/// sequentially read nxn array in row major order
void sequentialRead()
{
MDArray *array = new MDArray(fileName);
PagedStorageContainer::resetPerfCounts();
gettimeofday(&tim, NULL);
t1 = tim.tv_sec + tim.tv_usec/1000000.0;
MDIterator *it = array->createIterator(Dense, row);
while (it->moveNext())
{
it->get(coord, datum);
}
delete it;
delete array;
gettimeofday(&tim, NULL);
t2 = tim.tv_sec + tim.tv_usec/1000000.0;
readCount += PagedStorageContainer::readCount;
writeCount += PagedStorageContainer::writeCount;
accessTime += PagedStorageContainer::accessTime;
execTime += (t2 - t1);
fprintf(pFile, "%-14s %8i %8i %11.3f %11.3f\n", "seq read", PagedStorageContainer::readCount, PagedStorageContainer::writeCount, PagedStorageContainer::accessTime, t2-t1);
}
void StringFunction::operator()(MDArray& input_phy_points, MDArray& output_values,
const stk::mesh::Bucket& bucket, const MDArray& parametric_coordinates, double time_value_optional)
{
PRINT("tmp srk StringFunction::operator(bucket) getName()= " << getName() << " input_phy_points= " << input_phy_points << " output_values= " << output_values);
VERIFY_OP(input_phy_points.rank(), ==, 3, "StringFunction::operator() must pass in input_phy_points(numCells, numPointsPerCell, spaceDim)");
int nPoints = input_phy_points.dimension(1);
int nSpaceDim = input_phy_points.dimension(2);
int nOutDim = output_values.dimension(2);
MDArray input_phy_points_one(1, nPoints, nSpaceDim);
MDArray output_values_one (1, nPoints, nOutDim);
const unsigned num_elements_in_bucket = bucket.size();
for (unsigned iElement = 0; iElement < num_elements_in_bucket; iElement++)
{
stk::mesh::Entity& element = bucket[iElement];
for (int iPoint = 0; iPoint<nPoints; iPoint++)
{
for (int iSpaceDim=0; iSpaceDim < nSpaceDim; iSpaceDim++)
{
input_phy_points_one(0, iPoint, iSpaceDim) = input_phy_points(iElement, iPoint, iSpaceDim);
}
}
(*this)(input_phy_points_one, output_values_one, element, parametric_coordinates, time_value_optional);
for (int iPoint = 0; iPoint<nPoints; iPoint++)
{
for (int iDOF = 0; iDOF < nOutDim; iDOF++)
{
output_values(iElement, iPoint, iDOF) = output_values_one(0, iPoint, iDOF);
}
}
}
}
/// strided insertion, using the iterator
void stridedIteratorInsert()
{
MDArray *array;
Btree::MSplitter<Datum_t> leafSp;
Btree::MSplitter<PID_t> intSp;
if (type == DMA)
array = new MDArray(dim, type, row, fileName);
else if (type == BTREE)
array = new MDArray(dim, row, &leafSp, &intSp, fileName);
PagedStorageContainer::resetPerfCounts();
gettimeofday(&tim, NULL);
t1 = tim.tv_sec + tim.tv_usec/1000000.0;
MDIterator *it = array->createIterator(Dense, col);
while (it->moveNext())
{
it->put(12345);
}
delete it;
delete array;
gettimeofday(&tim, NULL);
t2 = tim.tv_sec + tim.tv_usec/1000000.0;
readCount += PagedStorageContainer::readCount;
writeCount += PagedStorageContainer::writeCount;
accessTime += PagedStorageContainer::accessTime;
execTime += (t2 - t1);
fprintf(pFile, "%-14s %8i %8i %11.3f %11.3f\n", "batch stride", PagedStorageContainer::readCount, PagedStorageContainer::writeCount, PagedStorageContainer::accessTime, t2-t1);
}
/// strided insertion into nxn array (insert going down columns first into row
//major array
void stridedInsert()
{
MDArray *array;
Btree::MSplitter<Datum_t> leafSp;
Btree::MSplitter<PID_t> intSp;
if (type == DMA)
array = new MDArray(dim, type, row, fileName);
else if (type == BTREE)
array = new MDArray(dim, row, &leafSp, &intSp, fileName);
PagedStorageContainer::resetPerfCounts();
gettimeofday(&tim, NULL);
t1 = tim.tv_sec + tim.tv_usec/1000000.0;
for (i64 j = 0; j < n; j++)
{
for (i64 i = 0; i < n; i++)
{
coord = MDCoord(2, i, j);
array->put(coord, 12345);
}
}
delete array;
gettimeofday(&tim, NULL);
t2 = tim.tv_sec + tim.tv_usec/1000000.0;
readCount += PagedStorageContainer::readCount;
writeCount += PagedStorageContainer::writeCount;
accessTime += PagedStorageContainer::accessTime;
