uniformly.