LArSoft  v09_90_00
Liquid Argon Software toolkit - https://larsoft.org/
LArSupportVectorMachine.h
Go to the documentation of this file.
1 
8 #ifndef LAR_SUPPORT_VECTOR_MACHINE_H
9 #define LAR_SUPPORT_VECTOR_MACHINE_H 1
10 
12 
14 
15 #include "Helpers/XmlHelper.h"
16 #include "Pandora/StatusCodes.h"
17 
18 #include <functional>
19 #include <map>
20 #include <vector>
21 
22 //------------------------------------------------------------------------------------------------------------------------------------------
23 
24 namespace lar_content
25 {
26 
31 {
32 public:
33  typedef std::function<double(const LArMvaHelper::MvaFeatureVector &, const LArMvaHelper::MvaFeatureVector &, const double)> KernelFunction;
34 
39  {
41  LINEAR = 1,
42  QUADRATIC = 2,
43  CUBIC = 3,
45  };
46 
51 
60  pandora::StatusCode Initialize(const std::string &parameterLocation, const std::string &svmName);
61 
69  bool Classify(const LArMvaHelper::MvaFeatureVector &features) const;
70 
79 
87  double CalculateProbability(const LArMvaHelper::MvaFeatureVector &features) const;
88 
94  bool IsInitialized() const;
95 
101  unsigned int GetNFeatures() const;
102 
108  void SetKernelFunction(KernelFunction kernelFunction);
109 
110 private:
115  {
116  public:
123  SupportVectorInfo(const double yAlpha, LArMvaHelper::MvaFeatureVector supportVector);
124 
125  double m_yAlpha;
127  };
128 
133  {
134  public:
141  FeatureInfo(const double muValue, const double sigmaValue);
142 
146  FeatureInfo();
147 
155  double StandardizeParameter(const double parameter) const;
156 
157  double m_muValue;
158  double m_sigmaValue;
159  };
160 
161  typedef std::vector<SupportVectorInfo> SVInfoList;
162  typedef std::vector<FeatureInfo> FeatureInfoVector;
163 
164  typedef std::map<KernelType, KernelFunction> KernelMap;
165 
167 
171 
173  unsigned int m_nFeatures;
174  double m_bias;
175  double m_scaleFactor;
176 
177  SVInfoList m_svInfoList;
178  FeatureInfoVector m_featureInfoList;
179 
181  KernelFunction m_kernelFunction;
182  KernelMap m_kernelMap;
183 
190  void ReadXmlFile(const std::string &svmFileName, const std::string &svmName);
191 
199  pandora::StatusCode ReadComponent(pandora::TiXmlElement *pCurrentXmlElement);
200 
208  pandora::StatusCode ReadMachine(const pandora::TiXmlHandle &currentHandle);
209 
217  pandora::StatusCode ReadFeatures(const pandora::TiXmlHandle &currentHandle);
218 
226  pandora::StatusCode ReadSupportVector(const pandora::TiXmlHandle &currentHandle);
227 
236 
246  static double QuadraticKernel(
247  const LArMvaHelper::MvaFeatureVector &supportVector, const LArMvaHelper::MvaFeatureVector &features, const double scaleFactor = 1.);
248 
258  static double CubicKernel(
259  const LArMvaHelper::MvaFeatureVector &supportVector, const LArMvaHelper::MvaFeatureVector &features, const double scaleFactor = 1.);
260 
270  static double LinearKernel(
271  const LArMvaHelper::MvaFeatureVector &supportVector, const LArMvaHelper::MvaFeatureVector &features, const double scaleFactor = 1.);
272 
282  static double GaussianRbfKernel(
283  const LArMvaHelper::MvaFeatureVector &supportVector, const LArMvaHelper::MvaFeatureVector &features, const double scaleFactor = 1.);
284 };
285 
286 //------------------------------------------------------------------------------------------------------------------------------------------
287 
289 {
290  return (this->CalculateClassificationScoreImpl(features) > 0.);
291 }
292 
293 //------------------------------------------------------------------------------------------------------------------------------------------
294 
296 {
297  return this->CalculateClassificationScoreImpl(features);
298 }
299 
300 //------------------------------------------------------------------------------------------------------------------------------------------
301 
303 {
304  if (!m_enableProbability)
305  {
306  std::cout << "LArSupportVectorMachine: cannot calculate probabilities for this SVM" << std::endl;
307  throw pandora::STATUS_CODE_NOT_INITIALIZED;
308  }
309 
310  // Use the logistic function to map the linearly-transformed score on the interval (-inf,inf) to a probability on [0,1] - the two free
311  // parameters in the linear transformation are trained such that the logistic map produces an accurate probability
