netket.stats.online_statistics

netket.stats.online_statistics(data, old_estimator=None, *, decay=None, max_lag=64)

Overloads:

• data (LocalEstimatorsBatch), old_estimator (OnlineStatsBatch | None), decay (float | None), max_lag (int) → OnlineStatsBatch

• data (LocalEstimators’ | jax.typing.ArrayLike), old_estimator (OnlineStats | None), decay (float | None), max_lag (int) → OnlineStats

Parameters:

• data (LocalEstimatorsBatch | LocalEstimators | jax.typing.ArrayLike)

• old_estimator (OnlineStats | OnlineStatsBatch | None)

• decay (float | None)

• max_lag (int)

Return type:

OnlineStats | OnlineStatsBatch

Accumulate streaming MCMC statistics across batches.

This is the functional API for OnlineStats. Each call merges a new batch of data into the running estimator using the parallel Welford algorithm.

Parameters:

• data (Union[LocalEstimatorsBatch, LocalEstimators, Array, ndarray, bool, number, bool, int, float, complex]) – Array of shape (n_samples,) or (n_chains, n_samples_per_chain), or a LocalEstimators / LocalEstimatorsBatch wrapper.

• old_estimator (OnlineStats | OnlineStatsBatch | None) – Previous accumulator instance, or None to start fresh.

• decay (float | None) – EMA decay factor applied per call (default 1.0 = no decay). When decay < 1.0, old data is down-weighted exponentially and tau_corr is set to NaN.

• max_lag (int) – Maximum lag for the online ACF estimator (default 64). Set to 0 to disable ACF tracking.

Returns:

Updated OnlineStats or OnlineStatsBatch instance, depending on the input type.

Return type:

OnlineStats | OnlineStatsBatch

Example:

estimator = None
for batch in training_loop:
    estimator = nk.stats.online_statistics(batch, estimator)
stats = estimator.get_stats()