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

Stores and learns the parameters between two linear layers of an RBM. More...

#include <RBMMatrixConnectionNatGrad.h>

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

List of all members.

Public Member Functions

 RBMMatrixConnectionNatGrad (real the_learning_rate=0)
 Default constructor.
virtual void update (const Mat &pos_down_values, const Mat &pos_up_values, const Mat &neg_down_values, const Mat &neg_up_values)
 Not implemented.
virtual void bpropUpdate (const Mat &inputs, const Mat &outputs, Mat &input_gradients, const Mat &output_gradients, bool accumulate=false)
 SOON TO BE DEPRECATED, USE bpropAccUpdate(const TVec<Mat*>& ports_value, const TVec<Mat*>& ports_gradient)
virtual string classname () const
virtual OptionListgetOptionList () const
virtual OptionMapgetOptionMap () const
virtual RemoteMethodMapgetRemoteMethodMap () const
virtual
RBMMatrixConnectionNatGrad
deepCopy (CopiesMap &copies) const
virtual void build ()
 Post-constructor.
virtual void makeDeepCopyFromShallowCopy (CopiesMap &copies)
 Transforms a shallow copy into a deep copy.

Static Public Member Functions

static string _classname_ ()
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 ()

Public Attributes

PP< NatGradEstimatornatgrad_template
 natural gradient estimator for neurons (if 0 then do not correct the gradient on neurons)
TVec< PP< NatGradEstimator > > cd_natgrad
TVec< PP< NatGradEstimator > > bp_natgrad
Mat weights_gradient
Vec natural_gradient

Static Public Attributes

static StaticInitializer _static_initializer_

Static Protected Member Functions

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

Private Types

typedef RBMMatrixConnection inherited

Private Member Functions

void build_ ()
 This does the actual building.

Detailed Description

Stores and learns the parameters between two linear layers of an RBM.

Definition at line 54 of file RBMMatrixConnectionNatGrad.h.


Member Typedef Documentation

Reimplemented from PLearn::RBMMatrixConnection.

Definition at line 56 of file RBMMatrixConnectionNatGrad.h.


Constructor & Destructor Documentation

PLearn::RBMMatrixConnectionNatGrad::RBMMatrixConnectionNatGrad ( real  the_learning_rate = 0)

Default constructor.

Definition at line 55 of file RBMMatrixConnectionNatGrad.cc.

                                                                               :
    inherited(the_learning_rate)
{
}

Member Function Documentation

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

Reimplemented from PLearn::RBMMatrixConnection.

Definition at line 53 of file RBMMatrixConnectionNatGrad.cc.

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

Reimplemented from PLearn::RBMMatrixConnection.

Definition at line 53 of file RBMMatrixConnectionNatGrad.cc.

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

Reimplemented from PLearn::RBMMatrixConnection.

Definition at line 53 of file RBMMatrixConnectionNatGrad.cc.

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

Reimplemented from PLearn::RBMMatrixConnection.

Definition at line 53 of file RBMMatrixConnectionNatGrad.cc.

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

Reimplemented from PLearn::RBMMatrixConnection.

Definition at line 53 of file RBMMatrixConnectionNatGrad.cc.

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

Reimplemented from PLearn::RBMMatrixConnection.

Definition at line 53 of file RBMMatrixConnectionNatGrad.cc.

void PLearn::RBMMatrixConnectionNatGrad::bpropUpdate ( const Mat inputs,
const Mat outputs,
Mat input_gradients,
const Mat output_gradients,
bool  accumulate = false 
) [virtual]

SOON TO BE DEPRECATED, USE bpropAccUpdate(const TVec<Mat*>& ports_value, const TVec<Mat*>& ports_gradient)

Reimplemented from PLearn::RBMMatrixConnection.

Definition at line 138 of file RBMMatrixConnectionNatGrad.cc.

References bp_natgrad, PLearn::RBMConnection::down_size, i, PLearn::RBMConnection::learning_rate, PLearn::TMat< T >::length(), PLearn::multiplyAcc(), natural_gradient, PLASSERT, PLASSERT_MSG, PLearn::RBMConnection::pos_count, PLearn::product(), PLearn::productAcc(), PLearn::productScaleAcc(), PLearn::TMat< T >::resize(), PLearn::RBMConnection::up_size, PLearn::RBMMatrixConnection::weights, weights_gradient, and PLearn::TMat< T >::width().

{
    PLASSERT( inputs.width() == down_size );
    PLASSERT( outputs.width() == up_size );
    PLASSERT( output_gradients.width() == up_size );

    if( accumulate )
    {
        PLASSERT_MSG( input_gradients.width() == down_size &&
                      input_gradients.length() == inputs.length(),
                      "Cannot resize input_gradients and accumulate into it" );

        // input_gradients += output_gradient * weights
        productAcc(input_gradients, output_gradients, weights);
    }
    else
    {
        input_gradients.resize(inputs.length(), down_size);
        // input_gradients = output_gradient * weights
        product(input_gradients, output_gradients, weights);
    }

    // weights_gradient = 1/n * output_gradients' * inputs
    productScaleAcc(weights_gradient, output_gradients, true, inputs, false,
                    1. / inputs.length(), 0.);
    for (int i=0;i<up_size;i++)
    {
        (*bp_natgrad[i])(pos_count,weights_gradient(i),natural_gradient);
        multiplyAcc(weights(i),natural_gradient,-learning_rate);
    }
    pos_count++;
}

Here is the call graph for this function:

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

Post-constructor.

