From 9374b1f189060b12ae2af9abdbe51681b3401134 Mon Sep 17 00:00:00 2001
From: kohlhaasrebecca <rebecca.kohlhaas@outlook.com>
Date: Thu, 14 Nov 2024 10:20:41 +0100
Subject: [PATCH] First tests for Sobol calculation
---
.../surrogate_models/polynomial_chaos.py | 15 +++++----
tests/test_PolynomialChaosEmulator.py | 31 +++++++++++++++++++
2 files changed, 40 insertions(+), 6 deletions(-)
diff --git a/src/bayesvalidrox/surrogate_models/polynomial_chaos.py b/src/bayesvalidrox/surrogate_models/polynomial_chaos.py
index ba8e0de1c..afedbe764 100644
--- a/src/bayesvalidrox/surrogate_models/polynomial_chaos.py
+++ b/src/bayesvalidrox/surrogate_models/polynomial_chaos.py
@@ -868,6 +868,8 @@ class PCE(MetaModel):
pce_stds_b = {}
for b_i in range(self.n_bootstrap_itrs):
+ pce_means_b[b_i] = {}
+ pce_stds_b[b_i] = {}
_coeffs_dicts = self._coeffs_dict[f'b_{b_i + 1}'].items()
for output, coef_dict in _coeffs_dicts:
@@ -927,7 +929,7 @@ class PCE(MetaModel):
return pce_means, pce_stds
- def calculate_sobol(self):
+ def calculate_sobol(self, Y_train = None):
"""
Provides Sobol' indices as a sensitivity measure to infer the importance
of the input parameters. See Eq. 27 in [1] for more details. For the
@@ -945,7 +947,9 @@ class PCE(MetaModel):
Parameters
----------
- None.
+ Y_train: dict, optional
+ Trainings outputs. They are needed when used in combination with PCA.
+ The default is None
Returns
-------
@@ -1057,8 +1061,7 @@ class PCE(MetaModel):
# Compute the sobol indices according to Ref. 2
if self.dim_red_method.lower() == "pca":
- # TODO: Update so that it does not use the ExpDesign anymore!
- n_c_points = self.engine.ExpDesign.Y[output].shape[1]
+ n_c_points = Y_train[output].shape[1]
pca = self.pca[f"b_{b_i+1}"][output]
comp_pca = pca.components_
n_comp = comp_pca.shape[0]
@@ -1076,7 +1079,7 @@ class PCE(MetaModel):
for t_idx in range(n_c_points):
var_yt = np.var(
- self.engine.ExpDesign.Y[output][:, t_idx]
+ Y_train[output][:, t_idx]
)
term1, term2 = 0.0, 0.0
if var_yt != 0.0:
@@ -1103,7 +1106,7 @@ class PCE(MetaModel):
s_zi = total_sobol_array[par_idx]
for t_idx in range(n_c_points):
- var_yt = np.var(self.engine.ExpDesign.Y[output][:, t_idx])
+ var_yt = np.var(Y_train[output][:, t_idx])
term1, term2 = 0.0, 0.0
if var_yt != 0.0:
for i in range(n_comp):
diff --git a/tests/test_PolynomialChaosEmulator.py b/tests/test_PolynomialChaosEmulator.py
index b62bd09da..64858588d 100644
--- a/tests/test_PolynomialChaosEmulator.py
+++ b/tests/test_PolynomialChaosEmulator.py
@@ -842,3 +842,34 @@ def test_calculate_moments_pca() -> None:
#%% Test PCE.update_metamodel
# TODO: taken from engine
+
+
+#%% Test PCE.calculate_sobol
+
+def test_calculate_sobol():
+ """
+ Calculate Sobol' indices of a pce-surrogate
+ """
+ inp = Input()
+ inp.add_marginals()
+ inp.Marginals[0].dist_type = 'normal'
+ inp.Marginals[0].parameters = [0, 1]
+ mm = PCE(inp)
+ mm.fit([[0.2], [0.8]], {'Z': [[0.4, 0.4], [0.5, 0.6]]})
+ sobol, totalsobol = mm.calculate_sobol()
+ # TODO are there theory-related checks that could be applied here?
+
+
+def test_calculate_sobol_pca():
+ """
+ Calculate Sobol' indices of a pce-surrogate with PCA
+ """
+ inp = Input()
+ inp.add_marginals()
+ inp.Marginals[0].dist_type = 'normal'
+ inp.Marginals[0].parameters = [0, 1]
+ mm = PCE(inp)
+ mm.dim_red_method = 'pca'
+ mm.fit([[0.2], [0.8]], {'Z': [[0.4, 0.4], [0.5, 0.6]]})
+ sobol, totalsobol = mm.calculate_sobol({'Z': np.array([[0.4, 0.4], [0.5, 0.6]])})
+ # TODO are there theory-related checks that could be applied here?
--
GitLab