Skip to content

Commit

Permalink
updated streaming mmd to work with blocks
Browse files Browse the repository at this point in the history
  • Loading branch information
@lambday @karlnapf
lambday authored and karlnapf committed Jul 1, 2016
1 parent 656b51e commit b779d7d
Show file tree
Hide file tree
Showing 2 changed files with 72 additions and 34 deletions.
81 changes: 55 additions & 26 deletions src/shogun/statistical_testing/MMD.cpp
View file
Expand Up @@ -28,7 +28,6 @@
* either expressed or implied, of the Shogun Development Team.
*/

#include <utility>
#include <vector>
#include <memory>
#include <type_traits>
Expand All @@ -42,8 +41,11 @@
#include <shogun/statistical_testing/LinearTimeMMD.h>
#include <shogun/statistical_testing/internals/NextSamples.h>
#include <shogun/statistical_testing/internals/DataManager.h>
#include <shogun/statistical_testing/internals/FeaturesUtil.h>
#include <shogun/statistical_testing/internals/KernelManager.h>
#include <shogun/statistical_testing/internals/ComputationManager.h>
#include <shogun/statistical_testing/internals/MaxMeasure.h>
#include <shogun/statistical_testing/internals/OptMeasure.h>
#include <shogun/statistical_testing/internals/mmd/BiasedFull.h>
#include <shogun/statistical_testing/internals/mmd/UnbiasedFull.h>
#include <shogun/statistical_testing/internals/mmd/UnbiasedIncomplete.h>
Expand All @@ -61,8 +63,8 @@ struct CMMD::Self
void create_variance_job();
void create_computation_jobs();

void merge_samples(NextSamples&, std::vector<std::shared_ptr<CFeatures>>&) const;
void compute_kernel(ComputationManager&, std::vector<std::shared_ptr<CFeatures>>&, CKernel*) const;
void merge_samples(NextSamples&, std::vector<CFeatures*>&) const;
void compute_kernel(ComputationManager&, std::vector<CFeatures*>&, CKernel*) const;
void compute_jobs(ComputationManager&) const;

std::pair<float64_t, float64_t> compute_statistic_variance();
Expand Down Expand Up @@ -134,41 +136,32 @@ void CMMD::Self::create_variance_job()
};
}

#define get_block_p(i) next_burst[0][i]
#define get_block_q(i) next_burst[1][i]
void CMMD::Self::merge_samples(NextSamples& next_burst, std::vector<std::shared_ptr<CFeatures>>& blocks) const
void CMMD::Self::merge_samples(NextSamples& next_burst, std::vector<CFeatures*>& blocks) const
{
blocks.resize(next_burst.num_blocks());

#pragma omp parallel for
for (size_t i = 0; i < blocks.size(); ++i)
for (size_t i=0; i<blocks.size(); ++i)
{
auto block_p=get_block_p(i);
auto block_q=get_block_q(i);

auto block_p_and_q=block_p->create_merged_copy(block_q.get());
SG_REF(block_p_and_q);

block_p=nullptr;
block_q=nullptr;

blocks[i]=std::shared_ptr<CFeatures>(block_p_and_q, [](CFeatures* ptr) { SG_UNREF(ptr); });
auto block_p=next_burst[0][i].get();
auto block_q=next_burst[1][i].get();
auto block_p_and_q=FeaturesUtil::create_merged_copy(block_p, block_q);
blocks[i]=block_p_and_q;
}
next_burst.clear();
}
#undef get_block_p
#undef get_block_q

void CMMD::Self::compute_kernel(ComputationManager& cm, std::vector<std::shared_ptr<CFeatures>>& blocks, CKernel* kernel) const
void CMMD::Self::compute_kernel(ComputationManager& cm, std::vector<CFeatures*>& blocks, CKernel* kernel) const
{
REQUIRE(kernel->get_kernel_type()!=K_CUSTOM, "Underlying kernel cannot be custom for streaming test!\n");
REQUIRE(kernel->get_kernel_type()!=K_CUSTOM, "Underlying kernel cannot be custom!\n");
cm.num_data(blocks.size());
const auto& dm=owner.get_data_manager();
#pragma omp parallel for
for (size_t i=0; i<blocks.size(); ++i)
{
try
{
auto kernel_clone=std::unique_ptr<CKernel>(static_cast<CKernel*>(kernel->clone()));
kernel_clone->init(blocks[i].get(), blocks[i].get());
kernel_clone->init(blocks[i], blocks[i]);
cm.data(i)=kernel_clone->get_kernel_matrix<float32_t>();
kernel_clone->remove_lhs_and_rhs();
}
Expand All @@ -177,6 +170,7 @@ void CMMD::Self::compute_kernel(ComputationManager& cm, std::vector<std::shared_
SG_SERROR("%s, Try using less number of blocks per burst!\n", e.get_exception_string());
}
}
blocks.resize(0);
}

