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::RBMMatrixTransposeConnection Class Reference

RBMConnection which uses the tranpose of some other RBMMatrixConnection's weights. More...

#include <RBMMatrixTransposeConnection.h>

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

List of all members.

Public Member Functions

 RBMMatrixTransposeConnection (PP< RBMMatrixConnection > the_rbm_matrix_connection=0, real the_learning_rate=0, bool call_build_=false)
 Default constructor.
virtual void accumulatePosStats (const Vec &down_values, const Vec &up_values)
 Accumulates positive phase statistics to *_pos_stats.
virtual void accumulatePosStats (const Mat &down_values, const Mat &up_values)
virtual void accumulateNegStats (const Vec &down_values, const Vec &up_values)
 Accumulates negative phase statistics to *_neg_stats.
virtual void accumulateNegStats (const Mat &down_values, const Mat &up_values)
virtual void update ()
 Updates parameters according to contrastive divergence gradient.
virtual void update (const Vec &pos_down_values, const Vec &pos_up_values, const Vec &neg_down_values, const Vec &neg_up_values)
 Updates parameters according to contrastive divergence gradient, not using the statistics but the explicit values passed.
virtual void clearStats ()
 Clear all information accumulated during stats.
virtual void computeProduct (int start, int length, const Vec &activations, bool accumulate=false) const
 Computes the vectors of activation of "length" units, starting from "start", and stores (or add) them into "activations".
virtual void computeProducts (int start, int length, Mat &activations, bool accumulate=false) const
 Same as 'computeProduct' but for mini-batches.
virtual void bpropUpdate (const Vec &input, const Vec &output, Vec &input_gradient, const Vec &output_gradient, bool accumulate=false)
 Adapt based on the output gradient: this method should only be called just after a corresponding fprop; it should be called with the same arguments as fprop for the first two arguments (and output should not have been modified since then).
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 void forget ()
 reset the parameters to the state they would be BEFORE starting training.
virtual int nParameters () const
 optionally perform some processing after training, or after a series of fprop/bpropUpdate calls to prepare the model for truly out-of-sample operation.
virtual Vec makeParametersPointHere (const Vec &global_parameters)
 Make the parameters data be sub-vectors of the given global_parameters.
virtual string classname () const
virtual OptionListgetOptionList () const
virtual OptionMapgetOptionMap () const
virtual RemoteMethodMapgetRemoteMethodMap () const
virtual
RBMMatrixTransposeConnection
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

Mat weights
 Matrix containing unit-to-unit weights ( $output_size \times input_size$)
PP< RBMMatrixConnectionrbm_matrix_connection
 RBMMatrixConnection from which the weights are taken.
bool learn_scale
 Indication that the scale of the weight matrix should be learned.
real scale
 Learned scale for weight matrix.
Mat weights_pos_stats
 Accumulates positive contribution to the weights' gradient.
Mat weights_neg_stats
 Accumulates negative contribution to the weights' gradient.
Mat weights_inc
 Used if momentum != 0.

Static Public Attributes

static StaticInitializer _static_initializer_

Static Protected Member Functions

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

Private Types

typedef RBMConnection inherited

Private Member Functions

void build_ ()
 This does the actual building.

Detailed Description

RBMConnection which uses the tranpose of some other RBMMatrixConnection's weights.

Definition at line 53 of file RBMMatrixTransposeConnection.h.


Member Typedef Documentation

Reimplemented from PLearn::RBMConnection.

Definition at line 55 of file RBMMatrixTransposeConnection.h.


Constructor & Destructor Documentation

PLearn::RBMMatrixTransposeConnection::RBMMatrixTransposeConnection ( PP< RBMMatrixConnection the_rbm_matrix_connection = 0,
real  the_learning_rate = 0,
bool  call_build_ = false 
)

Default constructor.

Definition at line 51 of file RBMMatrixTransposeConnection.cc.

References build_().

                      :
    inherited(the_learning_rate, call_build_),
    rbm_matrix_connection(the_rbm_matrix_connection),
    learn_scale( false ),
    scale( 1.0 )
{
    if (call_build_)
        build_();
}

Here is the call graph for this function:


Member Function Documentation

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

Reimplemented from PLearn::RBMConnection.

