1 # -*- coding: utf-8 -*-
3 # Copyright (C) 2008-2019 EDF R&D
5 # This library is free software; you can redistribute it and/or
6 # modify it under the terms of the GNU Lesser General Public
7 # License as published by the Free Software Foundation; either
8 # version 2.1 of the License.
10 # This library is distributed in the hope that it will be useful,
11 # but WITHOUT ANY WARRANTY; without even the implied warranty of
12 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 # Lesser General Public License for more details.
15 # You should have received a copy of the GNU Lesser General Public
16 # License along with this library; if not, write to the Free Software
17 # Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19 # See http://www.salome-platform.org/ or email : webmaster.salome@opencascade.com
21 # Author: Jean-Philippe Argaud, jean-philippe.argaud@edf.fr, EDF R&D
24 from daCore import BasicObjects
25 import numpy, scipy.optimize, scipy.version
27 # ==============================================================================
28 class ElementaryAlgorithm(BasicObjects.Algorithm):
30 BasicObjects.Algorithm.__init__(self, "4DVAR")
31 self.defineRequiredParameter(
32 name = "ConstrainedBy",
33 default = "EstimateProjection",
35 message = "Prise en compte des contraintes",
36 listval = ["EstimateProjection"],
38 self.defineRequiredParameter(
39 name = "EstimationOf",
42 message = "Estimation d'etat ou de parametres",
43 listval = ["State", "Parameters"],
45 self.defineRequiredParameter(
49 message = "Minimiseur utilisé",
50 listval = ["LBFGSB","TNC", "CG", "NCG", "BFGS"],
52 self.defineRequiredParameter(
53 name = "MaximumNumberOfSteps",
56 message = "Nombre maximal de pas d'optimisation",
59 self.defineRequiredParameter(
60 name = "CostDecrementTolerance",
63 message = "Diminution relative minimale du coût lors de l'arrêt",
66 self.defineRequiredParameter(
67 name = "ProjectedGradientTolerance",
70 message = "Maximum des composantes du gradient projeté lors de l'arrêt",
73 self.defineRequiredParameter(
74 name = "GradientNormTolerance",
77 message = "Maximum des composantes du gradient lors de l'arrêt",
80 self.defineRequiredParameter(
81 name = "StoreInternalVariables",
84 message = "Stockage des variables internes ou intermédiaires du calcul",
86 self.defineRequiredParameter(
87 name = "StoreSupplementaryCalculations",
90 message = "Liste de calculs supplémentaires à stocker et/ou effectuer",
95 "CostFunctionJAtCurrentOptimum",
97 "CostFunctionJbAtCurrentOptimum",
99 "CostFunctionJoAtCurrentOptimum",
105 self.defineRequiredParameter( # Pas de type
107 message = "Liste des valeurs de bornes",
109 self.requireInputArguments(
110 mandatory= ("Xb", "Y", "HO", "EM", "R", "B" ),
111 optional = ("U", "CM"),
114 def run(self, Xb=None, Y=None, U=None, HO=None, EM=None, CM=None, R=None, B=None, Q=None, Parameters=None):
115 self._pre_run(Parameters, Xb, Y, R, B, Q)
117 # Correction pour pallier a un bug de TNC sur le retour du Minimum
118 if "Minimizer" in self._parameters and self._parameters["Minimizer"] == "TNC":
119 self.setParameterValue("StoreInternalVariables",True)
123 Hm = HO["Direct"].appliedControledFormTo
125 Mm = EM["Direct"].appliedControledFormTo
127 if CM is not None and "Tangent" in CM and U is not None:
128 Cm = CM["Tangent"].asMatrix(Xb)
134 if hasattr(U,"store") and 1<=_step<len(U) :
135 _Un = numpy.asmatrix(numpy.ravel( U[_step] )).T
136 elif hasattr(U,"store") and len(U)==1:
137 _Un = numpy.asmatrix(numpy.ravel( U[0] )).T
139 _Un = numpy.asmatrix(numpy.ravel( U )).T
144 if Cm is not None and _un is not None: # Attention : si Cm est aussi dans M, doublon !
145 _Cm = Cm.reshape(_xn.size,_un.size) # ADAO & check shape
151 # Remarque : les observations sont exploitées à partir du pas de temps
152 # numéro 1, et sont utilisées dans Yo comme rangées selon ces indices.
153 # Donc le pas 0 n'est pas utilisé puisque la première étape commence
154 # avec l'observation du pas 1.
