From f92f927d04c3f6e40188f403fbf1e59ab316f50d Mon Sep 17 00:00:00 2001
From: faridm69 <faridmohammadi69@gmail.com>
Date: Tue, 17 Nov 2020 17:54:38 +0100
Subject: [PATCH] [surrogate] speeded up the eval_metamodel function.
---
.../surrogate_models/RegressionFastARD.py | 2 ++
.../surrogate_models/surrogate_models.py | 26 ++++++++++++-------
.../Test_AnalyticalFunction.py | 14 +++++-----
3 files changed, 26 insertions(+), 16 deletions(-)
diff --git a/BayesValidRox/surrogate_models/RegressionFastARD.py b/BayesValidRox/surrogate_models/RegressionFastARD.py
index 08c2a6de1..2f40199e1 100755
--- a/BayesValidRox/surrogate_models/RegressionFastARD.py
+++ b/BayesValidRox/surrogate_models/RegressionFastARD.py
@@ -245,6 +245,8 @@ class RegressionFastARD(LinearModel,RegressorMixin):
start = np.argmax(proj)
active[start] = True
A[start] = XXd[start]/( proj[start] - var_y)
+
+
warning_flag = 0
scores_ = []
diff --git a/BayesValidRox/surrogate_models/surrogate_models.py b/BayesValidRox/surrogate_models/surrogate_models.py
index 9b819535f..a8bc70145 100644
--- a/BayesValidRox/surrogate_models/surrogate_models.py
+++ b/BayesValidRox/surrogate_models/surrogate_models.py
@@ -410,7 +410,7 @@ class Metamodel:
return univ_vals
#--------------------------------------------------------------------------------------------------------
- def PCE_create_Psi(self,BasisIndices,univ_p_val):
+ def PCE_create_Psi(self,BasisIndices,univ_p_val, active=None):
"""
This function assemble the design matrix Psi from the given basis index
set INDICES and the univariate polynomial evaluations univ_p_val.
@@ -429,7 +429,7 @@ class Metamodel:
# number of basis elements
P = BasisIndices.shape[0]
-
+
# check that the variables have consistent sizes
if NofPa != BasisIndices.shape[1]:
raise ValueError("The shapes of BasisIndices and univ_p_val don't match!!")
@@ -438,7 +438,8 @@ class Metamodel:
Psi = np.ones((NofSamples, P),dtype=np.float32)
# Assemble the Psi matrix
for m in range(BasisIndices.shape[1]):
- aa = np.where(BasisIndices[:,m]>0)[0]
+ aa = np.where(BasisIndices[:,m]>0)[0]
+ if active is not None: aa = list(set(aa).intersection(active)) # Take active terms
try:
basisIdx = BasisIndices[aa,m]
Psi[:,aa] = np.multiply(Psi[:,aa] , np.reshape(univ_p_val[:, m, basisIdx], Psi[:,aa].shape))
@@ -525,7 +526,7 @@ class Metamodel:
clf_poly = VBLinearRegression(fit_intercept=False)
elif RegMethod.lower() == 'ebl':
- clf_poly = EBLinearRegression(fit_intercept=False)
+ clf_poly = EBLinearRegression(fit_intercept=False,optimizer = 'em')
# Fit
clf_poly.fit(PSI, ModelOutput)
@@ -551,7 +552,6 @@ class Metamodel:
PolynomialDegrees_Sparse = BasisIndices.toarray() if sparsity else BasisIndices
PSI_Sparse = PSI
coeffs = clf_poly.coef_
-
# Raise an error, if all coeffs are zero
#raise ValueError('All the coefficients of the regression model are zero!')
@@ -633,8 +633,8 @@ class Metamodel:
score, LCerror = self.CorrectedLeaveoutError(sparsePSI, Coeffs, ModelOutput, clf_poly)
# Check the convergence of noise for FastARD
- # if self.RegMethod == 'FastARD' and clf_poly.alpha_ < np.finfo(np.float32).eps:
- # score = -np.inf
+ if self.RegMethod == 'FastARD' and clf_poly.alpha_ < np.finfo(np.float32).eps:
+ score = -np.inf
qNormScores[qidx] = score
qAllCoeffs[str(qidx+1)] = Coeffs
@@ -2178,7 +2178,15 @@ class Metamodel:
y_mean, y_std = gp.predict(self.Samples, return_std=True)
else: # Perdiction with PCE or pcekriging
- PSI_Val = self.PCE_create_Psi(self.BasisDict[Outkey][Inkey], univ_p_val)
+ # Find active terms
+ try:
+ active_terms = np.nonzero(self.clf_poly[Outkey][Inkey].coef_)[0]
+ except:
+ active_terms = np.nonzero(self.CoeffsDict[Outkey][Inkey].coef_)[0]
+
+ # Assemble Psi matrix
+ PSI_Val = self.PCE_create_Psi(self.BasisDict[Outkey][Inkey], univ_p_val,
+ active=active_terms)
# Perdiction
try: # with error bar
@@ -2194,7 +2202,7 @@ class Metamodel:
gp = self.gp_poly[Outkey][Inkey]
y_gp_mean, y_gp_std = gp.predict(self.Samples, return_std=True)
y_mean += y_gp_mean
- y_std = y_gp_std
+ y_std += y_gp_std
PCEOutputs_mean[:, idx] = y_mean
PCEOutputs_std[:, idx] = y_std
diff --git a/BayesValidRox/tests/AnalyticalFunction/Test_AnalyticalFunction.py b/BayesValidRox/tests/AnalyticalFunction/Test_AnalyticalFunction.py
index 776394a0d..d00381c51 100755
--- a/BayesValidRox/tests/AnalyticalFunction/Test_AnalyticalFunction.py
+++ b/BayesValidRox/tests/AnalyticalFunction/Test_AnalyticalFunction.py
@@ -112,13 +112,13 @@ if __name__ == "__main__":
# The degree with the lowest Leave-One-Out cross-validation (LOO)
# error (or the highest score=1-LOO)estimator is chosen as the final
# metamodel.
- MetaModelOpts.MinPceDegree = 1#1
- MetaModelOpts.MaxPceDegree = 12#12
+ MetaModelOpts.MinPceDegree = 1
+ MetaModelOpts.MaxPceDegree = 8 #8
# q-quasi-norm 0<q<1 (default=1)
- MetaModelOpts.q = 1.0 if ndim<5 else 0.75 #np.linspace(0.3,0.8, 3)
-
+ MetaModelOpts.q = np.linspace(0.3,0.8,3) if ndim<5 else 0.75
+
# Print summary of the regression results
- # MetaModelOpts.DisplayFlag = True
+ MetaModelOpts.DisplayFlag = True
# ------------------------------------------------
# ------ Experimental Design Configuration -------
@@ -129,7 +129,7 @@ if __name__ == "__main__":
# One-shot (normal) or Sequential Adaptive (sequential) Design
MetaModelOpts.ExpDesign.Method = 'normal'
- NrofInitSamples = 400 #5*ndim #50
+ NrofInitSamples = 350 #5*ndim #50
MetaModelOpts.ExpDesign.NrSamples = NrofInitSamples
# Sampling methods
@@ -245,7 +245,7 @@ if __name__ == "__main__":
# Select the inference method
BayesOpts.SamplingMethod = 'MCMC'
- BayesOpts.MCMCnwalkers = 200
+ BayesOpts.MCMCnwalkers = 300
# BayesOpts.MCMCinitSamples = np.zeros((1,10)) + 1e-3 * np.random.randn(200, 10)
# BayesOpts.MCMCnSteps = 1000
import emcee
--
GitLab