execTime += (t2 - t1);
fprintf(pFile, "%-14s %8i %8i %11.3f %11.3f\n", "stride insert", PagedStorageContainer::readCount, PagedStorageContainer::writeCount, PagedStorageContainer::accessTime, t2-t1);
}
/// Dimensions of parametric_coordinates and input_phy_points are required to be ([P],[D])
/// output_values: ([P], [DOF])
void StringFunction::operator()(MDArray& input_phy_points, MDArray& output_values,
const stk::mesh::Entity& element, const MDArray& parametric_coordinates, double time_value_optional)
{
PRINT("tmp srk StringFunction::operator(element) getName()= " << getName() << " input_phy_points= " << input_phy_points << " output_values= " << output_values);
argsAreValid(input_phy_points, output_values);
argsAreValid(parametric_coordinates, output_values);
m_element = &element;
m_have_element = true;
VERIFY_OP(parametric_coordinates.rank(), ==, 2, "StringFunction::operator() parametric_coordinates rank bad");
m_parametric_coordinates = MDArray(1, parametric_coordinates.dimension(1));
int nPoints=parametric_coordinates.dimension(0);
int spaceDim = parametric_coordinates.dimension(1);
for (int iPoint = 0; iPoint < nPoints; iPoint++)
{
for (int iSpace=0; iSpace < spaceDim; iSpace++)
m_parametric_coordinates(0, iSpace)=parametric_coordinates(iPoint, iSpace);
(*this)(input_phy_points, output_values, time_value_optional);
}
// reset this else we won't be able to reuse this object correctly
m_have_element = false;
m_element = 0;
// * Dimensions of parametric_coordinates are required to be ([P],[D])
//FieldFunction:: void operator()(const stk::mesh::Entity *element, const MDArray& parametric_coordinates, MDArray& out);
}
/// random read from nxn array. all indices are read once
void randomRead()
{
i64 k[n*n];
for (int i = 0; i < n*n; i++)
{
k[i] = i;
}
MDArray *array = new MDArray(fileName);
PagedStorageContainer::resetPerfCounts();
permute(k, n*n);
gettimeofday(&tim, NULL);
t1 = tim.tv_sec + tim.tv_usec/1000000.0;
for (int i = 0; i < n*n; i++)
{
coord = MDCoord(2, k[i]/n, k[i]%n);
array->get(coord, datum);
}
delete array;
gettimeofday(&tim, NULL);
t2 = tim.tv_sec + tim.tv_usec/1000000.0;
readCount += PagedStorageContainer::readCount;
writeCount += PagedStorageContainer::writeCount;
accessTime += PagedStorageContainer::accessTime;
execTime += (t2 - t1);
fprintf(pFile, "%-14s %8i %8i %11.3f %11.3f\n", "rand read", PagedStorageContainer::readCount, PagedStorageContainer::writeCount, PagedStorageContainer::accessTime, t2-t1);
}
void MultinomialDensities::estimate(vector<MDArray<double> > & ess)
{
assert(!ess.empty());
MDArray<double> e = ess[0];
e.add_inplace(pseudo_count);
set_parameters(e);
}
TEST(MDArray, Read)
{
i64 rows = 20L;
i64 cols = 20L;
MDCoord dim(2, rows, cols);
StorageType type = DMA;
i64 blockDims[] = {10L, 10L};
u8 blockOrders[] = {0, 1};
u8 microOrders[] = {0, 1};
//permute(blockOrders, dim.nDim);
//permute(microOrders, dim.nDim);
Linearization *block = new BlockBased(dim.nDim, dim.coords, blockDims,
blockOrders, microOrders);
const char *fileName = "test1.bin";
MDCoord key;
Datum_t datum = 1.0;
MDArray *array = new MDArray(dim, type, block, fileName);
MDIterator *it = array->createIterator(Dense, block);
srand(time(NULL));
const int num=100;
std::map<int, MDCoord> keys;
std::map<int, Datum_t> data;
int total = rows*cols;
for(int i=0; i<num; i++) {
int k = rand() % total;
if (data.find(k) == data.end()) {
data[k] = rand();
it->setIndexRange(k, k+1);
it->moveNext();
it->put(data[k]);
MDCoord key;
it->get(key, datum);
keys[k] = key;
}
else
i--;
}
delete it;
delete array;
array = new MDArray(fileName);
Datum_t datum1;
for (std::map<int,MDCoord>::iterator s = keys.begin();
s != keys.end();
++s) {
//cout<<"checking "<<s->second.toString()<<endl;
array->get(s->second, datum1);
ASSERT_DOUBLE_EQ(data[s->first], datum1);
}
delete array;
delete block;
}
int main(int argc, char **argv)
{
