pypomp.maths.logmeanexp¶

pypomp.maths.logmeanexp(x: Any, axis: None = None, ignore_nan: bool = False) → float[source]¶

pypomp.maths.logmeanexp(x: Any, axis: int | tuple[int, ...], ignore_nan: bool = False) → ndarray

Compute the log of the mean likelihood from log-likelihood values.

Calculates log(mean(exp(x))) in a numerically stable way using the log-sum-exp trick. This is appropriate when the estimator is unbiased on the natural (probability) scale, e.g. for averaging particle filter log-likelihood estimates across replicates.

Parameters:

x (array-like) – Collection of log-likelihood values.
axis (int, tuple of int, or None, optional) – Axis or axes along which to compute the mean. If None (default), compute over the entire array.
ignore_nan (bool, optional) – If True, treat NaN entries as -inf (i.e. zero probability) before computing. Defaults to False.

Returns:

The log-mean-exp value. A scalar float when axis=None, otherwise a numpy.ndarray with the reduced dimension removed.

Return type:

float or np.ndarray