The normal implementation should call simply inherited::build(), then this class's build_(). This method should be callable again at later times, after modifying some option fields to change the "architecture" of the object.

Reimplemented from PLearn::RBMMatrixConnection.

Definition at line 86 of file RBMMatrixConnectionNatGrad.cc.

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

Here is the call graph for this function:

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

This does the actual building.

Reimplemented from PLearn::RBMMatrixConnection.

Definition at line 73 of file RBMMatrixConnectionNatGrad.cc.

References bp_natgrad, cd_natgrad, deepCopy(), PLearn::RBMConnection::down_size, i, natgrad_template, natural_gradient, PLearn::TMat< T >::resize(), PLearn::TVec< T >::resize(), PLearn::RBMConnection::up_size, and weights_gradient.

Referenced by build().

Here is the call graph for this function:

Here is the caller graph for this function:

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

Reimplemented from PLearn::RBMMatrixConnection.

Definition at line 53 of file RBMMatrixConnectionNatGrad.cc.

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

Declares the class options.

Reimplemented from PLearn::RBMMatrixConnection.

Definition at line 60 of file RBMMatrixConnectionNatGrad.cc.

References PLearn::declareOption(), PLearn::RBMMatrixConnection::declareOptions(), PLearn::OptionBase::learntoption, and natgrad_template.

{
    declareOption(ol, "natgrad_template", &RBMMatrixConnectionNatGrad::natgrad_template,
                  OptionBase::learntoption,
                  "An object of type NatGradEstimator which will be copied for each row of the\n"
                  "weights matrix; each will compute the adjustment to the update direction\n"
                  "based on the natural gradient.\n");


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

Here is the call graph for this function:

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

Reimplemented from PLearn::RBMMatrixConnection.

Definition at line 95 of file RBMMatrixConnectionNatGrad.h.

:
    //#####  Protected Member Functions  ######################################
RBMMatrixConnectionNatGrad * PLearn::RBMMatrixConnectionNatGrad::deepCopy ( CopiesMap copies) const [virtual]

Reimplemented from PLearn::RBMMatrixConnection.

Definition at line 53 of file RBMMatrixConnectionNatGrad.cc.

Referenced by build_().

Here is the caller graph for this function:

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

Reimplemented from PLearn::RBMMatrixConnection.

Definition at line 53 of file RBMMatrixConnectionNatGrad.cc.

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

Reimplemented from PLearn::RBMMatrixConnection.

Definition at line 53 of file RBMMatrixConnectionNatGrad.cc.

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

Reimplemented from PLearn::RBMMatrixConnection.

Definition at line 53 of file RBMMatrixConnectionNatGrad.cc.

void PLearn::RBMMatrixConnectionNatGrad::makeDeepCopyFromShallowCopy ( CopiesMap copies) [virtual]
void PLearn::RBMMatrixConnectionNatGrad::update ( const Mat pos_down_values,
const Mat pos_up_values,
const Mat neg_down_values,
const Mat neg_up_values 
) [virtual]

Not implemented.

Reimplemented from PLearn::RBMMatrixConnection.

Definition at line 104 of file RBMMatrixConnectionNatGrad.cc.

References cd_natgrad, i, PLearn::RBMConnection::learning_rate, PLearn::TMat< T >::length(), PLearn::RBMConnection::momentum, PLearn::multiplyAcc(), natural_gradient, PLASSERT, PLERROR, PLearn::RBMConnection::pos_count, PLearn::productScaleAcc(), PLearn::RBMConnection::up_size, PLearn::RBMMatrixConnection::weights, weights_gradient, and PLearn::TMat< T >::width().

{
    // weights -= learning_rate * ( h_0 v_0' - h_1 v_1' );
    // or:
    // weights[i][j] += learning_rate * (h_1[i] v_1[j] - h_0[i] v_0[j]);

    PLASSERT( pos_up_values.width() == weights.length() );
    PLASSERT( neg_up_values.width() == weights.length() );
    PLASSERT( pos_down_values.width() == weights.width() );
    PLASSERT( neg_down_values.width() == weights.width() );
    if( momentum == real(0.) )
    {
        // We use the average gradient over a mini-batch.
        real mbnorm = 1. / pos_down_values.length();
        productScaleAcc(weights_gradient, pos_up_values, true, pos_down_values, false,
                        mbnorm, 0.);
        productScaleAcc(weights_gradient, neg_up_values, true, neg_down_values, false,
                        -mbnorm, 1.);

        for (int i=0;i<up_size;i++)
        {
            (*cd_natgrad[i])(pos_count,weights_gradient(i),natural_gradient);
            multiplyAcc(weights(i),natural_gradient,-learning_rate);
        }
        pos_count++;
    }
    else
        PLERROR("RBMMatrixConnectionNatGrad::update with momentum - Not implemented");
}

Here is the call graph for this function:


Member Data Documentation

Reimplemented from PLearn::RBMMatrixConnection.

Definition at line 95 of file RBMMatrixConnectionNatGrad.h.

Definition at line 67 of file RBMMatrixConnectionNatGrad.h.

Referenced by bpropUpdate(), build_(), and makeDeepCopyFromShallowCopy().

Definition at line 66 of file RBMMatrixConnectionNatGrad.h.

Referenced by build_(), makeDeepCopyFromShallowCopy(), and update().

natural gradient estimator for neurons (if 0 then do not correct the gradient on neurons)

Definition at line 63 of file RBMMatrixConnectionNatGrad.h.

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


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