const int required = 2;
char fileName[100] = "/riot/mb";
unsigned int tm;
if (argc >= required+1)
tm = atoi(argv[required]);
else
tm = time(NULL);
srand(tm);
cerr<<"seed = "<<tm<<endl;
if (argc < required) {
cerr<<"Usage: "<<argv[0]<<"<splitter type>"<<endl
<<"splitter type: M,A,D"<<endl;
return 0;
}
char splitterType = argv[1][0];
i64 arraydims[] = {20000,20000};
i64 blockdims[] = {31,31};
u8 orders[] = {0, 1};
Linearization<2> *lin = new BlockBased<2>(arraydims, blockdims, orders,
orders);
MDArray<2> *array;
StorageParam sp;
sp.fileName = fileName;
sp.intSp = 'M';
sp.leafSp = splitterType;
if (splitterType == 'D') {
// dma
sp.type = DMA;
}
else if (splitterType == 'M') {
sp.type = BTREE;
sp.useDenseLeaf = false;
}
else if (splitterType == 'A') {
sp.type = BTREE;
sp.useDenseLeaf = true;
}
else {
cerr<<"wrong splitter type"<<endl;
exit(1);
}
array = new MDArray<2>(&sp, MDCoord<2>(arraydims), lin);
for (i64 i=0; i<20000; ++i) {
for (i64 j=0; j<20000; ++j) {
MDCoord<2> c(i,j);
array->put(c, 1.0);
}
}
delete lin;
delete array;
}
void ParentMap::set(uint h, MDArray<double> & v) {
vector<uint> & v1 = indices[h];
uint i = v1.back();
assert(v.get_shape().size() == 1);
uint dim = v.get_shape().front();
uint k = 0;
for (uint j = i; j < i + dim; j++, k++) {
*(seq[j]) = v[k];
}
}
MDArray<double> Shrinkage::combine(MDArray<double> target,MDArray<double> MLE,double lambda)
{
assert(target.get_shape() == MLE.get_shape());
// assert(lambda>=0 && lambda<=1);
lambda = std::max(lambda, 0.0);
lambda = std::min(lambda, 1.0);
return add_matrices(mult_num_to_matrix(target,lambda),mult_num_to_matrix(MLE,1-lambda));
}
Shrinkage::Shrinkage(MDArray<double> data, MDArray<double> MLE)
{
N=data.get_shape()[0];
P=data.get_shape()[1];
assert(MLE.get_shape()[0]==P && MLE.get_shape()[1]==P);
this->data=data;
wk=wk_StDM=vec(N,P,P);
S_MLE=vec(P,P);
StDM=vec(N,P);
set_S_MLE(MLE);
}
void FieldFunction::operator()(MDArray& input_phy_points, MDArray& output_field_values,
const stk_classic::mesh::Bucket& bucket, const MDArray& parametric_coordinates, double time_value_optional)
{
EXCEPTWATCH;
#ifndef NDEBUG
int num_elements_in_bucket = bucket.size();
VERIFY_OP(input_phy_points.dimension(0), ==, num_elements_in_bucket, "FieldFunction::operator() mismatch in input_phy_points and num_elements_in_bucket");
VERIFY_OP(output_field_values.dimension(0), ==, num_elements_in_bucket, "FieldFunction::operator() mismatch in input_phy_points and num_elements_in_bucket");
#endif
helper(input_phy_points, output_field_values, bucket, parametric_coordinates, time_value_optional);
}
void printMDArray(const MDArray<nDim> &array)
{
MDCoord<nDim> dim = array.getDims();
Datum_t datum;
for (int i=0; i<dim[0]; i++) {
for (int j=0; j<dim[1]; j++) {
MDCoord<2> c(i,j);
array.get(c, datum);
cout<<datum<<"\t";
}
cout<<endl;
}
}
void IntrepidSideCell<MDArray>::mapToCellPhysicalFrame(
const MDArray& parametric_coords, MDArray& physical_coords )
{
DTK_REQUIRE( 2 == parametric_coords.rank() );
DTK_REQUIRE( 3 == physical_coords.rank() );
DTK_REQUIRE( parametric_coords.dimension(1) ==
Teuchos::as<int>(this->d_topology.getDimension()) );
DTK_REQUIRE( physical_coords.dimension(0) ==
this->d_cell_node_coords.dimension(0) );
DTK_REQUIRE( physical_coords.dimension(1) ==
parametric_coords.dimension(0) );
DTK_REQUIRE( physical_coords.dimension(2) ==
Teuchos::as<int>(d_parent_topology.getDimension()) );
MDArray mapped_coords( parametric_coords.dimension(0),
d_parent_topology.getDimension() );
Intrepid::CellTools<Scalar>::mapToReferenceSubcell(
mapped_coords, parametric_coords, this->d_topology.getDimension(),
d_side_id, d_parent_topology );
Intrepid::CellTools<Scalar>::mapToPhysicalFrame(
physical_coords, mapped_coords,
this->d_cell_node_coords, d_parent_topology );
}
TEST(Gen, GenerateData)
{
i64 rows = 5L;
i64 cols = 5L;
MDCoord dim(2, rows, cols);
StorageType type = DMA;