Definition at line 49 of file RBMMatrixTransposeConnection.cc.

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

Reimplemented from PLearn::RBMConnection.

Definition at line 49 of file RBMMatrixTransposeConnection.cc.

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

Reimplemented from PLearn::RBMConnection.

Definition at line 49 of file RBMMatrixTransposeConnection.cc.

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

Reimplemented from PLearn::RBMConnection.

Definition at line 49 of file RBMMatrixTransposeConnection.cc.

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

Reimplemented from PLearn::Object.

Definition at line 49 of file RBMMatrixTransposeConnection.cc.

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

Reimplemented from PLearn::RBMConnection.

Definition at line 49 of file RBMMatrixTransposeConnection.cc.

virtual void PLearn::RBMMatrixTransposeConnection::accumulateNegStats ( const Mat down_values,
const Mat up_values 
) [inline, virtual]

Implements PLearn::RBMConnection.

Definition at line 110 of file RBMMatrixTransposeConnection.h.

References PLASSERT_MSG.

    {
        PLASSERT_MSG( false, "Not implemented" );
    }
void PLearn::RBMMatrixTransposeConnection::accumulateNegStats ( const Vec down_values,
const Vec up_values 
) [virtual]

Accumulates negative phase statistics to *_neg_stats.

Implements PLearn::RBMConnection.

Definition at line 149 of file RBMMatrixTransposeConnection.cc.

References PLearn::externalProductAcc(), PLearn::RBMConnection::neg_count, and weights_neg_stats.

{
    // weights_neg_stats += down_values * up_values'
    externalProductAcc( weights_neg_stats, down_values, up_values );

    neg_count++;
}

Here is the call graph for this function:

virtual void PLearn::RBMMatrixTransposeConnection::accumulatePosStats ( const Mat down_values,
const Mat up_values 
) [inline, virtual]

Implements PLearn::RBMConnection.

Definition at line 100 of file RBMMatrixTransposeConnection.h.

References PLASSERT_MSG.

    {
        PLASSERT_MSG( false, "Not implemented" );
    }
void PLearn::RBMMatrixTransposeConnection::accumulatePosStats ( const Vec down_values,
const Vec up_values 
) [virtual]

Accumulates positive phase statistics to *_pos_stats.

Implements PLearn::RBMConnection.

Definition at line 140 of file RBMMatrixTransposeConnection.cc.

References PLearn::externalProductAcc(), PLearn::RBMConnection::pos_count, and weights_pos_stats.

{
    // weights_pos_stats += down_values * up_values'
    externalProductAcc( weights_pos_stats, down_values, up_values );

    pos_count++;
}

Here is the call graph for this function:

void PLearn::RBMMatrixTransposeConnection::bpropUpdate ( const Vec input,
const Vec output,
Vec input_gradient,
const Vec output_gradient,
bool  accumulate = false 
) [virtual]

Adapt based on the output gradient: this method should only be called just after a corresponding fprop; it should be called with the same arguments as fprop for the first two arguments (and output should not have been modified since then).

this version allows to obtain the input gradient as well

Since sub-classes are supposed to learn ONLINE, the object is 'ready-to-be-used' just after any bpropUpdate. N.B. A DEFAULT IMPLEMENTATION IS PROVIDED IN THE SUPER-CLASS, WHICH JUST CALLS bpropUpdate(input, output, input_gradient, output_gradient) AND IGNORES INPUT GRADIENT. this version allows to obtain the input gradient as well N.B. THE DEFAULT IMPLEMENTATION IN SUPER-CLASS JUST RAISES A PLERROR.

Reimplemented from PLearn::OnlineLearningModule.

Definition at line 356 of file RBMMatrixTransposeConnection.cc.

References PLearn::TVec< T >::data(), PLearn::RBMConnection::down_size, PLearn::externalProductScaleAcc(), i, in, j, learn_scale, PLearn::RBMConnection::learning_rate, PLearn::TMat< T >::length(), PLASSERT, PLASSERT_MSG, PLearn::product(), PLearn::productAcc(), rbm_matrix_connection, PLearn::TVec< T >::resize(), scale, PLearn::TVec< T >::size(), PLearn::RBMConnection::up_size, weights, and PLearn::TMat< T >::width().

