diff --git a/src/bayesvalidrox/surrogate_models/polynomial_chaos.py b/src/bayesvalidrox/surrogate_models/polynomial_chaos.py
index 6cf4c1c09b96b612e9f054be2adfa07734734405..860eb80f5cc6ea1c2e3670df07c222f819b0b464 100644
--- a/src/bayesvalidrox/surrogate_models/polynomial_chaos.py
+++ b/src/bayesvalidrox/surrogate_models/polynomial_chaos.py
@@ -31,7 +31,7 @@ from .orthogonal_matching_pursuit import OrthogonalMatchingPursuit
from .reg_fast_ard import RegressionFastARD
from .reg_fast_laplace import RegressionFastLaplace
-from .surrogate_models import MetaModel, _preprocessing_fit, _bootstrap
+from .surrogate_models import MetaModel, _preprocessing_fit, _bootstrap_fit
from .supplementary import corr_loocv_error, create_psi
@@ -115,6 +115,7 @@ class PCE(MetaModel):
self._pce_reg_method = pce_reg_method
self._pce_deg = pce_deg
self._pce_q_norm = pce_q_norm
+ self.regression_dict = {}
self._pce_reg_options = {}
# These three parameters belong to the 'error_model'.
@@ -134,7 +135,9 @@ class PCE(MetaModel):
self._coeffs_dict = None
self._q_norm_dict = None
self._deg_dict = None
-
+
+ # Initialize the regression options as a dictionary
+ self.set_regression_options()
def check_is_gaussian(self, _pce_reg_method, n_bootstrap_itrs) -> bool:
@@ -155,54 +158,57 @@ class PCE(MetaModel):
if _pce_reg_method.lower() == 'lars':
return False
return True
-
+
+
def set_regression_options(self) -> None:
"""
- Chooses the generic settings for the regression type.
- TODO: currently not useable!
+ Collects the generic settings for the regression in a dictionary.
+ This includes the regression objects and their arguments.
Returns
-------
None
"""
- match self._pce_reg_method.lower():
- case 'ols':
- self._pce_reg_options = {'fit_intercept':False}
- case 'brr':
- self._pce_reg_options = {'n_iter':1000, 'tol':1e-7,
+ # Collect all regression objects
+ self.regression_dict['ols'] = lm.LinearRegression
+ self.regression_dict['brr'] = lm.BayesianRidge
+ self.regression_dict['ard'] = lm.ARDRegression
+ self.regression_dict['fastard'] = RegressionFastARD
+ self.regression_dict['bcs'] = RegressionFastLaplace
+ self.regression_dict['lars'] = lm.LassoLarsCV
+ self.regression_dict['sgdr'] = lm.SGDRegressor
+ self.regression_dict['omp'] = OrthogonalMatchingPursuit
+ self.regression_dict['vbl'] = VBLinearRegression
+ self.regression_dict['ebl'] = EBLinearRegression
+
+ # Collect the corresponding options
+ self._pce_reg_options['ols'] = {'fit_intercept':False}
+ self._pce_reg_options['brr'] = {'n_iter':1000, 'tol':1e-7,
'fit_intercept':False,
#'compute_score':compute_score,
'alpha_1':1e-04, 'alpha_2':1e-04,
#'lambda_1':lambda_, 'lambda_2':lambda_
}
- case 'ard':
- self._pce_reg_options = {'fit_intercept':False,
+ self._pce_reg_options['ard'] = {'fit_intercept':False,
#'compute_score':compute_score,
'n_iter':1000, 'tol':0.0001,
'alpha_1':1e-3, 'alpha_2':1e-3,
#'lambda_1':lambda_, 'lambda_2':lambda_
}
- case 'fastard':
- self._pce_reg_options = {'fit_intercept':False,