i64 blockDims[] = {3L, 2L};
u8 blockOrders[] = {1, 0};
u8 microOrders[] = {0, 1};
//permute(blockOrders, dim.nDim);
//permute(microOrders, dim.nDim);
Linearization *block = new BlockBased(dim.nDim, dim.coords, blockDims,
blockOrders, microOrders);
//Linearization *block = new BlockBased(dim.nDim, dim.coords, blockDims,
// blockOrders, microOrders);
const char *fileName = "b.bin";
MDCoord key;
Datum_t datum = 0.0;
MDArray *array = new MDArray(dim, type, block, fileName);
MDIterator *it = array->createIterator(Dense, block);
while(it->moveNext()) {
it->put(datum);
datum += 1;
}
delete it;
delete array;
array = new MDArray(fileName);
it = array->createIterator(Dense, block);
while(it->moveNext()) {
it->get(key, datum);
cout<<"key "<<key.toString()<<"\t datum "<<datum<<endl;
}
delete it;
delete array;
delete block;
}
void checkRect(const MDArray<2> &array, const MDCoord<2> &begin, const MDCoord<2> &end, const Datum_t *data)
{
int k = 0;
Datum_t datum;
for (i64 j=begin[1]; j<=end[1]; ++j)
for (i64 i=begin[0]; i<=end[0]; ++i) {
array.get(MDCoord<2>(i,j), datum);
ASSERT_EQ(data[k++], datum);
}
}
Shrinkage::Shrinkage(MDArray<double> data)
{
N=data.get_shape()[0];
P=data.get_shape()[1];
this->data=data;
wk=wk_StDM=vec(N,P,P);
S_MLE=vec(P,P);
StDM=vec(N,P);
generate_MLE();
}
/// random insertion into nxn array, all indices are written to once
void randomInsert()
{
i64 k[n*n];
int stride = 5;
int num = n*n/stride;
for (int i = 0; i < num; i++)
{
k[i] = i*stride;
}
MDArray *array;
Btree::MSplitter<Datum_t> leafSp;
Btree::MSplitter<PID_t> intSp;
if (type == DMA)
array = new MDArray(dim, type, row, fileName);
else if (type == BTREE)
array = new MDArray(dim, row, &leafSp, &intSp, fileName);
PagedStorageContainer::resetPerfCounts();
permute(k, num);
i64 *i = k;
gettimeofday(&tim, NULL);
t1 = tim.tv_sec + tim.tv_usec/1000000.0;
while (i < k + num)
{
coord = MDCoord(2, *i/n, *i%n);
array->put(coord, 12345);
i++;
}
delete array;
gettimeofday(&tim, NULL);
t2 = tim.tv_sec + tim.tv_usec/1000000.0;
readCount += PagedStorageContainer::readCount;
writeCount += PagedStorageContainer::writeCount;
accessTime += PagedStorageContainer::accessTime;
execTime += (t2 - t1);
fprintf(pFile, "sparsity %4.3f\n", 1.0/stride);
fprintf(pFile, "%-14s %8i %8i %11.3f %11.3f\n", "rand insert", PagedStorageContainer::readCount, PagedStorageContainer::writeCount, PagedStorageContainer::accessTime, t2-t1);
}
void MultinomialDensities::set_parameters(MDArray<double> & cpd)
{
// Check cpd is right shape
vector<uint> shape = cpd.get_shape();
assert(shape[1] == dim);
this->cpd = cpd;
this->cpd.normalize();
// make new MultinomialDensity objects
_create_densities();
}
static inline void first_dimensions(MDArray& arr, int arr_offset, int *n_points, int max_rank=3)
{
for (int ii = 0; ii < max_rank; ii++)
{
n_points[ii] = 1;
}
for (int ii = 0; ii < arr_offset; ii++)
{
n_points[ii] = arr.dimension(ii);
}
}
void Shrinkage::generate_wk(MDArray<double> mat,int SDM_type)
{
int n=mat.get_shape()[0];
int p=mat.get_shape()[1];
vector<uint> index;
double sum,term1,term2;
MDArray<double> sub_means(get_sub_means(mat));
for(int i=0;i<p;i++)
for(int j=0;j<p;j++){
sum=0;
for(int k=0;k<n;k++){
term1=mat.get(k,i)-sub_means[i];
term2=mat.get(k,j)-sub_means[j];
index=vec((uint)k, (uint)i, (uint)j);
if(SDM_type)
wk_StDM[index]=term1*term2;
else
wk[index]=term1*term2;
}
}
}
void PoissonSolver::solve(MDArray<Real>& a_Rhs) const
{
//create a contructor
Box bx = a_Rhs.getBox();
MDArray<complex<double> > fftwForward(bx);
MDArray<complex<double> > fourierCoef(bx);
double h = 1/m_M;
for (int k = 0; k < bx.sizeOf();k++)
{
fftwForward[k] = complex<double>(a_Rhs[k],0.);
}
FFTMD fftmd = FFTMD(m_fft1dptr);
fftmd.forwardCC(fftwForward);
for (int i= 0; i<m_M; i++)
{
for (int j= 0; j<m_M; j++)
{
int index[2] ={i,j};
if(i ==0 && j ==0)
{
fourierCoef[index] =complex<double>(0,0);
}
else
{
