ssfortran.fit_many

ssfortran.fit_many(models, maxiter=500, m=10, factr=10000000.0, pgtol=1e-05, compute_cov=True, gradient='auto')

Fit independent models in parallel.

The library fits the models on OpenMP threads, with the GIL released.

Parameters:

modelssequence of StructuralModel or MappedModel

The models; each is left at its estimates.

maxiterint, optional

Maximum number of iterations.

mint, optional

Number of L-BFGS corrections.

factrfloat, optional

Relative reduction tolerance, in units of the machine epsilon.

pgtolfloat, optional

Projected gradient tolerance.

compute_covbool, optional

Compute standard errors.

gradient{“auto”, “analytic”, “numerical”}, optional

As for Model.fit().

Returns:

list of FitResults or None

One result per model; None where the likelihood could not be evaluated.

Raises:

TypeError

If a model is an MLEModel: Python callbacks cannot run on the library’s threads.

Notes

The library must be built with OpenMP (the default with CMake). For large models set OPENBLAS_NUM_THREADS=1 to avoid nested threading in BLAS.

Examples

>>> rng = np.random.default_rng(5)
>>> ys = [np.cumsum(rng.standard_normal(80)) for _ in range(4)]
>>> results = ss.fit_many([ss.StructuralModel(y, [ss.Irregular(), ss.Level()])
...                        for y in ys])
>>> [r.converged for r in results]
[True, True, True, True]