{
    PLASSERT( input.size() == down_size );
    PLASSERT( output.size() == up_size );
    PLASSERT( output_gradient.size() == up_size );
    PLASSERT_MSG( rbm_matrix_connection, "RBMMatrixTransposeConnection must be given an rbm_matrix_connection.\n");

    if( accumulate )
    {
        PLASSERT_MSG( input_gradient.size() == down_size,
                      "Cannot resize input_gradient AND accumulate into it" );

        // input_gradient += weights' * output_gradient
        productAcc( input_gradient, weights, output_gradient );
    }
    else
    {
        input_gradient.resize( down_size );

        // input_gradient = weights' * output_gradient
        product( input_gradient, weights, output_gradient );
    }

    // weights -= learning_rate * output_gradient * input'
    externalProductScaleAcc( weights, input, output_gradient, -learning_rate );
    if( learn_scale )
    {
        real* in = input.data();
        real* out_g;
        real* wj;
        for( int j=0; j<weights.width(); j++)
        {
            out_g = output_gradient.data();
            wj = weights[j];
            for( int i=0; i<weights.length(); i++ )
                scale -= learning_rate * out_g[i] * wj[i] * in[j];
        }
    }
}

Here is the call graph for this function:

void PLearn::RBMMatrixTransposeConnection::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::OnlineLearningModule.

Definition at line 400 of file RBMMatrixTransposeConnection.cc.

References PLearn::RBMConnection::down_size, i, in, j, learn_scale, PLearn::RBMConnection::learning_rate, PLearn::TMat< T >::length(), PLASSERT, PLASSERT_MSG, PLearn::productTranspose(), PLearn::productTransposeAcc(), rbm_matrix_connection, PLearn::TMat< T >::resize(), scale, PLearn::transposeProductScaleAcc(), PLearn::RBMConnection::up_size, weights, and PLearn::TMat< T >::width().

{
    PLASSERT( inputs.width() == down_size );
    PLASSERT( outputs.width() == up_size );
    PLASSERT( output_gradients.width() == up_size );
    PLASSERT_MSG( rbm_matrix_connection, "RBMMatrixTransposeConnection must be given an rbm_matrix_connection.\n");

    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
        productTransposeAcc(input_gradients, output_gradients, weights);
    }
    else
    {
        input_gradients.resize(inputs.length(), down_size);
        // input_gradients = output_gradient * weights
        productTranspose(input_gradients, output_gradients, weights);
    }

    // weights -= learning_rate/n * output_gradients' * inputs
    transposeProductScaleAcc(weights, inputs, output_gradients,
                             -learning_rate / inputs.length(), real(1));

    if( learn_scale )
    {
        for( int t=0; t<inputs.length(); t++)
        {
            real* in = inputs[t];
            real* out_g;
            real* wj;
            for( int j=0; j<weights.width(); j++)
            {
                out_g = output_gradients[t];
                wj = weights[j];
                for( int i=0; i<weights.length(); i++ )
                    scale -= learning_rate * out_g[i] * wj[i] * in[j];
            }
        }
    }
}

Here is the call graph for this function:

void PLearn::RBMMatrixTransposeConnection::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::RBMConnection.

Definition at line 122 of file RBMMatrixTransposeConnection.cc.

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

Referenced by PLearn::DeepBeliefNet::train().

Here is the call graph for this function:

Here is the caller graph for this function:

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

This does the actual building.

Reimplemented from PLearn::RBMConnection.

Definition at line 93 of file RBMMatrixTransposeConnection.cc.

References clearStats(), PLearn::RBMConnection::down_size, PLearn::OnlineLearningModule::input_size, PLearn::RBMConnection::momentum, PLearn::OnlineLearningModule::output_size, PLearn::OnlineLearningModule::random_gen, rbm_matrix_connection, PLearn::TMat< T >::resize(), PLearn::RBMConnection::up_size, weights, weights_inc, weights_neg_stats, and weights_pos_stats.

