diff --git a/NonLinearSolver/modelReduction/ANNUtils.h b/NonLinearSolver/modelReduction/ANNUtils.h
index a113ff1272c40f22c3b51c6731725f9ab140722b..1620de713e6dab6f5e329171d048980db497bf34 100644
--- a/NonLinearSolver/modelReduction/ANNUtils.h
+++ b/NonLinearSolver/modelReduction/ANNUtils.h
@@ -65,14 +65,15 @@ class RectifierActivationFunction: public activationFunction
{
protected:
double _fact;
+ double _offset;
public:
- RectifierActivationFunction(double f=10.): activationFunction(),_fact(f){}
+ RectifierActivationFunction(double f=10.): activationFunction(),_fact(f),_offset(0.){}
RectifierActivationFunction(const RectifierActivationFunction& src): activationFunction(src){}
virtual ~RectifierActivationFunction(){}
virtual activationFunction::functionType getType() const {return activationFunction::rectifier;};
virtual std::string getName() const;
- virtual double getVal(double x) const {return (log(1.+exp(_fact*x)))/_fact;};
- virtual double getReciprocalVal(double y) const {return (log(exp(_fact*y)-1.))/_fact;}; // inverse function
+ virtual double getVal(double x) const {return (_offset+log(1.+exp(_fact*x)))/_fact;};
+ virtual double getReciprocalVal(double y) const {return (log(exp(_fact*(y-_offset))-1.))/_fact;}; // inverse function
virtual double getDiff(double x) const {return exp(_fact*x)/(1.+exp(_fact*x));};
virtual activationFunction* clone() const {return new RectifierActivationFunction(*this);};
};
diff --git a/NonLinearSolver/modelReduction/DeepMaterialNetworks.cpp b/NonLinearSolver/modelReduction/DeepMaterialNetworks.cpp
index 84bd3862c8fe106609a3499e0831067c38b8a2a8..07fbc4265ce7d72423706d7cd8549eac24c5aecd 100644
--- a/NonLinearSolver/modelReduction/DeepMaterialNetworks.cpp
+++ b/NonLinearSolver/modelReduction/DeepMaterialNetworks.cpp
@@ -980,9 +980,14 @@ void BimaterialHomogenization::compute(const std::vector<fullMatrix<double> >& C
{
DCeffDnormal.resize(3);
}
- };
- computeBimaterial(Cin[0],Cin[1],f[0],f[1],normal,Ceff,
+ computeBimaterial(Cin[0],Cin[1],f[0],f[1],normal,Ceff,
stiff,&DCeffDcin[0],&DCeffDcin[1],&DCeffDf[0],&DCeffDf[1],&DCeffDnormal[0],&DCeffDnormal[1],&DCeffDnormal[2]);
+ }
+ else
+ {
+ computeBimaterial(Cin[0],Cin[1],f[0],f[1],normal,Ceff);
+ };
+
};
void LaminateHomogenization::compute(const fullMatrix<double>& C1, const fullMatrix<double>& C2, const fullMatrix<double>& C3,
@@ -1105,10 +1110,17 @@ void LaminateHomogenization::compute(const std::vector<fullMatrix<double> >& Cin
}
else if (numPhase == 2)
{
- computeBimaterial(Cin[0],Cin[1],f[0],f[1],normal,Ceff,
- stiff,&DCeffDcin[0],&DCeffDcin[1],
- &DCeffDf[0],&DCeffDf[1],&DCeffDnormal[0],
- &DCeffDnormal[1],&DCeffDnormal[2]);
+ if (stiff)
+ {
+ computeBimaterial(Cin[0],Cin[1],f[0],f[1],normal,Ceff,
+ stiff,&DCeffDcin[0],&DCeffDcin[1],
+ &DCeffDf[0],&DCeffDf[1],&DCeffDnormal[0],
+ &DCeffDnormal[1],&DCeffDnormal[2]);
+ }
+ else
+ {
+ computeBimaterial(Cin[0],Cin[1],f[0],f[1],normal,Ceff);
+ }
}
else
{
@@ -1173,7 +1185,14 @@ void LaminateHomogenization::compute(const std::vector<fullMatrix<double> >& Cin
};
}
// two phase
- computeBimaterial(CC[i],Cin[i+1],f0i,f1i,normal,CC[i+1],stiff,&A[i],&B[i],&F0[i],&F1[i],&N0[i],&N1[i],&N2[i]);
+ if (stiff)
+ {
+ computeBimaterial(CC[i],Cin[i+1],f0i,f1i,normal,CC[i+1],stiff,&A[i],&B[i],&F0[i],&F1[i],&N0[i],&N1[i],&N2[i]);
+ }
+ else
+ {
+ computeBimaterial(CC[i],Cin[i+1],f0i,f1i,normal,CC[i+1]);
+ }
};
//
Ceff = CC[numPhase-1];
@@ -1265,10 +1284,17 @@ void LaminateHomogenization::compute(const std::vector<fullMatrix<double> >& Cin
}
else if (numPhase == 2)
{
- computeBimaterial(Cin[0],Cin[1],f[0],f[1],normal[0],Ceff,
+ if (stiff)
+ {
+ computeBimaterial(Cin[0],Cin[1],f[0],f[1],normal[0],Ceff,