complex <double> div(2*cos(2*M_PI*i*h) - 2*cos(2*M_PI*j*h - 4),0);
fourierCoef[index] = fftwForward[index]/div;
}
}
}
fftmd.inverseCC(fourierCoef);
for (int k= 0; k < bx.sizeOf(); k++)
{
a_Rhs[k] = real(fourierCoef[k])/m_N;
}
}
void ForwardBacktracker::backtrack_pass(uint start, uint end, uint seq_len, MDArray<eMISMASK> & mismask, RandomGen* rg) {
uint choice = 0;
MDArray<double> transition_matrix(vec(hidden_node_size));
// The backtrack loop
for(int l=end-1; l >= ((int) start); l--) {
uint i = l - start;
DiscreteNode *node;
// Setup the node, cpd and nodes
if(i==0) {
node = hd_0;
}
else {
node = hd_1;
}
// Choose whether to sample from the forward probability
// distribution or take the previous choice into account
if ( ((uint) l)==end-1 ) {
transition_matrix = forward.get_view(i);
}
else {
// TODO: Make this sum faster
for (uint j=0; j<hidden_node_size; j++)
transition_matrix[j] = forward.get(i,j) * transition_matrices[i+1]->get(j, choice);
transition_matrix.normalize();
}
// Sample the hidden node from the forward probability distribution
transition_matrix.cumsum();
assert(rg);
double r = rg->get_rand();
choice = transition_matrix.bisect(r);
node->parentmap.set(l, choice);
// Sample the children
for(vector<Node*>::iterator child=node->children_1.begin(); child < node->children_1.end(); child++) {
if (mismask.get(l, (*child)->node_index) != MOCAPY_OBSERVED) {
(*child)->sample(l);
}
}
}
}
void PoissonSolver::solve(MDArray<Real>& a_Rhs) const {
//create a contructor
Box bx = a_Rhs.getBox();
MDArray<complex<double> > fftwForward(bx);
MDArray<complex<double> > fourierCoef(bx);
double h = 1.0/((double)m_N);
h = .0001;
//double h = 1/m_N; // integer division, becomes 0.0 in the end!
for (int k = 0; k < bx.sizeOf(); k++) {
fftwForward[k] = complex<double>(a_Rhs[k], 0.0);
}
FFTMD fftmd = FFTMD(m_fft1dptr);
fftmd.forwardCC(fftwForward);
for (int k=0; k<bx.sizeOf(); k++) {
int index[2];
bx.tupleIndex(k,index);
int i = index[0];
int j = index[1];
if (k == 0) {
fourierCoef[index] = complex<double>(0.0,0.0);
} else {
complex<double> div((2.0*cos(2.0*M_PI*((double)i)*h) - 2.0*cos(2.0*M_PI*((double)j)*h) - 4.0)/h/h,0.0);
fourierCoef[index] =fftwForward[index]/div;
}
}
fftmd.inverseCC(fourierCoef);
for (int k=0; k<bx.sizeOf(); k++) {
a_Rhs[k] = real(fourierCoef[k]) * h * h;
}
}
TEST(DISABLED_MDArray, Create)
{
i64 rows = 500L;
i64 cols = 500L;
MDCoord dim(2, rows, cols);
StorageType type = DMA;
Linearization *row = new RowMajor(dim);
Linearization *col = new ColMajor(dim);
i64 blockDims[] = {100L, 100L};
u8 blockOrders[] = {0, 1};
u8 microOrders[] = {0, 1};
permute(blockOrders, dim.nDim);
permute(microOrders, dim.nDim);
Linearization *block = new BlockBased(dim.nDim, dim.coords, blockDims,
blockOrders, microOrders);
const char *fileName = "test.bin";
MDArray *array = new MDArray(dim, type, row, fileName);
MDCoord coord1 = MDCoord(2, 120, 19);
ASSERT_TRUE(AC_OK == array->put(coord1, 12345));
Datum_t datum;
ASSERT_TRUE(AC_OK == array->get(coord1, datum));
ASSERT_DOUBLE_EQ(12345, datum);
MDCoord coord2 = MDCoord(2, 400, 133);
ASSERT_TRUE(AC_OK == array->put(coord2, 12.345));
MDCoord coord3 = MDCoord(2, 256, 19);
ASSERT_TRUE(AC_OK == array->put(coord3, 1000));
ASSERT_TRUE(AC_OK == array->get(coord2, datum));
ASSERT_DOUBLE_EQ(12.345, datum);
ASSERT_TRUE(AC_OK == array->get(coord3, datum));
ASSERT_DOUBLE_EQ(1000, datum);
// dense itor, write
MDIterator *it = array->createIterator(Dense, row);
MDCoord coord;
int i = 0;
while (it->moveNext())
{
it->put(i);
i++;
}
ASSERT_EQ(rows*cols, i);
delete it;
// block itor, read
it = array->createIterator(Dense, block);
i = 0;
while(it->moveNext())
{
it->get(coord, datum);
ASSERT_DOUBLE_EQ(row->linearize(coord), datum);
i++;
}
ASSERT_EQ(rows*cols, i);
delete it;
/*
// natural itor, read
it = array->createNaturalIterator(Dense);
i = 1;
while(it->moveNext())
{
it->get(coord, datum);