Referenced by build(), and RBMMatrixTransposeConnection().

{
    if( !rbm_matrix_connection )
        return;

    // If we have a random_gen and rbm_matrix_connection does not, share it
    if( random_gen && !(rbm_matrix_connection->random_gen) )
    {
        rbm_matrix_connection->random_gen = random_gen;
        rbm_matrix_connection->forget();
    }
    weights = rbm_matrix_connection->weights;
    down_size = rbm_matrix_connection->up_size;
    up_size = rbm_matrix_connection->down_size;

    // For compatibility with OnlineLearningModule inherited functions
    input_size = down_size;
    output_size = up_size;


    weights_pos_stats.resize( down_size, up_size );
    weights_neg_stats.resize( down_size, up_size );

    if( momentum != 0. )
        weights_inc.resize( down_size, up_size );

    clearStats();
}

Here is the call graph for this function:

Here is the caller graph for this function:

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

Reimplemented from PLearn::Object.

Definition at line 49 of file RBMMatrixTransposeConnection.cc.

void PLearn::RBMMatrixTransposeConnection::clearStats ( ) [virtual]

Clear all information accumulated during stats.

Implements PLearn::RBMConnection.

Definition at line 270 of file RBMMatrixTransposeConnection.cc.

References PLearn::TMat< T >::clear(), PLearn::RBMConnection::neg_count, PLearn::RBMConnection::pos_count, weights_neg_stats, and weights_pos_stats.

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

{
    weights_pos_stats.clear();
    weights_neg_stats.clear();

    pos_count = 0;
    neg_count = 0;
}

Here is the call graph for this function:

Here is the caller graph for this function:

void PLearn::RBMMatrixTransposeConnection::computeProduct ( int  start,
int  length,
const Vec activations,
bool  accumulate = false 
) const [virtual]

Computes the vectors of activation of "length" units, starting from "start", and stores (or add) them into "activations".

"start" indexes an up unit if "going_up", else a down unit.

Implements PLearn::RBMConnection.

Definition at line 279 of file RBMMatrixTransposeConnection.cc.

References PLearn::RBMConnection::down_size, PLearn::RBMConnection::going_up, PLearn::RBMConnection::input_vec, learn_scale, PLearn::TVec< T >::length(), PLASSERT, PLASSERT_MSG, PLearn::product(), PLearn::productAcc(), rbm_matrix_connection, scale, PLearn::TMat< T >::subMatColumns(), PLearn::TMat< T >::subMatRows(), PLearn::transposeProduct(), PLearn::transposeProductAcc(), PLearn::RBMConnection::up_size, and weights.

{
    PLASSERT( activations.length() == length );
    PLASSERT_MSG( rbm_matrix_connection, "RBMMatrixTransposeConnection must be given an rbm_matrix_connection.\n");

    if( going_up )
    {
        PLASSERT( start+length <= up_size );
        // activations[i-start] += sum_j weights(i,j) input_vec[j]

        if( accumulate )
            transposeProductAcc( activations,
                                 weights.subMatColumns(start,length),
                                 input_vec );
        else
            transposeProduct( activations,
                              weights.subMatColumns(start,length),
                              input_vec );
    }
    else
    {
        PLASSERT( start+length <= down_size );
        // activations[i-start] += sum_j weights(j,i) input_vec[j]
        if( accumulate )
            productAcc( activations,
                        weights.subMatRows(start,length),
                        input_vec );
        else
            product( activations,
                     weights.subMatRows(start,length),
                     input_vec );
    }
    if( learn_scale)
        activations *= scale;
}

Here is the call graph for this function:

void PLearn::RBMMatrixTransposeConnection::computeProducts ( int  start,
int  length,
Mat activations,
bool  accumulate = false 
) const [virtual]

Same as 'computeProduct' but for mini-batches.

Implements PLearn::RBMConnection.

Definition at line 317 of file RBMMatrixTransposeConnection.cc.

