PLearn 0.1
Public Types | Public Member Functions | Static Public Member Functions | Public Attributes | Static Public Attributes | Static Protected Member Functions | Private Member Functions
PLearn::HistogramDistribution Class Reference

#include <HistogramDistribution.h>

Inheritance diagram for PLearn::HistogramDistribution:
Inheritance graph
[legend]
Collaboration diagram for PLearn::HistogramDistribution:
Collaboration graph
[legend]

List of all members.

Public Types

typedef PDistribution inherited

Public Member Functions

 HistogramDistribution ()
 HistogramDistribution (VMat data, PP< Binner > binner_=0, PP< Smoother > smoother_=0)
 This constructor calls train as part of the construction process.
virtual void build ()
 Simply calls inherited::build() then build_().
virtual void makeDeepCopyFromShallowCopy (CopiesMap &copies)
 Transforms a shallow copy into a deep copy.
virtual string classname () const
virtual OptionListgetOptionList () const
virtual OptionMapgetOptionMap () const
virtual RemoteMethodMapgetRemoteMethodMap () const
virtual HistogramDistributiondeepCopy (CopiesMap &copies) const
virtual void train ()
 trains the model
virtual void computeOutput (const Vec &input, Vec &output)
 computes the output of a trained model
virtual real log_density (const Vec &x) const
 return log of probability density log(p(x))
virtual real density (const Vec &x) const
 return probability density p(x) [ default version returns exp(log_density(x)) ]
virtual real survival_fn (const Vec &x) const
 return survival fn = P(X>x)
virtual real cdf (const Vec &x) const
 return survival fn = P(X<x)
virtual void expectation (Vec &mu) const
 return E[X]
virtual void variance (Mat &cov) const
 return Var[X]
virtual real prob_in_range (const Vec &x0, const Vec &x1) const
 return P(x0 < X < x1)
int find_bin (real x) const
 Find the bin where x belongs; -1 if x is out of range.
void calc_density_from_survival ()
 calculate bin_density from survival_values
void calc_survival_from_density ()
 calculate survival_values from bin_density

Static Public Member Functions

static string _classname_ ()
 Declares name and deepCopy methods.
static OptionList_getOptionList_ ()
static RemoteMethodMap_getRemoteMethodMap_ ()
static Object_new_instance_for_typemap_ ()
static bool _isa_ (const Object *o)
static void _static_initialize_ ()
static const PPathdeclaringFile ()
static void calc_density_from_survival (const Vec &survival, Vec &density_, const Vec &positions)
 calculate density from survival - static, on 2 Vecs
static void calc_survival_from_density (const Vec &density_, Vec &survival, const Vec &positions)
 calculate survival from density - static, on 2 Vecs

Public Attributes

Vec bin_positions
 there is one more bin position than number of bins, all the bins are supposed adjacent
Vec bin_density
 the density is supposed constant within each bin: p(x) = bin_density[i] if bin_positions[i] < x <= bin_positions[i+1]
Vec survival_values
 redundant with density is the pre-computed (optional) survival fn
PP< Binnerbinner
 this Binner is used to do binning at training time.
PP< Smoothersmoother
 this Smoother is used at training time.
bool smooth_density_instead_of_survival_fn
 whether to smooth the density or the survival function

Static Public Attributes

static StaticInitializer _static_initializer_

Static Protected Member Functions

static void declareOptions (OptionList &ol)
 Declares this class' options.

Private Member Functions

void build_ ()
 This does the actual building.

Detailed Description

Definition at line 50 of file HistogramDistribution.h.


Member Typedef Documentation

Reimplemented from PLearn::PDistribution.

Definition at line 77 of file HistogramDistribution.h.


Constructor & Destructor Documentation

PLearn::HistogramDistribution::HistogramDistribution ( )

Definition at line 47 of file HistogramDistribution.cc.

{}
PLearn::HistogramDistribution::HistogramDistribution ( VMat  data,
PP< Binner binner_ = 0,
PP< Smoother smoother_ = 0 
)

This constructor calls train as part of the construction process.