+ self._pce_reg_options['fastard'] = {'fit_intercept':False,
'normalize':True,
#'compute_score':compute_score,
'n_iter':300, 'tol':1e-10}
- case 'bcs':
- self._pce_reg_options = {'fit_intercept':False,
+ self._pce_reg_options['bcs'] = {'fit_intercept':False,
#'bias_term':bias_term,
'n_iter':1000, 'tol':1e-7}
- case 'lars':
- self._pce_reg_options = {'fit_intercept':False}
- case 'sgdr':
- self._pce_reg_options = {'fit_intercept':False,
+ self._pce_reg_options['lars'] = {'fit_intercept':False}
+ self._pce_reg_options['sgdr'] = {'fit_intercept':False,
'max_iter':5000, 'tol':1e-7}
- case 'omp':
- self._pce_reg_options = {'fit_intercept':False}
- case 'vbl':
- self._pce_reg_options = {'fit_intercept':False}
- case 'ebl':
- self._pce_reg_options = {'optimizer':'em'}
+ self._pce_reg_options['omp'] = {'fit_intercept':False}
+ self._pce_reg_options['vbl'] = {'fit_intercept':False}
+ self._pce_reg_options['ebl'] = {'optimizer':'em'}
def build_metamodel(self) -> None:
@@ -228,7 +234,7 @@ class PCE(MetaModel):
self._clf_poly = self.auto_vivification()
self._LCerror = self.auto_vivification()
- self._deg_array = self.__select_degree(self.ndim, self.n_samples)
+ self._deg_array = self.__select_degree()#self.ndim, self.n_samples)
# Generate all basis indices
self._all_basis_indices = self.auto_vivification()
@@ -236,7 +242,7 @@ class PCE(MetaModel):
keys = self._all_basis_indices.keys()
if deg not in np.fromiter(keys, dtype=float):
# Generate the polynomial basis indices
- for qidx, q in enumerate(self._pce_q_norm):
+ for _, q in enumerate(self._pce_q_norm):
basis_indices = glexindex(start=0, stop=deg + 1,
dimensions=self.ndim,
cross_truncation=q,
@@ -245,7 +251,7 @@ class PCE(MetaModel):
@_preprocessing_fit
- @_bootstrap
+ @_bootstrap_fit
def fit(self, X: np.array, y: dict, parallel=False, verbose=False, b_i = 0):
"""
Fits the surrogate to the given data (samples X, outputs y).
@@ -355,7 +361,7 @@ class PCE(MetaModel):
# Calulate the cofficients of surrogate model
updated_out = self.regression(
- psi, y[:, i], basis_indices, reg_method='OLS',
+ psi, y[:, i], basis_indices,
sparsity=False
)
@@ -364,6 +370,7 @@ class PCE(MetaModel):
return final_out_dict
+
# -------------------------------------------------------------------------
def univ_basis_vals(self, samples, n_max=None):
"""
@@ -398,6 +405,7 @@ class PCE(MetaModel):
return univ_basis
+
# -------------------------------------------------------------------------
def regression(self, X, y, basis_indices, sparsity=True):
"""
@@ -423,72 +431,45 @@ class PCE(MetaModel):
Fitted estimator, spareMulti-Index, sparseX and coefficients.
"""
- # Set parameters
- bias_term = self.dim_red_method.lower() != 'pca'
- compute_score = bool(self.verbose)
-
- # inverse of the observed variance of the data
- if np.var(y) != 0:
- lambda_ = 1 / np.var(y)
- else:
- lambda_ = 1e-6
-
# Bayes sparse adaptive aPCE
- # TODO: give the parameters for this as dictionary in the future
_clf_poly = None
- # WARNING: Match-case writing is only available for py 3.10 and above!