ASSERT_EQ(i, row->linearize(coord));
ASSERT_EQ(i, datum);
i++;
}
ASSERT_EQ(rows*cols, i);
delete it;
*/
delete row;
delete col;
delete block;
delete array;
}
int main(void) {
mocapy_seed((uint)5556575);
// Number of trainining sequences
int N = 500;
// Sequence lengths
int T = 100;
// Gibbs sampling parameters
int MCMC_BURN_IN = 10;
// HMM hidden and observed node sizes
uint H_SIZE=2;
bool init_random=false;
CPD th0_cpd;
th0_cpd.set_shape(2); th0_cpd.set_values(vec(0.1, 0.9));
CPD th1_cpd;
th1_cpd.set_shape(2,2); th1_cpd.set_values( vec(0.95, 0.05, 0.1, 0.9));
// The target DBN (This DBN generates the data)
Node* th0 = NodeFactory::new_discrete_node(H_SIZE, "th0", init_random, th0_cpd);
Node* th1 = NodeFactory::new_discrete_node(H_SIZE, "th1", init_random, th1_cpd);
Node* to0 = NodeFactory::new_kent_node("to0");
DBN tdbn;
tdbn.set_slices(vec(th0, to0), vec(th1, to0));
tdbn.add_intra("th0", "to0");
tdbn.add_inter("th0", "th1");
tdbn.construct();
// The model DBN (this DBN will be trained)
// For mh0, get the CPD from th0 and fix parameters
Node* mh0 = NodeFactory::new_discrete_node(H_SIZE, "mh0", init_random, CPD(), th0, true );
Node* mh1 = NodeFactory::new_discrete_node(H_SIZE, "mh1", init_random);
Node* mo0 = NodeFactory::new_kent_node("mo0");
DBN mdbn;
mdbn.set_slices(vec(mh0, mo0), vec(mh1, mo0));
mdbn.add_intra("mh0", "mo0");
mdbn.add_inter("mh0", "mh1");
mdbn.construct();
cout << "*** TARGET ***" << endl;
cout << *th0 << endl;
cout << *th1 << endl;
cout << *to0 << endl;
cout << "*** MODEL ***" << endl;
cout << *mh0 << endl;
cout << *mh1 << endl;
cout << *mo0 << endl;
vector<Sequence> seq_list;
vector< MDArray<eMISMASK> > mismask_list;
cout << "Generating data" << endl;
MDArray<eMISMASK> mismask;
mismask.repeat(T, vec(MOCAPY_HIDDEN, MOCAPY_OBSERVED));
// Generate the data
double sum_LL(0);
for (int i=0; i<N; i++) {
pair<Sequence, double> seq_ll = tdbn.sample_sequence(T);
sum_LL += seq_ll.second;
seq_list.push_back(seq_ll.first);
mismask_list.push_back(mismask);
}
cout << "Average LL: " << sum_LL/N << endl;
GibbsRandom mcmc = GibbsRandom(&mdbn);
EMEngine em = EMEngine(&mdbn, &mcmc, &seq_list, &mismask_list);
cout << "Starting EM loop" << endl;
double bestLL=-1000;
uint it_no_improvement(0);
uint i(0);
// Start EM loop
while (it_no_improvement<10) {
em.do_E_step(1, MCMC_BURN_IN, true);
double ll = em.get_loglik();
cout << "LL= " << ll;
if (ll > bestLL) {
cout << " * saving model *" << endl;
mdbn.save("kent_hmm.dbn");
bestLL = ll;
it_no_improvement=0;
}
else { it_no_improvement++; cout << endl; }
i++;
em.do_M_step();
}
cout << "DONE" << endl;
mdbn.load("kent_hmm.dbn");
cout << "*** TARGET ***" << endl;
cout << *th0 << endl;
cout << *th1 << endl;
cout << *to0 << endl;
cout << "*** MODEL ***" << endl;
cout << *mh0 << endl;
cout << *mh1 << endl;
cout << *mo0 << endl;
delete th0;
delete th1;
delete to0;
delete mh0;
delete mh1;
delete mo0;
return EXIT_SUCCESS;
}
void StringFunction::operator()(MDArray& inp, MDArray& out, double time_value_optional)
{
EXCEPTWATCH;
argsAreValid(inp, out);
PRINT("tmp srk StringFunction::operator() getName()= " << getName() << " inp= " << inp << " out= " << out);
int domain_rank = m_domain_dimensions.size();
int codomain_rank = m_codomain_dimensions.size();
// FIXME move to argsAreValid
VERIFY_OP(domain_rank, ==, 1, "StringFunction::operator(): must specify domain Dimensions as rank 1 (for now)");
VERIFY_OP(codomain_rank, ==, 1, "StringFunction::operator(): must specify codomain Dimensions as rank 1 (for now)");
int inp_rank = inp.rank();
int out_rank = out.rank();
int inp_offset = inp_rank - domain_rank;
int out_offset = out_rank - codomain_rank;
// returns 1,1,1,... etc. if that dimension doesn't exist
enum { maxRank = 3};
VERIFY_OP_ON(inp_rank, <=, maxRank, "StringFunction::operator() input array rank too large");
VERIFY_OP_ON(out_rank, <=, maxRank, "StringFunction::operator() output array rank too large");