The computeOutput function can then be used right away.

Definition at line 49 of file HistogramDistribution.cc.

References PLearn::PLearner::setTrainingSet(), and train().

    :bin_positions(data.length()+1), bin_density(data.length()), survival_values(data.length()),
     binner(binner_), smoother(smoother_)
{
    setTrainingSet(data);
    train();
}

Here is the call graph for this function:


Member Function Documentation

string PLearn::HistogramDistribution::_classname_ ( ) [static]

Declares name and deepCopy methods.

Reimplemented from PLearn::PDistribution.

Definition at line 65 of file HistogramDistribution.cc.

OptionList & PLearn::HistogramDistribution::_getOptionList_ ( ) [static]

Reimplemented from PLearn::PDistribution.

Definition at line 65 of file HistogramDistribution.cc.

RemoteMethodMap & PLearn::HistogramDistribution::_getRemoteMethodMap_ ( ) [static]

Reimplemented from PLearn::PDistribution.

Definition at line 65 of file HistogramDistribution.cc.

bool PLearn::HistogramDistribution::_isa_ ( const Object o) [static]

Reimplemented from PLearn::PDistribution.

Definition at line 65 of file HistogramDistribution.cc.

Object * PLearn::HistogramDistribution::_new_instance_for_typemap_ ( ) [static]

Reimplemented from PLearn::PDistribution.

Definition at line 65 of file HistogramDistribution.cc.

StaticInitializer HistogramDistribution::_static_initializer_ & PLearn::HistogramDistribution::_static_initialize_ ( ) [static]

Reimplemented from PLearn::PDistribution.

Definition at line 65 of file HistogramDistribution.cc.

void PLearn::HistogramDistribution::build ( ) [virtual]

Simply calls inherited::build() then build_().

Reimplemented from PLearn::PDistribution.

Definition at line 102 of file HistogramDistribution.cc.

References PLearn::PDistribution::build(), and build_().

Here is the call graph for this function:

void PLearn::HistogramDistribution::build_ ( ) [private]

This does the actual building.

Reimplemented from PLearn::PDistribution.

Definition at line 97 of file HistogramDistribution.cc.

Referenced by build().

{
}

Here is the caller graph for this function:

void PLearn::HistogramDistribution::calc_density_from_survival ( )

calculate bin_density from survival_values

Definition at line 288 of file HistogramDistribution.cc.

References bin_density, bin_positions, and survival_values.

Referenced by train().

{
    calc_density_from_survival(survival_values, bin_density, bin_positions);
    /*
      int n= bin_positions.length()-1;
      bin_density.resize(n);
      real sum= 0.0;
      for(int i= 0; i < n; ++i)
      if(bin_positions[i+1] != bin_positions[i])
      if(i == n-1)
      sum+= (bin_density[i]= survival_values[i] / (bin_positions[i+1]-bin_positions[i]));
      else
      sum+= (bin_density[i]= (survival_values[i] - survival_values[i+1]) / (bin_positions[i+1]-bin_positions[i]));
      else
      bin_density[i]= 0.0;
    */
}

Here is the caller graph for this function:

void PLearn::HistogramDistribution::calc_density_from_survival ( const Vec survival,
Vec density_,
const Vec positions 
) [static]

calculate density from survival - static, on 2 Vecs

Definition at line 321 of file HistogramDistribution.cc.

References i, PLearn::TVec< T >::length(), n, PLearn::TVec< T >::resize(), and PLearn::sum().

{
    int n= positions.length()-1;
    density_.resize(n);
    real sum= 0.0;
    for(int i= 0; i < n; ++i)
        if(positions[i+1] != positions[i])
            if(i == n-1)
                sum+= (density_[i]= survival[i] / (positions[i+1]-positions[i]));
            else
                sum+= (density_[i]= (survival[i] - survival[i+1]) / (positions[i+1]-positions[i]));
        else
            density_[i]= 0.0;
}

Here is the call graph for this function:

void PLearn::HistogramDistribution::calc_survival_from_density ( )

calculate survival_values from bin_density

Definition at line 307 of file HistogramDistribution.cc.