reg_method = self._pce_reg_method.lower()
- if reg_method == 'ols':
- _clf_poly = lm.LinearRegression(fit_intercept=False)
- elif reg_method == 'brr':
- _clf_poly = lm.BayesianRidge(n_iter=1000, tol=1e-7,
- fit_intercept=False,
- # normalize=True,
- compute_score=compute_score,
- alpha_1=1e-04, alpha_2=1e-04,
- lambda_1=lambda_, lambda_2=lambda_)
- _clf_poly.converged = True
+ kwargs = self._pce_reg_options[reg_method]
+
+ # Add the last parameters
+ if reg_method in ['brr', 'ard', 'fastard']:
+ compute_score = bool(self.verbose)
+ kwargs['compute_score'] = compute_score
- elif reg_method == 'ard':
+ if reg_method in ['brr', 'ard']:
+ lambda_ = 1e-6
+ if np.var(y) != 0:
+ lambda_ = 1 / np.var(y)
+ kwargs['lambda_1'] = lambda_
+ kwargs['lambda_2'] = lambda_
+
+ if reg_method == 'bcs':
+ bias_term = self.dim_red_method.lower() != 'pca'
+ kwargs['bias_term'] = bias_term
+
+ # Handle any exceptions
+ if reg_method == 'ard':
if X.shape[0] < 2:
raise ValueError('Regression with ARD can only be performed for more than 2 samples')
- _clf_poly = lm.ARDRegression(fit_intercept=False,
- # normalize=True,
- compute_score=compute_score,
- n_iter=1000, tol=0.0001,
- alpha_1=1e-3, alpha_2=1e-3,
- lambda_1=lambda_, lambda_2=lambda_)
-
- elif reg_method == 'fastard':
- _clf_poly = RegressionFastARD(fit_intercept=False,
- normalize=True,
- compute_score=compute_score,
- n_iter=300, tol=1e-10)
-
- elif reg_method == 'bcs':
+ if reg_method == 'bcs':
+ # TODO: move this error message into RegressionFastLaplace
if X.shape[0] < 10:
raise ValueError('Regression with BCS can only be performed for more than 10 samples')
- _clf_poly = RegressionFastLaplace(fit_intercept=False,
- bias_term=bias_term,
- n_iter=1000, tol=1e-7)
-
- elif reg_method == 'lars':
+ if reg_method == 'lars':
if X.shape[0] < 10:
raise ValueError('Regression with LARS can only be performed for more than 5 samples')
- _clf_poly = lm.LassoLarsCV(fit_intercept=False)
-
- elif reg_method == 'sgdr':
- _clf_poly = lm.SGDRegressor(fit_intercept=False,
- max_iter=5000, tol=1e-7)
-
- elif reg_method == 'omp':
- _clf_poly = OrthogonalMatchingPursuit(fit_intercept=False)
- elif reg_method == 'vbl':
- _clf_poly = VBLinearRegression(fit_intercept=False)
+ # Init the regression object
+ _clf_poly = self.regression_dict[reg_method](**kwargs)
- elif reg_method == 'ebl':
- _clf_poly = EBLinearRegression(optimizer='em')
+ # Apply any other settings
+ if reg_method == 'brr':
+ _clf_poly.converged = True
# Do the fit
_clf_poly.fit(X, y)
@@ -517,6 +498,7 @@ class PCE(MetaModel):
return_out_dict['coeffs'] = coeffs
return return_out_dict
+
# --------------------------------------------------------------------------------------------------------
def adaptive_regression(self, X, y, varIdx, verbose=False):
"""
@@ -761,7 +743,7 @@ class PCE(MetaModel):
mean = np.empty((len(samples), len(values)))
std = np.empty((len(samples), len(values)))
idx = 0
- for in_key, InIdxValues in values.items():
+ for in_key, _ in values.items():
# Assemble Psi matrix
basis = self._basis_dict[f'b_{b_i + 1}'][output][in_key]
@@ -826,7 +808,7 @@ class PCE(MetaModel):
return mean_pred, std_pred
# -------------------------------------------------------------------------
- def __select_degree(self, ndim, n_samples):
+ def __select_degree(self):#, ndim, n_samples):
"""
Selects degree based on the number of samples and parameters in the
sequential design.
@@ -936,7 +918,7 @@ class PCE(MetaModel):
pce_mean = np.zeros((len(coef_dict)))
pce_var = np.zeros((len(coef_dict)))
- for index, values in coef_dict.items():
+ for index, _ in coef_dict.items():
idx = int(index.split('_')[1]) - 1
coeffs = self._coeffs_dict[f'b_{b_i + 1}'][output][index]
diff --git a/src/bayesvalidrox/surrogate_models/surrogate_models.py b/src/bayesvalidrox/surrogate_models/surrogate_models.py
index 5446c9968e72122ad8fc138e35193ddc94be27d3..133c420ad82787e5f424f0d613eaef620e63dcbb 100644
--- a/src/bayesvalidrox/surrogate_models/surrogate_models.py
+++ b/src/bayesvalidrox/surrogate_models/surrogate_models.py
@@ -115,7 +115,7 @@ def _preprocessing_fit(fit_function):
return decorator
-def _bootstrap(fit_function):
+def _bootstrap_fit(fit_function):
"""
Decorator that applies bootstrap iterations around the training of a
MetaModel.
@@ -293,7 +293,7 @@ class MetaModel:
# Other params
self.InputSpace = None
- self.out_names = None
+ self.out_names = []
self.n_samples = None
self.LCerror = None # TODO: only used in loocv-exploration?! Not assigned correctly in PCE class???
self.LOOCV_score_dict = None # TODO: generic function that calculates this score!?