312  const double scaledScore = m_probAParameter * this->CalculateClassificationScoreImpl(features) + m_probBParameter;
313 
314  return 1. / (1. + std::exp(scaledScore));
315 }
316 
317 //------------------------------------------------------------------------------------------------------------------------------------------
318 
320 {
321  return m_isInitialized;
322 }
323 
324 //------------------------------------------------------------------------------------------------------------------------------------------
325 
326 inline unsigned int SupportVectorMachine::GetNFeatures() const
327 {
328  return m_nFeatures;
329 }
330 
331 //------------------------------------------------------------------------------------------------------------------------------------------
332 
334 {
335  m_kernelFunction = std::move(kernelFunction);
336 }
337 
338 //------------------------------------------------------------------------------------------------------------------------------------------
339 
341  const LArMvaHelper::MvaFeatureVector &supportVector, const LArMvaHelper::MvaFeatureVector &features, const double scaleFactor)
342 {
343  const double denominator(scaleFactor * scaleFactor);
344  if (denominator < std::numeric_limits<double>::epsilon())
345  throw pandora::StatusCodeException(pandora::STATUS_CODE_INVALID_PARAMETER);
346 
347  double total(0.);
348  for (unsigned int i = 0; i < features.size(); ++i)
349  total += supportVector.at(i).Get() * features.at(i).Get();
350 
351  return total / denominator;
352 }
353 
354 //------------------------------------------------------------------------------------------------------------------------------------------
355 
357  const LArMvaHelper::MvaFeatureVector &supportVector, const LArMvaHelper::MvaFeatureVector &features, const double scaleFactor)
358 {
359  const double denominator(scaleFactor * scaleFactor);
360  if (denominator < std::numeric_limits<double>::epsilon())
361  throw pandora::StatusCodeException(pandora::STATUS_CODE_INVALID_PARAMETER);
362 
363  double total(0.);
364  for (unsigned int i = 0; i < features.size(); ++i)
365  total += supportVector.at(i).Get() * features.at(i).Get();
366 
367  total = total / denominator + 1.;
368  return total * total;
369 }
370 
371 //------------------------------------------------------------------------------------------------------------------------------------------
372 
374  const LArMvaHelper::MvaFeatureVector &supportVector, const LArMvaHelper::MvaFeatureVector &features, const double scaleFactor)
375 {
376  const double denominator(scaleFactor * scaleFactor);
377  if (denominator < std::numeric_limits<double>::epsilon())
378  throw pandora::StatusCodeException(pandora::STATUS_CODE_INVALID_PARAMETER);
379 
380  double total(0.);
381  for (unsigned int i = 0; i < features.size(); ++i)
382  total += supportVector.at(i).Get() * features.at(i).Get();
383 
384  total = total / denominator + 1.;
385  return total * total * total;
386 }
387 
388 //------------------------------------------------------------------------------------------------------------------------------------------
389 
391  const LArMvaHelper::MvaFeatureVector &supportVector, const LArMvaHelper::MvaFeatureVector &features, const double scaleFactor)
392 {
393  double total(0.);
394  for (unsigned int i = 0; i < features.size(); ++i)
395  total += (supportVector.at(i).Get() - features.at(i).Get()) * (supportVector.at(i).Get() - features.at(i).Get());
396 
397  return std::exp(-scaleFactor * total);
398 }
399 
400 //------------------------------------------------------------------------------------------------------------------------------------------
401 
403  m_yAlpha(yAlpha),
404  m_supportVector(std::move(supportVector))
405 {
406 }
407 
408 //------------------------------------------------------------------------------------------------------------------------------------------
409 
410 inline SupportVectorMachine::FeatureInfo::FeatureInfo(const double muValue, const double sigmaValue) :
411  m_muValue(muValue),
412  m_sigmaValue(sigmaValue)
413 {
414 }
415 
416 //------------------------------------------------------------------------------------------------------------------------------------------
417 
419  m_muValue(0.),
420  m_sigmaValue(0.)