References bin_density, bin_positions, and survival_values.

Referenced by train().

{
    calc_survival_from_density(bin_density, survival_values, bin_positions);
    /*
      int n= bin_positions.length()-1;
      survival_values.resize(n);
      real prec= 0.0;
      for(int i= n-1; i >= 0; --i)
      prec= survival_values[i]= bin_density[i]*(bin_positions[i+1]-bin_positions[i]) + prec;
      for(int i= 0; i < n; ++i)
      survival_values[i]/= prec;
    */
}

Here is the caller graph for this function:

void PLearn::HistogramDistribution::calc_survival_from_density ( const Vec density_,
Vec survival,
const Vec positions 
) [static]

calculate survival from density - static, on 2 Vecs

Definition at line 336 of file HistogramDistribution.cc.

References i, PLearn::TVec< T >::length(), n, and PLearn::TVec< T >::resize().

{
    int n= positions.length()-1;
    survival.resize(n);
    real prec= 0.0;
    for(int i= n-1; i >= 0; --i)
        prec= survival[i]= density_[i]*(positions[i+1]-positions[i]) + prec;
    for(int i= 0; i < n; ++i)
        survival[i]/= prec;
}

Here is the call graph for this function:

real PLearn::HistogramDistribution::cdf ( const Vec x) const [virtual]

return survival fn = P(X<x)

Reimplemented from PLearn::PDistribution.

Definition at line 231 of file HistogramDistribution.cc.

References survival_fn().

Referenced by computeOutput().

{
    return 1.0-survival_fn(x);
}

Here is the call graph for this function:

Here is the caller graph for this function:

string PLearn::HistogramDistribution::classname ( ) const [virtual]

Reimplemented from PLearn::PDistribution.

Definition at line 65 of file HistogramDistribution.cc.

void PLearn::HistogramDistribution::computeOutput ( const Vec input,
Vec output 
) [virtual]

computes the output of a trained model

Definition at line 173 of file HistogramDistribution.cc.

References cdf(), density(), expectation(), log_density(), m, PLearn::PDistribution::outputs_def, PLERROR, PLearn::TVec< T >::size(), survival_fn(), and variance().

{
    if(input.size() != 1 || output.size() != 1)
        PLERROR("In HistogramDistribution::use  implemented only for reals; i.e. input.size()=output.size()=1.  "
                "Got input.size()=%d and output.size()=%d", input.size(), output.size());
    // outputs_def: 'l'->log_density, 'd' -> density, 'c' -> cdf, 's' -> survival_fn, 'e' -> expectation, 'v' -> variance
    if(outputs_def == "l") output[0]= log_density(input);
    else if(outputs_def == "d") output[0]= density(input);
    else if(outputs_def == "c") output[0]= cdf(input);
    else if(outputs_def == "s") output[0]= survival_fn(input);
    else if(outputs_def == "e") { Vec mu(1); expectation(mu); output[0]= mu[0]; }
    else if(outputs_def == "v") { Mat m(1,1); variance(m); output[0]= m(0,0); }
    else PLERROR("In HistogramDistribution::use  unknown value for outputs_def= \"%s\"", outputs_def.c_str());
}

Here is the call graph for this function:

void PLearn::HistogramDistribution::declareOptions ( OptionList ol) [static, protected]

Declares this class' options.

Reimplemented from PLearn::PDistribution.

Definition at line 67 of file HistogramDistribution.cc.

References bin_density, bin_positions, binner, PLearn::OptionBase::buildoption, PLearn::declareOption(), PLearn::PDistribution::declareOptions(), PLearn::OptionBase::learntoption, smooth_density_instead_of_survival_fn, smoother, and survival_values.