int n_inp_points[maxRank] = {1,1,1};
int n_out_points[maxRank] = {1,1,1};
first_dimensions(inp, inp_offset, n_inp_points, maxRank);
first_dimensions(out, out_offset, n_out_points, maxRank);
int nInDim = m_domain_dimensions[0];
int nOutDim = m_codomain_dimensions[0];
MDArray inp_loc(nInDim);
MDArray out_loc(nOutDim);
int iDim[maxRank-1] = {0,0};
double *xyzt[] = {&m_x, &m_y, &m_z, &m_t};
PRINT("getName()= " << getName() << " n_inp_points= " << n_inp_points[0] << " " << n_inp_points[1] << " " << n_inp_points[2] << " nInDim= " << nInDim);
m_t = time_value_optional;
for (iDim[0] = 0; iDim[0] < n_inp_points[0]; iDim[0]++)
{
for (iDim[1] = 0; iDim[1] < n_inp_points[1]; iDim[1]++)
{
for (int id = 0; id < nInDim; id++)
{
switch (inp_rank)
{
case 3:
inp_loc(id) = inp(iDim[0], iDim[1], id);
*xyzt[id] = inp(iDim[0], iDim[1], id);
break;
case 2:
inp_loc(id) = inp(iDim[0], id);
*xyzt[id] = inp(iDim[0], id);
break;
case 1:
inp_loc(id) = inp( id);
*xyzt[id] = inp( id);
break;
}
}
if (m_func_to_value.size())
{
evalFunctions(inp_loc, time_value_optional);
}
//Save the evaluations slightly differently whether it's a scalar or vector valued function
if ((int)m_v.size() != nOutDim)
{
m_v.resize(nOutDim);
}
PRINT("getName()= " << getName() << " about to evaluate m_functionExpr... iDim[0] = " << iDim[0] << " iDim[1]= " << iDim[1]);
if(nOutDim == 1)
m_v[0] = m_functionExpr.evaluate();
else
m_functionExpr.evaluate();
PRINT("getName()= " << getName() << " done to evaluate m_functionExpr... iDim[0] = " << iDim[0] << " iDim[1]= " << iDim[1]);
for (int iOutDim = 0; iOutDim < nOutDim; iOutDim++)
{
switch (out_rank)
{
case 1: out(iOutDim) = m_v[iOutDim]; break;
case 2: out(iDim[0], iOutDim) = m_v[iOutDim]; break;
case 3: out(iDim[0], iDim[1], iOutDim) = m_v[iOutDim]; break;
default:
VERIFY_1("StringFunction::operator() bad output rank");
}
}
}
}
PRINT("getName()= " << getName() << " done in operator()");
}
void NewtonSolver<NonlinearProblem>::solve( MDArray& u,
NonlinearProblem& problem,
const double tolerance,
const int max_iters )
{
DTK_REQUIRE( 2 == u.rank() );
// Allocate nonlinear residual, Jacobian, Newton update, and work arrays.
int d0 = u.dimension(0);
int d1 = u.dimension(1);
MDArray F( d0, d1 );
MDArray J( d0, d1, d1 );
MDArray J_inv( d0, d1, d1 );
MDArray update( d0, d1 );
MDArray u_old = u;
MDArray conv_check( d0 );
// Compute the initial state.
NPT::updateState( problem, u );
// Computen the initial nonlinear residual and scale by -1 to get -F(u).
NPT::evaluateResidual( problem, u, F );
Intrepid::RealSpaceTools<Scalar>::scale(
F, -Teuchos::ScalarTraits<Scalar>::one() );
// Compute the initial Jacobian.
NPT::evaluateJacobian( problem, u, J );
// Check for degeneracy of the Jacobian. If it is degenerate then the
// problem is ill conditioned and return very large numbers in the state
// vector that correspond to no solution.
MDArray det( 1 );
Intrepid::RealSpaceTools<Scalar>::det( det, J );
if ( std::abs(det(0)) < tolerance )
{
for ( int m = 0; m < d0; ++m )
{
for ( int n = 0; n < d1; ++n )
{
u(m,n) = std::numeric_limits<Scalar>::max();
}
}
return;
}
// Nonlinear solve.
for ( int k = 0; k < max_iters; ++k )
{
// Solve the linear model, delta_u = J^-1 * -F(u).
Intrepid::RealSpaceTools<Scalar>::inverse( J_inv, J );
Intrepid::RealSpaceTools<Scalar>::matvec( update, J_inv, F );
// Update the solution, u += delta_u.
Intrepid::RealSpaceTools<Scalar>::add( u, update );
// Check for convergence.
Intrepid::RealSpaceTools<Scalar>::subtract( u_old, u );
Intrepid::RealSpaceTools<Scalar>::vectorNorm(
conv_check, u_old, Intrepid::NORM_TWO );
if ( tolerance > conv_check(0) )
{
break;
}
// Reset for the next iteration.
u_old = u;
// Update any state-dependent data from the last iteration using the
// new solution vector.
NPT::updateState( problem, u );
// Compute the nonlinear residual and scale by -1 to get -F(u).