156 # Nombre de pas identique au nombre de pas d'observations
157 # -------------------------------------------------------
158 if hasattr(Y,"stepnumber"):
159 duration = Y.stepnumber()
163 # Précalcul des inversions de B et R
164 # ----------------------------------
168 # Définition de la fonction-coût
169 # ------------------------------
170 self.DirectCalculation = [None,] # Le pas 0 n'est pas observé
171 self.DirectInnovation = [None,] # Le pas 0 n'est pas observé
173 _X = numpy.asmatrix(numpy.ravel( x )).T
174 if self._parameters["StoreInternalVariables"] or \
175 self._toStore("CurrentState") or \
176 self._toStore("CurrentOptimum"):
177 self.StoredVariables["CurrentState"].store( _X )
178 Jb = 0.5 * (_X - Xb).T * BI * (_X - Xb)
179 self.DirectCalculation = [None,]
180 self.DirectInnovation = [None,]
183 for step in range(0,duration-1):
184 self.DirectCalculation.append( _Xn )
185 if hasattr(Y,"store"):
186 _Ynpu = numpy.asmatrix(numpy.ravel( Y[step+1] )).T
188 _Ynpu = numpy.asmatrix(numpy.ravel( Y )).T
192 if self._parameters["EstimationOf"] == "State":
193 _Xn = Mm( (_Xn, _Un) ) + CmUn(_Xn, _Un)
194 elif self._parameters["EstimationOf"] == "Parameters":
197 if self._parameters["Bounds"] is not None and self._parameters["ConstrainedBy"] == "EstimateProjection":
198 _Xn = numpy.max(numpy.hstack((_Xn,numpy.asmatrix(self._parameters["Bounds"])[:,0])),axis=1)
199 _Xn = numpy.min(numpy.hstack((_Xn,numpy.asmatrix(self._parameters["Bounds"])[:,1])),axis=1)
201 # Etape de différence aux observations
202 if self._parameters["EstimationOf"] == "State":
203 _YmHMX = _Ynpu - numpy.asmatrix(numpy.ravel( Hm( (_Xn, None) ) )).T
204 elif self._parameters["EstimationOf"] == "Parameters":
205 _YmHMX = _Ynpu - numpy.asmatrix(numpy.ravel( Hm( (_Xn, _Un) ) )).T - CmUn(_Xn, _Un)
206 self.DirectInnovation.append( _YmHMX )
207 # Ajout dans la fonctionnelle d'observation
208 Jo = Jo + _YmHMX.T * RI * _YmHMX
210 J = float( Jb ) + float( Jo )
211 self.StoredVariables["CostFunctionJb"].store( Jb )
212 self.StoredVariables["CostFunctionJo"].store( Jo )
213 self.StoredVariables["CostFunctionJ" ].store( J )
214 if self._toStore("IndexOfOptimum") or \
215 self._toStore("CurrentOptimum") or \
216 self._toStore("CostFunctionJAtCurrentOptimum") or \
217 self._toStore("CostFunctionJbAtCurrentOptimum") or \
218 self._toStore("CostFunctionJoAtCurrentOptimum"):
219 IndexMin = numpy.argmin( self.StoredVariables["CostFunctionJ"][nbPreviousSteps:] ) + nbPreviousSteps
220 if self._toStore("IndexOfOptimum"):
221 self.StoredVariables["IndexOfOptimum"].store( IndexMin )
222 if self._toStore("CurrentOptimum"):
223 self.StoredVariables["CurrentOptimum"].store( self.StoredVariables["CurrentState"][IndexMin] )
224 if self._toStore("CostFunctionJAtCurrentOptimum"):
225 self.StoredVariables["CostFunctionJAtCurrentOptimum" ].store( self.StoredVariables["CostFunctionJ" ][IndexMin] )
226 if self._toStore("CostFunctionJbAtCurrentOptimum"):
227 self.StoredVariables["CostFunctionJbAtCurrentOptimum"].store( self.StoredVariables["CostFunctionJb"][IndexMin] )
228 if self._toStore("CostFunctionJoAtCurrentOptimum"):
229 self.StoredVariables["CostFunctionJoAtCurrentOptimum"].store( self.StoredVariables["CostFunctionJo"][IndexMin] )
232 def GradientOfCostFunction(x):
233 _X = numpy.asmatrix(numpy.ravel( x )).T
234 GradJb = BI * (_X - Xb)
236 for step in range(duration-1,0,-1):
237 # Etape de récupération du dernier stockage de l'évolution
238 _Xn = self.DirectCalculation.pop()
239 # Etape de récupération du dernier stockage de l'innovation
240 _YmHMX = self.DirectInnovation.pop()
241 # Calcul des adjoints
242 Ha = HO["Adjoint"].asMatrix(ValueForMethodForm = _Xn)
243 Ha = Ha.reshape(_Xn.size,_YmHMX.size) # ADAO & check shape