stiff,&DCeffDcin[0],&DCeffDcin[1],
&DCeffDf[0],&DCeffDf[1],&DCeffDnormal[0][0],
&DCeffDnormal[0][1],&DCeffDnormal[0][2]);
+ }
+ else
+ {
+ computeBimaterial(Cin[0],Cin[1],f[0],f[1],normal[0],Ceff);
+ }
}
else
{
@@ -1333,7 +1359,14 @@ void LaminateHomogenization::compute(const std::vector<fullMatrix<double> >& Cin
};
}
// two phase
- computeBimaterial(CC[i],Cin[i+1],f0i,f1i,normal[i],CC[i+1],stiff,&A[i],&B[i],&F0[i],&F1[i],&N0[i],&N1[i],&N2[i]);
+ if (stiff)
+ {
+ computeBimaterial(CC[i],Cin[i+1],f0i,f1i,normal[i],CC[i+1],stiff,&A[i],&B[i],&F0[i],&F1[i],&N0[i],&N1[i],&N2[i]);
+ }
+ else
+ {
+ computeBimaterial(CC[i],Cin[i+1],f0i,f1i,normal[i],CC[i+1]);
+ }
};
//
Ceff = CC[numPhase-1];
@@ -1696,7 +1729,8 @@ void TrainingDeepMaterialNetwork::train(double lr, int maxEpoch, std::string los
WBest = Wcur;
Wprev = Wcur;
g.resize(numDof,true);
- double costfuncPrev = 0.;
+ double costfuncPrev = evaluateTrainingSet(*lossEval);
+ Msg::Info("costfuncPrev = %e",costfuncPrev);
//OfflineData
int epoch = 0;
@@ -1705,6 +1739,7 @@ void TrainingDeepMaterialNetwork::train(double lr, int maxEpoch, std::string los
int Ntrain = _XTrain.size();
double lrCur = lr;
int numEpochIncrease = 0;
+ int numEpochLrMin = 0;
int numEpochDecrease = 0;
while (true)
{
@@ -1773,17 +1808,23 @@ void TrainingDeepMaterialNetwork::train(double lr, int maxEpoch, std::string los
double costfunc = evaluateTrainingSet(*lossEval);
if (numberBatches==Ntrain)
{
- if ((costfunc > costfuncPrev) && (epoch > 0))
+ if (costfunc > costfuncPrev)
{
numEpochIncrease ++;
if (numEpochIncrease == 1)
{
numEpochIncrease = 0;
- Msg::Info("reduce learning rate from %.5e to %.5e",lrCur,0.8*lrCur);
+ Msg::Info("costfunc = %e, reduce learning rate from %.5e to %.5e",costfunc,lrCur,0.8*lrCur);
lrCur *= 0.8;
if (lrCur < lrmin)
{
lrCur = lrmin;
+ numEpochLrMin ++;
+ if (numEpochLrMin>10)
+ {
+ Msg::Info("maximal number of iterations %d at minimal learning step is reached !!",numEpochLrMin);
+ break;
+ }
}
numEpochDecrease = 0;
Wcur = Wprev;
@@ -1844,8 +1885,10 @@ void TrainingDeepMaterialNetwork::train(double lr, int maxEpoch, std::string los
if (removeZeroContribution)
{
+
if ((epoch+1) %numStepRemove==0)
{
+ double costfuncPrev_pre = evaluateTrainingSet(*lossEval);
Msg::Info("tree removing");
bool removed = _T->removeLeavesWithZeroContribution(tolRemove);
if (removed)
@@ -1858,7 +1901,10 @@ void TrainingDeepMaterialNetwork::train(double lr, int maxEpoch, std::string los
Wprev = Wcur;
g.resize(numDof,true);
}
+ costfuncPrev = evaluateTrainingSet(*lossEval);
+ Msg::Info("pre-removing costfuncPrev = %e and after removing costfuncPrev = %e",costfuncPrev_pre,costfuncPrev);
}
+
}
epoch++;
@@ -3158,7 +3204,7 @@ void DeepMaterialNetwork::downscale(const STensor3& Fcur, STensor3& F, const Tr
double wChild = getWeight(node);
int childOrder = node->childOrder;
//
- double f = wChild/wParent;
+ //double f = wChild/wParent;
double fact = 1./wChild;
/*
diff --git a/NonLinearSolver/modelReduction/DeepMaterialNetworks.h b/NonLinearSolver/modelReduction/DeepMaterialNetworks.h
index bf1ead0059c58c8c25b700dc38b7c3769a9988a0..51ce1bf3d9116b9ca7f414b932f11ad09c04786c 100644
--- a/NonLinearSolver/modelReduction/DeepMaterialNetworks.h
+++ b/NonLinearSolver/modelReduction/DeepMaterialNetworks.h
@@ -142,10 +142,10 @@ class BimaterialHomogenization : public Homogenization
fullMatrix<double>& DNDn2) const;
virtual void computeBimaterial(const fullMatrix<double>& C1, const fullMatrix<double>& C2, double f1, double f2, const SVector3& normal,
fullMatrix<double>& Ceff,
- bool stiff,
- hyperFullMatrix* DCeffDC1, hyperFullMatrix* DCeffDC2,
- fullMatrix<double>* DCeffDf1, fullMatrix<double>* DCeffDf2,
