environmentaltools.common.optimize_rbf_epsilon

environmentaltools.common.optimize_rbf_epsilon(coords, data, n_train, method='gaussian', smooth=0.5, eps0=1, optimizer='local', metric='rmse')[source]

Optimize epsilon and smooth parameters for RBF by minimizing validation error (RMSE or MAE). Allows local (SLSQP) or global (differential_evolution) optimization.

Parameters:

  • coords (np.ndarray) – Input coordinates (n_samples, n_features).

  • data (np.ndarray) – Target values (n_samples,).

  • n_train (int) – Number of samples for training (rest for validation).

  • method (str, optional) – RBF function type. Default ‘gaussian’.

  • smooth (float, optional) – Initial smooth value. Default 0.5.

  • eps0 (float, optional) – Initial epsilon value. Default 1.

  • optimizer (str, optional) – ‘local’ (SLSQP) or ‘global’ (differential_evolution).

  • metric (str, optional) – ‘rmse’ or ‘mae’.

Returns:

(epsilon_opt, smooth_opt)

Return type:

tuple