244 Ma = EM["Adjoint"].asMatrix(ValueForMethodForm = _Xn)
245 Ma = Ma.reshape(_Xn.size,_Xn.size) # ADAO & check shape
246 # Calcul du gradient par etat adjoint
247 GradJo = GradJo + Ha * RI * _YmHMX # Equivaut pour Ha lineaire à : Ha( (_Xn, RI * _YmHMX) )
248 GradJo = Ma * GradJo # Equivaut pour Ma lineaire à : Ma( (_Xn, GradJo) )
249 GradJ = numpy.asmatrix( numpy.ravel( GradJb ) - numpy.ravel( GradJo ) ).T
252 # Point de démarrage de l'optimisation : Xini = Xb
253 # ------------------------------------
254 if isinstance(Xb, type(numpy.matrix([]))):
255 Xini = Xb.A1.tolist()
259 # Minimisation de la fonctionnelle
260 # --------------------------------
261 nbPreviousSteps = self.StoredVariables["CostFunctionJ"].stepnumber()
263 if self._parameters["Minimizer"] == "LBFGSB":
264 # Minimum, J_optimal, Informations = scipy.optimize.fmin_l_bfgs_b(
265 if "0.19" <= scipy.version.version <= "1.1.0":
266 import lbfgsbhlt as optimiseur
268 import scipy.optimize as optimiseur
269 Minimum, J_optimal, Informations = optimiseur.fmin_l_bfgs_b(
272 fprime = GradientOfCostFunction,
274 bounds = self._parameters["Bounds"],
275 maxfun = self._parameters["MaximumNumberOfSteps"]-1,
276 factr = self._parameters["CostDecrementTolerance"]*1.e14,
277 pgtol = self._parameters["ProjectedGradientTolerance"],
278 iprint = self._parameters["optiprint"],
280 nfeval = Informations['funcalls']
281 rc = Informations['warnflag']
282 elif self._parameters["Minimizer"] == "TNC":
283 Minimum, nfeval, rc = scipy.optimize.fmin_tnc(
286 fprime = GradientOfCostFunction,
288 bounds = self._parameters["Bounds"],
289 maxfun = self._parameters["MaximumNumberOfSteps"],
290 pgtol = self._parameters["ProjectedGradientTolerance"],
291 ftol = self._parameters["CostDecrementTolerance"],
292 messages = self._parameters["optmessages"],
294 elif self._parameters["Minimizer"] == "CG":
295 Minimum, fopt, nfeval, grad_calls, rc = scipy.optimize.fmin_cg(
298 fprime = GradientOfCostFunction,
300 maxiter = self._parameters["MaximumNumberOfSteps"],
301 gtol = self._parameters["GradientNormTolerance"],
302 disp = self._parameters["optdisp"],
305 elif self._parameters["Minimizer"] == "NCG":
306 Minimum, fopt, nfeval, grad_calls, hcalls, rc = scipy.optimize.fmin_ncg(
309 fprime = GradientOfCostFunction,
311 maxiter = self._parameters["MaximumNumberOfSteps"],
312 avextol = self._parameters["CostDecrementTolerance"],
313 disp = self._parameters["optdisp"],
316 elif self._parameters["Minimizer"] == "BFGS":
317 Minimum, fopt, gopt, Hopt, nfeval, grad_calls, rc = scipy.optimize.fmin_bfgs(
320 fprime = GradientOfCostFunction,
322 maxiter = self._parameters["MaximumNumberOfSteps"],
323 gtol = self._parameters["GradientNormTolerance"],
324 disp = self._parameters["optdisp"],
328 raise ValueError("Error in Minimizer name: %s"%self._parameters["Minimizer"])
330 IndexMin = numpy.argmin( self.StoredVariables["CostFunctionJ"][nbPreviousSteps:] ) + nbPreviousSteps
331 MinJ = self.StoredVariables["CostFunctionJ"][IndexMin]
333 # Correction pour pallier a un bug de TNC sur le retour du Minimum
334 # ----------------------------------------------------------------
335 if self._parameters["StoreInternalVariables"] or self._toStore("CurrentState"):
336 Minimum = self.StoredVariables["CurrentState"][IndexMin]
338 # Obtention de l'analyse
339 # ----------------------
340 Xa = numpy.asmatrix(numpy.ravel( Minimum )).T
342 self.StoredVariables["Analysis"].store( Xa.A1 )
344 # Calculs et/ou stockages supplémentaires
345 # ---------------------------------------
346 if self._toStore("BMA"):
347 self.StoredVariables["BMA"].store( numpy.ravel(Xb) - numpy.ravel(Xa) )
352 # ==============================================================================
353 if __name__ == "__main__":
354 print('\n AUTODIAGNOSTIC\n')