{
    declareOption(ol, "bin_positions", &HistogramDistribution::bin_positions, OptionBase::learntoption,
                  "The n+1 positions that define n bins. There is one more bin position "
                  "than number of bins, all the bins are supposed adjacent.");

    declareOption(ol, "bin_density", &HistogramDistribution::bin_density, OptionBase::learntoption,
                  "Density of the distribution for each bin.  The density is supposed "
                  "constant within each bin:\n"
                  "\t p(x) = bin_density[i] if bin_positions[i] < x <= bin_positions[i+1].");

    declareOption(ol, "survival_values", &HistogramDistribution::survival_values, OptionBase::learntoption,
                  "Redundant with density is the pre-computed survival function.");

    declareOption(ol, "binner", &HistogramDistribution::binner, OptionBase::buildoption,
                  "Used to do binning at training time (although a fixed binning scheme can be\n"
                  "obtained by using a ManualBinner.B)");

    declareOption(ol, "smoother", &HistogramDistribution::smoother, OptionBase::buildoption,
                  "Used to smooth learned density (or survival) at train time, after the empirical\n"
                  "frequencies of each bin have been collected\n");

    declareOption(ol, "smooth_density_instead_of_survival_fn",
                  &HistogramDistribution::smooth_density_instead_of_survival_fn, OptionBase::buildoption,
                  "whether to smooth the density or the survival function, with the smoother\n");

    // Now call the parent class' declareOptions
    inherited::declareOptions(ol);
}

Here is the call graph for this function:

static const PPath& PLearn::HistogramDistribution::declaringFile ( ) [inline, static]

Reimplemented from PLearn::PDistribution.

Definition at line 120 of file HistogramDistribution.h.

:
HistogramDistribution * PLearn::HistogramDistribution::deepCopy ( CopiesMap copies) const [virtual]

Reimplemented from PLearn::PDistribution.

Definition at line 65 of file HistogramDistribution.cc.

real PLearn::HistogramDistribution::density ( const Vec x) const [virtual]

return probability density p(x) [ default version returns exp(log_density(x)) ]

Reimplemented from PLearn::PDistribution.

Definition at line 205 of file HistogramDistribution.cc.

References bin_density, find_bin(), PLERROR, and PLearn::TVec< T >::size().

Referenced by computeOutput(), and log_density().

{
    if(x.size() != 1)
        PLERROR("HistogramDistribution::density implemented only for univariate data (vec size == 1).");
    return bin_density[find_bin(x[0])];
}

Here is the call graph for this function:

Here is the caller graph for this function:

void PLearn::HistogramDistribution::expectation ( Vec mu) const [virtual]

return E[X]

Reimplemented from PLearn::PDistribution.

Definition at line 236 of file HistogramDistribution.cc.

References bin_density, bin_positions, i, PLERROR, PLearn::TVec< T >::size(), and PLearn::sum().

Referenced by computeOutput().

{
    if(mu.size() != 1)
        PLERROR("HistogramDistribution::expectation implemented only for univariate data (vec size == 1).");
    real sum= 0.0;
    for(int i= 0; i < bin_density.size(); ++i)
        sum+= bin_density[i] * (bin_positions[i+1]-bin_positions[i]) * (bin_positions[i]+bin_positions[i+1])/2;
    //    sum+= bin_density[i] * bin_positions[i+1];
    mu[0]=sum;
}

Here is the call graph for this function:

Here is the caller graph for this function:

int PLearn::HistogramDistribution::find_bin ( real  x) const

Find the bin where x belongs; -1 if x is out of range.

Definition at line 268 of file HistogramDistribution.cc.

References b, bin_positions, and PLearn::TVec< T >::length().

Referenced by density(), and survival_fn().

{
    int b= 0, e= bin_positions.length()-2, p= b+(e-b)/2;

    if(x < bin_positions[b] || x >= bin_positions[e+1])
        return -1;

    while(b < e)
    {
        if(bin_positions[p] == x)
            return p;
        if(bin_positions[p] > x)
            e= p-1;
        else
            b= p+1;
        p= b+(e-b)/2;
    }
    return p;
}

Here is the call graph for this function:

Here is the caller graph for this function:

OptionList & PLearn::HistogramDistribution::getOptionList ( ) const [virtual]

Reimplemented from PLearn::PDistribution.

