pwtools.rbf.hyperopt.fit_opt

pwtools.rbf.hyperopt.fit_opt(points, values, method='de', what='pr', cv={'n_repeats': 1, 'n_splits': 5}, cv_kwds=None, opt_kwds={}, rbf_kwds={})

Optimize Rbf’s hyper-parameter \(p\) or both \((p,r)\).

Use a cross validation error metric or the direct fit error if cv is None. Uses FitError.

Note: While we do have some defaults for initial guess or bounds, depending on the optimizer, you are strongly advised to set your own in opt_kwds.

Parameters:

• points – see Rbf

• values – see Rbf

• method (str) –

  • ‘de’ : scipy.optimize.differential_evolution()

  • ’fmin’: scipy.optimize.fmin()

  • ’brute’: scipy.optimize.brute()

• what (str) –

  • ‘p’ : optimize only p, set fixed r in rbf_kwds in this case, else we’ll use Rbf’s default)

  • ’pr’ : optimize p and r

• cv – see FitError

• rbf_kwds (dict) – see FitError

• opt_kwds (dict) – kwds for the optimizer (see method)

• rbf_kwds – Constant params for Rbf

Returns:

rbfi – Rbf instance initialized with points, values and found optimal p (and r).

Return type:

Rbf