void CMMD::Self::compute_jobs(ComputationManager& cm) const
Expand Down Expand Up @@ -206,11 +200,10 @@ std::pair<float64_t, float64_t> CMMD::Self::compute_statistic_variance()
cm.enqueue_job(statistic_job);
cm.enqueue_job(variance_job);

std::vector<std::shared_ptr<CFeatures>> blocks;
std::vector<CFeatures*> blocks;

dm.start();
auto next_burst=dm.next();

while (!next_burst.empty())
{
merge_samples(next_burst, blocks);
Expand Down Expand Up @@ -278,7 +271,7 @@ SGVector<float64_t> CMMD::Self::sample_null()
create_statistic_job();
cm.enqueue_job(permutation_job);

std::vector<std::shared_ptr<CFeatures>> blocks;
std::vector<CFeatures*> blocks;

dm.start();
auto next_burst=dm.next();
Expand Down Expand Up @@ -328,6 +321,31 @@ void CMMD::add_kernel(CKernel* kernel)
self->kernel_selection_mgr.push_back(kernel);
}

void CMMD::select_kernel(EKernelSelectionMethod kmethod)
{
SG_DEBUG("Entering!\n");
SG_DEBUG("Selecting kernels from a total of %d kernels!\n", self->kernel_selection_mgr.num_kernels());
switch (kmethod)
{
case EKernelSelectionMethod::MAXIMIZE_MMD:
{
MaxMeasure policy(self->kernel_selection_mgr, this);
get_kernel_manager().kernel_at(0)=policy.select_kernel();
break;
}
case EKernelSelectionMethod::OPTIMIZE_MMD:
{
OptMeasure policy(self->kernel_selection_mgr, this);
get_kernel_manager().kernel_at(0)=policy.select_kernel();
break;
}
default:
SG_ERROR("Unsupported kernel selection method specified!\n");
break;
}
SG_DEBUG("Leaving!\n");
}

float64_t CMMD::compute_statistic()
{
return self->compute_statistic_variance().first;
Expand All @@ -338,6 +356,11 @@ float64_t CMMD::compute_variance()
return self->compute_statistic_variance().second;
}

std::pair<float64_t, float64_t> CMMD::compute_statistic_variance()
{
return self->compute_statistic_variance();
}

SGVector<float64_t> CMMD::sample_null()
{
return self->sample_null();
Expand All @@ -363,6 +386,12 @@ bool CMMD::use_gpu() const
return self->use_gpu;
}

void CMMD::cleanup()
{
for (size_t i=0; i<get_kernel_manager().num_kernels(); ++i)
get_kernel_manager().restore_kernel_at(i);
}

void CMMD::set_statistic_type(EStatisticType stype)
{
self->statistic_type=stype;
Expand Down
25 changes: 17 additions & 8 deletions src/shogun/statistical_testing/MMD.h
View file
Expand Up @@ -31,6 +31,7 @@
#ifndef MMD_H_
#define MMD_H_

#include <utility>
#include <memory>
#include <functional>
#include <shogun/statistical_testing/TwoSampleTest.h>
Expand All @@ -42,6 +43,13 @@ class CKernel;
template <typename> class SGVector;
template <typename> class SGMatrix;

namespace internal
{

class OptMeasure;

}

enum class EStatisticType
{
UNBIASED_FULL,
Expand All @@ -67,24 +75,29 @@ enum class EKernelSelectionMethod
{
MEDIAN_HEURISRIC,
MAXIMIZE_MMD,
OPTIMIZE_MMD,
MAXIMIZE_POWER
};

class CMMD : public CTwoSampleTest
{
using operation=std::function<float32_t(SGMatrix<float32_t>)>;
friend class internal::OptMeasure;
public:
CMMD();
virtual ~CMMD();

void add_kernel(CKernel *kernel);
/*
void select_kernel(EKernelSelectionMethod kmethod);
CKernel* get_kernel() const;
*/

virtual float64_t compute_statistic() override;
virtual float64_t compute_variance();

virtual SGVector<float64_t> sample_null() override;

void use_gpu(bool gpu);
void cleanup();

void set_statistic_type(EStatisticType stype);
const EStatisticType get_statistic_type() const;

Expand All @@ -97,10 +110,6 @@ class CMMD : public CTwoSampleTest
void set_null_approximation_method(ENullApproximationMethod nmethod);
const ENullApproximationMethod get_null_approximation_method() const;

virtual SGVector<float64_t> sample_null() override;

void use_gpu(bool gpu);

virtual const char* get_name() const;
protected:
virtual const operation get_direct_estimation_method() const = 0;
Expand All @@ -110,7 +119,7 @@ class CMMD : public CTwoSampleTest
private:
struct Self;
std::unique_ptr<Self> self;

virtual std::pair<float64_t, float64_t> compute_statistic_variance();
};

}
Expand Down

0 comments on commit b779d7d

Please sign in to comment.