421 {
422 }
423 
424 //------------------------------------------------------------------------------------------------------------------------------------------
425 
426 inline double SupportVectorMachine::FeatureInfo::StandardizeParameter(const double parameter) const
427 {
428  if (m_sigmaValue < std::numeric_limits<double>::epsilon())
429  throw pandora::StatusCodeException(pandora::STATUS_CODE_INVALID_PARAMETER);
430 
431  return (parameter - m_muValue) / m_sigmaValue;
432 }
433 
434 } // namespace lar_content
435 
436 #endif // #ifndef LAR_SUPPORT_VECTOR_MACHINE_H
static double CubicKernel(const LArMvaHelper::MvaFeatureVector &supportVector, const LArMvaHelper::MvaFeatureVector &features, const double scaleFactor=1.)
An inhomogeneous cubic kernel.
bool Classify(const LArMvaHelper::MvaFeatureVector &features) const
Make a classification for a set of input features, based on the trained model.
pandora::StatusCode ReadComponent(pandora::TiXmlElement *pCurrentXmlElement)
Read the component at the current xml element.
pandora::StatusCode ReadSupportVector(const pandora::TiXmlHandle &currentHandle)
Read the support vector component at the current xml handle.
MvaTypes::MvaFeatureVector MvaFeatureVector
Definition: LArMvaHelper.h:75
std::vector< SupportVectorInfo > SVInfoList
bool m_enableProbability
Whether to enable probability calculations.
double m_scaleFactor
The kernel scale factor.
FeatureInfoVector m_featureInfoList
The list of FeatureInfo objects.
double CalculateClassificationScore(const LArMvaHelper::MvaFeatureVector &features) const
Calculate the classification score for a set of input features, based on the trained model...
MvaInterface class.
STL namespace.
std::vector< FeatureInfo > FeatureInfoVector
SVInfoList m_svInfoList
The list of SupportVectorInfo objects.
bool IsInitialized() const
Query whether this svm is initialized.
static double QuadraticKernel(const LArMvaHelper::MvaFeatureVector &supportVector, const LArMvaHelper::MvaFeatureVector &features, const double scaleFactor=1.)
An inhomogeneous quadratic kernel.
void SetKernelFunction(KernelFunction kernelFunction)
Set the kernel function to use.
static double LinearKernel(const LArMvaHelper::MvaFeatureVector &supportVector, const LArMvaHelper::MvaFeatureVector &features, const double scaleFactor=1.)
A linear kernel.
pandora::StatusCode ReadMachine(const pandora::TiXmlHandle &currentHandle)
Read the machine component at the current xml handle.
double m_muValue
The average value of this feature.
bool m_standardizeFeatures
Whether to standardize the features.
LArMvaHelper::MvaFeatureVector m_supportVector
The support vector.
void ReadXmlFile(const std::string &svmFileName, const std::string &svmName)
Read the svm parameters from an xml file.
KernelMap m_kernelMap
Map from the kernel types to the kernel functions.
KernelFunction m_kernelFunction
The kernel function.
double StandardizeParameter(const double parameter) const
Standardize a parameter corresponding to this feature.
FeatureInfo()
Default constructor to allow default-construction of (uninitialized) svms.
KernelType m_kernelType
The kernel type.
pandora::StatusCode Initialize(const std::string &parameterLocation, const std::string &svmName)
Initialize the svm using a serialized model.
double m_probAParameter
The first-order score coefficient for mapping to a probability using the logistic function...
unsigned int m_nFeatures
The number of features.
double m_yAlpha
The alpha-value multiplied by the y-value for the support vector.
std::function< double(const LArMvaHelper::MvaFeatureVector &, const LArMvaHelper::MvaFeatureVector &, const double)> KernelFunction
std::map< KernelType, KernelFunction > KernelMap
double CalculateProbability(const LArMvaHelper::MvaFeatureVector &features) const
Calculate the classification probability for a set of input features, based on the trained model...
static double GaussianRbfKernel(const LArMvaHelper::MvaFeatureVector &supportVector, const LArMvaHelper::MvaFeatureVector &features, const double scaleFactor=1.)
A gaussian RBF kernel.
double CalculateClassificationScoreImpl(const LArMvaHelper::MvaFeatureVector &features) const
Implementation method for calculating the classification score using the trained model.
bool m_isInitialized
Whether this svm has been initialized.
double m_probBParameter
The score offset parameter for mapping to a probability using the logistic function.
pandora::StatusCode ReadFeatures(const pandora::TiXmlHandle &currentHandle)
Read the feature component at the current xml handle.
SupportVectorInfo(const double yAlpha, LArMvaHelper::MvaFeatureVector supportVector)
Constructor.
unsigned int GetNFeatures() const
Get the number of features.
Header file for the lar multivariate analysis interface class.