Definition at line 65 of file HistogramDistribution.cc.

OptionMap & PLearn::HistogramDistribution::getOptionMap ( ) const [virtual]

Reimplemented from PLearn::PDistribution.

Definition at line 65 of file HistogramDistribution.cc.

RemoteMethodMap & PLearn::HistogramDistribution::getRemoteMethodMap ( ) const [virtual]

Reimplemented from PLearn::PDistribution.

Definition at line 65 of file HistogramDistribution.cc.

real PLearn::HistogramDistribution::log_density ( const Vec x) const [virtual]

return log of probability density log(p(x))

Reimplemented from PLearn::PDistribution.

Definition at line 199 of file HistogramDistribution.cc.

References density(), and pl_log.

Referenced by computeOutput().

{
    return pl_log(density(x));
}

Here is the call graph for this function:

Here is the caller graph for this function:

void PLearn::HistogramDistribution::makeDeepCopyFromShallowCopy ( CopiesMap copies) [virtual]

Transforms a shallow copy into a deep copy.

Reimplemented from PLearn::PDistribution.

Definition at line 188 of file HistogramDistribution.cc.

References bin_density, bin_positions, binner, PLearn::deepCopyField(), PLearn::PDistribution::makeDeepCopyFromShallowCopy(), smoother, and survival_values.

Here is the call graph for this function:

real PLearn::HistogramDistribution::prob_in_range ( const Vec x0,
const Vec x1 
) const [virtual]

return P(x0 < X < x1)

Definition at line 262 of file HistogramDistribution.cc.

References survival_fn().

{
    return survival_fn(x0) - survival_fn(x1);
}

Here is the call graph for this function:

real PLearn::HistogramDistribution::survival_fn ( const Vec x) const [virtual]

return survival fn = P(X>x)

Reimplemented from PLearn::PDistribution.

Definition at line 213 of file HistogramDistribution.cc.

References bin_positions, find_bin(), PLERROR, PLearn::TVec< T >::size(), and survival_values.

Referenced by cdf(), computeOutput(), and prob_in_range().

{
    if(x.size() != 1)
        PLERROR("HistogramDistribution::survival_fn implemented only for univariate data (vec size == 1).");
    int bin= find_bin(x[0]);
    if(bin < 0)
        if(x[0] < bin_positions[0])
            return 1.0;
        else
            return 0.0;

    if(x[0] < bin_positions[bin] && bin >= 1)
        return survival_values[bin-1] + (x[0] - bin_positions[bin-1]) *
            (survival_values[bin] - survival_values[bin-1]) / (bin_positions[bin] - bin_positions[bin-1]);

    return survival_values[bin];
}

Here is the call graph for this function:

Here is the caller graph for this function:

void PLearn::HistogramDistribution::train ( ) [virtual]

trains the model

Reimplemented from PLearn::PDistribution.

Definition at line 108 of file HistogramDistribution.cc.

References bin_density, bin_positions, binner, calc_density_from_survival(), calc_survival_from_density(), PLearn::VMat::getColumn(), i, PLearn::PLearner::inputsize(), PLearn::TVec< T >::length(), PLearn::VMat::length(), PLERROR, PLearn::TVec< T >::resize(), smooth_density_instead_of_survival_fn, smoother, survival_values, PLearn::PLearner::targetsize(), PLearn::PLearner::train_set, and PLearn::VMat::width().

Referenced by HistogramDistribution().

{

    /*
      - prend la distri empirique
      | trie les points
      | merge les bins (possiblement sous contraintes)
      |     - points de coupure predefinis (option include_cutpoints) ManualBinner
      |     - largeur des bins > a une valeur minimale
      |     - bins contenir un minimum de points
      Binner

      Smoother
      (recalcule la densite)

      calculer survival_values
    */

    if(train_set->width() != inputsize()+targetsize())
        PLERROR("In HistogramDistribution::train(VMat training_set) training_set->width() != inputsize()+targetsize()");
    if(train_set->width() != 1)
        PLERROR("In HistogramDistribution::train() train_set->width() must be 1 (column vec.)");
    if(binner == 0)
        PLERROR("In HistogramDistribution::train() Can't train without a Binner.");