NPT::evaluateResidual( problem, u, F );
Intrepid::RealSpaceTools<Scalar>::scale(
F, -Teuchos::ScalarTraits<Scalar>::one() );
// Compute the Jacobian.
NPT::evaluateJacobian( problem, u, J );
}
// Check for convergence.
DTK_ENSURE( tolerance > conv_check(0) );
}
int main(void) {
mocapy_seed((uint)924573);
// Number of trainining sequences
int N = 2000;
// Allocate storage in each node.
int max_num_dat_points = N;
// Sequence lengths
int T = 1;
// Gibbs sampling parameters
int MCMC_BURN_IN = 10;
// HMM hidden and observed node sizes
uint H_SIZE=2;
uint O_SIZE=20;
bool init_random=true;
// The target DBN (This DBN generates the data)
Node* th0 = NodeFactory::new_discrete_node(H_SIZE, "th0", init_random);
Node* th1 = NodeFactory::new_discrete_node(H_SIZE, "th1", init_random);
Node* to0 = NodeFactory::new_GD_node("to0", O_SIZE , max_num_dat_points);
cout<< "Done making GD node" << endl;
DBN tdbn;
tdbn.set_slices(vec(th0, to0), vec(th1, to0));
tdbn.add_intra("th0", "to0");
tdbn.add_inter("th0", "th1");
cout<< "model has been defined" << endl;
tdbn.construct();
// The model DBN (this DBN will be trained)
// For mh0, get the CPD from th0 and fix parameters
Node* mh0 = NodeFactory::new_discrete_node(H_SIZE, "mh0", init_random, CPD(), th0, true );
Node* mh1 = NodeFactory::new_discrete_node(H_SIZE, "mh1", init_random);
Node* mo0 = NodeFactory::new_GD_node("mo0",O_SIZE, max_num_dat_points);
DBN mdbn;
mdbn.set_slices(vec(mh0, mo0), vec(mh1, mo0));
mdbn.add_intra("mh0", "mo0");
mdbn.add_inter("mh0", "mh1");
mdbn.construct();
cout << "*** TARGET ***" << endl;
cout << *th0 << endl;
cout << *th1 << endl;
cout << *to0 << endl;
cout << "*** MODEL ***" << endl;
cout << *mh0 << endl;
cout << *mh1 << endl;
cout << *mo0 << endl;
vector<Sequence> seq_list;
vector< MDArray<eMISMASK> > mismask_list;
cout << "Generating data" << endl;
MDArray<eMISMASK> mismask;
mismask.repeat(T, vec(MOCAPY_HIDDEN, MOCAPY_OBSERVED));
// Setting parameters
//CPD pars2;
CPD pars;
int t = 4;
pars.set_shape(H_SIZE,2*(O_SIZE-1));
//pars2.set_shape(H_SIZE,2*(O_SIZE-1));
vector<double> parvec;
for(uint i = 0;i < O_SIZE-1;i++){
parvec.push_back(t* 1.0);
}
for(double i = O_SIZE-1;i >= 1;i--){
parvec.push_back(t*i);
}
// pars.set_values(parvec);
double s = 2;
for(uint i = 0;i < O_SIZE-1;i++){
parvec.push_back(s);
}
for(double i = s*(O_SIZE-1);i >= s*1;i=i-s){
parvec.push_back(i);
}
pars.set_values(parvec);
((GDNode*)to0)->get_densities()->set_parameters(pars);
pair<Sequence, double> seq_ll;
// Generate the data
double sum_LL(0);
for (int i=0; i<N; i++) {
seq_ll = tdbn.sample_sequence(T);
sum_LL += seq_ll.second;
seq_list.push_back(seq_ll.first);
mismask_list.push_back(mismask);
}
cout << "Average LL: " << sum_LL/N << endl;
GibbsRandom mcmc = GibbsRandom(&mdbn);
EMEngine em = EMEngine(&mdbn, &mcmc, &seq_list, &mismask_list);
cout << "Starting EM loop" << endl;
double bestLL=-100000;
uint it_no_improvement(0);
uint i(0);
bool l;
// Start EM loop
while (it_no_improvement<100) {
em.do_E_step(1, MCMC_BURN_IN, true);
double ll = em.get_loglik();
cout << "LL= " << ll;
if (ll > bestLL) {
cout << " * saving model *" << endl;
cout << *mo0 << endl;
l = true;
//mdbn.save("discrete_hmm.dbn");
bestLL = ll;
it_no_improvement=0;
}
else {
it_no_improvement++; cout << endl;
}
i++;
em.do_M_step();
if (l){
l=false;
}
}
cout << "DONE" << endl;
// mdbn.load("discrete_hmm.dbn");
cout << "*** TARGET ***" << endl;
cout << *th0 << endl;
cout << *th1 << endl;
cout << *to0 << endl;
cout << "*** MODEL ***" << endl;
cout << *mh0 << endl;
cout << *mh1 << endl;
cout << *mo0 << endl;
delete th0;
delete th1;
delete to0;
delete mh0;
delete mh1;
delete mo0;
return EXIT_SUCCESS;
}