- fullMatrix<double>* DCeffDnormal1,fullMatrix<double>* DCeffDnormal2, fullMatrix<double>* DCeffDnormal3) const;
+ bool stiff=false,
+ hyperFullMatrix* DCeffDC1=NULL, hyperFullMatrix* DCeffDC2=NULL,
+ fullMatrix<double>* DCeffDf1=NULL, fullMatrix<double>* DCeffDf2=NULL,
+ fullMatrix<double>* DCeffDnormal1=NULL,fullMatrix<double>* DCeffDnormal2=NULL, fullMatrix<double>* DCeffDnormal3=NULL) const;
#endif //SWIG
};
diff --git a/NonLinearSolver/modelReduction/Tree.cpp b/NonLinearSolver/modelReduction/Tree.cpp
index af87281f50b70c238cef90e39e9ba6ed117dad84..ed8d0a0e8ed4f125dac7a22c8804d6a51b0f1c53 100644
--- a/NonLinearSolver/modelReduction/Tree.cpp
+++ b/NonLinearSolver/modelReduction/Tree.cpp
@@ -244,7 +244,7 @@ void Tree::createSpecialTripleTree(int numNodes, int numDirVars, const std::stri
}
};
-void Tree::createRandomTreeSameDepthForNodes(int nbLeaves, int numChildMax, int leafPerNode, int numDirVars,const std::string affunc, bool randomChild)
+void Tree::createRandomTreeSameDepthForNodes(int nbLeaves, int numChildMax, int leafPerNode, int numDirVars,const std::string affunc, bool randomChild, bool randomMat)
{
auto makeGroup = [&](const std::vector<int>& input, std::vector<int>& outPut, bool first)
{
@@ -269,7 +269,7 @@ void Tree::createRandomTreeSameDepthForNodes(int nbLeaves, int numChildMax, int
if (first)
{
numChilds = leafPerNode;
- if (randomChild)
+ if (randomMat)
{
numChilds = rand() % (leafPerNode-1) + 2; // number takes a random value from 2 to leafPerNode
}
@@ -971,7 +971,7 @@ int Tree::getNumberOfNodes() const
return allNodes.size();
};
-bool Tree::removeZeroLeaves(double tol)
+bool Tree::removeZeroLeaves(double tol, bool iteration)
{
bool ok = false;
int maxInter = 100;
@@ -1122,6 +1122,11 @@ bool Tree::removeZeroLeaves(double tol)
Msg::Exit(0);
}
};
+
+ if (!iteration)
+ {
+ break;
+ }
};
return false;
};
@@ -1904,7 +1909,8 @@ void Tree::initialize(bool rand)
for (int i=0; i< allLeaves.size(); i++)
{
TreeNode* n = allLeaves[i];
- n->weight *= (1./toltalWeightLeaves);
+ double vv = (n->af->getVal(n->weight))/toltalWeightLeaves;
+ n->weight *= n->af->getReciprocalVal(vv);
// propagate data
double w = n->weight;
while (n->parent != NULL)
diff --git a/NonLinearSolver/modelReduction/Tree.h b/NonLinearSolver/modelReduction/Tree.h
index c3747732f16f9c2431a1d83e5bcf986dd424c6d0..e5d7abd6622e4ddd268951038980af00e6afd7ac 100644
--- a/NonLinearSolver/modelReduction/Tree.h
+++ b/NonLinearSolver/modelReduction/Tree.h
@@ -53,7 +53,7 @@ class Tree
void createPerfectTree(int maxDepth, int numChildRegular, int numChildLeaf, int numDirVars, const std::string affunc="relu", int lastRepeat=1);
void createMixedTree(int maxDepth, int numChildMax, int leafPerNode, int numDirVars,const std::string affunc="relu");
void createRandomTree(int nbLeaves, int numChildMax, int leafPerNode, int numDirVars,const std::string affunc="relu", bool randomSubdivion=true, bool randomChild=true, bool sameNbChildLayer=true);
- void createRandomTreeSameDepthForNodes(int nbLeaves, int numChildMax, int leafPerNode, int numDirVars,const std::string affunc="relu", bool randomChild=true);
+ void createRandomTreeSameDepthForNodes(int nbLeaves, int numChildMax, int leafPerNode, int numDirVars,const std::string affunc="relu", bool randomChild=true, bool randomMat=true);
void createSpecialBinaryTree(int numNodes, int numDirVars,const std::string affunc="relu");
void createSpecialTripleTree(int numNodes, int numDirVars,const std::string affunc="relu");
void printTree() const;
@@ -78,7 +78,7 @@ class Tree
void printNodeFraction() const;
bool removeLeavesWithZeroContribution(double tol=1e-6);
- bool removeZeroLeaves(double tol);
+ bool removeZeroLeaves(double tol, bool iteration=true);
bool treeMerging();
#ifndef SWIG