    Vec data(train_set.length());
    data << train_set.getColumn(train_set.width()-1);

    PP<RealMapping> binning= binner->getBinning(train_set);
    binning->setMappingForOther(0.0);
    binning->transform(data);

    bin_positions= binning->getCutPoints();
    bin_density.resize(bin_positions.length()-1);
    survival_values.resize(bin_positions.length()-1);

    for(int i= 0; i < data.length(); ++i)
        ++survival_values[static_cast<int>(data[i])];
    for(int i= survival_values.length()-2; i >= 0; --i)
        survival_values[i]+= survival_values[i+1];
    for(int i= survival_values.length()-1; i >= 0; --i)
        survival_values[i]/= survival_values[0];

    if(smoother)
    {
        if (smooth_density_instead_of_survival_fn)
        {
            calc_density_from_survival();
            Vec df(bin_density.length());
            df << bin_density;
            smoother->smooth(df, bin_density, bin_positions, bin_positions);
            calc_survival_from_density();
        }
        else
        {
            Vec sv(survival_values.length());
            sv << survival_values;
            smoother->smooth(sv, survival_values, bin_positions, bin_positions);
            calc_density_from_survival();
        }
    }
    else
        calc_density_from_survival();
}

Here is the call graph for this function:

Here is the caller graph for this function:

void PLearn::HistogramDistribution::variance ( Mat cov) const [virtual]

return Var[X]

Reimplemented from PLearn::PDistribution.

Definition at line 247 of file HistogramDistribution.cc.

References PLearn::abs(), bin_density, bin_positions, i, n, PLERROR, PLearn::TVec< T >::size(), PLearn::TMat< T >::size(), and PLearn::sum().

Referenced by computeOutput().

{
    if(cov.size() != 1)
        PLERROR("HistogramDistribution::variance implemented only for univariate data");
    real sumsq= 0.0, sum= 0.0, s;
    int n= bin_density.size();
    for(int i= 0; i < n; ++i)
    {
        s= bin_density[i] * (bin_positions[i+1]-bin_positions[i]) * (bin_positions[i]+bin_positions[i+1])/2;
        sum+= s;
        sumsq+= s*s;
    }
    cov(0,0) = abs(sumsq-(sum*sum)/n)/n;
}

Here is the call graph for this function:

Here is the caller graph for this function:


Member Data Documentation

Reimplemented from PLearn::PDistribution.

Definition at line 120 of file HistogramDistribution.h.

the density is supposed constant within each bin: p(x) = bin_density[i] if bin_positions[i] < x <= bin_positions[i+1]

Definition at line 63 of file HistogramDistribution.h.

Referenced by calc_density_from_survival(), calc_survival_from_density(), declareOptions(), density(), expectation(), makeDeepCopyFromShallowCopy(), train(), and variance().

there is one more bin position than number of bins, all the bins are supposed adjacent

Definition at line 59 of file HistogramDistribution.h.

Referenced by calc_density_from_survival(), calc_survival_from_density(), declareOptions(), expectation(), find_bin(), makeDeepCopyFromShallowCopy(), survival_fn(), train(), and variance().

this Binner is used to do binning at training time.

Definition at line 69 of file HistogramDistribution.h.

Referenced by declareOptions(), makeDeepCopyFromShallowCopy(), and train().

whether to smooth the density or the survival function

Definition at line 75 of file HistogramDistribution.h.

Referenced by declareOptions(), and train().

this Smoother is used at training time.

Definition at line 72 of file HistogramDistribution.h.

Referenced by declareOptions(), makeDeepCopyFromShallowCopy(), and train().

redundant with density is the pre-computed (optional) survival fn

Definition at line 66 of file HistogramDistribution.h.

Referenced by calc_density_from_survival(), calc_survival_from_density(), declareOptions(), makeDeepCopyFromShallowCopy(), survival_fn(), and train().


The documentation for this class was generated from the following files:
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Defines