References PLearn::RBMConnection::down_size, PLearn::RBMConnection::going_up, PLearn::RBMConnection::inputs_mat, learn_scale, PLearn::TMat< T >::length(), PLASSERT, PLASSERT_MSG, PLearn::product(), PLearn::productAcc(), PLearn::productTranspose(), PLearn::productTransposeAcc(), rbm_matrix_connection, PLearn::TMat< T >::resize(), scale, PLearn::TMat< T >::subMatColumns(), PLearn::TMat< T >::subMatRows(), PLearn::RBMConnection::up_size, and weights.

{
    PLASSERT_MSG( rbm_matrix_connection, "RBMMatrixTransposeConnection must be given an rbm_matrix_connection.\n");
    activations.resize(inputs_mat.length(), length);
    if( going_up )
    {
        PLASSERT( start+length <= up_size );
        // activations(k, i-start) += sum_j weights(i,j) inputs_mat(k, j)

        if( accumulate )
            productAcc(activations,
                    inputs_mat,
                    weights.subMatColumns(start,length));
        else
            product(activations,
                    inputs_mat,
                    weights.subMatColumns(start,length));
    }
    else
    {
        PLASSERT( start+length <= down_size );
        // activations(k, i-start) += sum_j weights(j,i) inputs_mat(k, j)
        if( accumulate )
            productTransposeAcc(activations,
                    inputs_mat,
                    weights.subMatRows(start,length) );
        else
            productTranspose(activations,
                    inputs_mat,
                    weights.subMatRows(start,length) );
    }

    if( learn_scale)
        activations *= scale;
}

Here is the call graph for this function:

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

Declares the class options.

Reimplemented from PLearn::RBMConnection.

Definition at line 64 of file RBMMatrixTransposeConnection.cc.

References PLearn::OptionBase::buildoption, PLearn::declareOption(), PLearn::RBMConnection::declareOptions(), PLearn::RBMConnection::down_size, learn_scale, PLearn::OptionBase::learntoption, rbm_matrix_connection, PLearn::redeclareOption(), scale, and PLearn::RBMConnection::up_size.

{
    declareOption(ol, "rbm_matrix_connection",
                  &RBMMatrixTransposeConnection::rbm_matrix_connection,
                  OptionBase::buildoption,
                  "RBMMatrixConnection from which the weights are taken");

    declareOption(ol, "learn_scale",
                  &RBMMatrixTransposeConnection::learn_scale,
                  OptionBase::buildoption,
                  "Indication that the scale of the weight matrix should be "
                  "learned.\n");

    declareOption(ol, "scale",
                  &RBMMatrixTransposeConnection::scale,
                  OptionBase::learntoption,
                  "Learned scale for weight matrix.\n");

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

    redeclareOption(ol, "up_size", &RBMConnection::up_size,
                    OptionBase::learntoption,
                    "Is set to rbm_matrix_connection->down_size.");
    redeclareOption(ol, "down_size", &RBMConnection::down_size,
                    OptionBase::learntoption,
                    "Is set to rbm_matrix_connection->up_size.");
}

Here is the call graph for this function:

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

Reimplemented from PLearn::RBMConnection.

Definition at line 190 of file RBMMatrixTransposeConnection.h.

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

Reimplemented from PLearn::RBMConnection.

Definition at line 49 of file RBMMatrixTransposeConnection.cc.

void PLearn::RBMMatrixTransposeConnection::forget ( ) [virtual]

reset the parameters to the state they would be BEFORE starting training.

Note that this method is necessarily called from build().

Implements PLearn::OnlineLearningModule.

Definition at line 451 of file RBMMatrixTransposeConnection.cc.

References clearStats(), learn_scale, PLASSERT_MSG, PLWARNING, PLearn::OnlineLearningModule::random_gen, rbm_matrix_connection, and scale.

{
    PLASSERT_MSG( rbm_matrix_connection, "RBMMatrixTransposeConnection must be given an rbm_matrix_connection.\n");
    clearStats();
    if( !random_gen )
    {
        PLWARNING("RBMMatrixTransposeConnection: cannot forget() without"
                  " random_gen");
        return;
    }
    if( !(rbm_matrix_connection->random_gen) )
        rbm_matrix_connection->random_gen = random_gen;
    rbm_matrix_connection->forget();
    if( learn_scale )
        scale = 1;
}

Here is the call graph for this function:

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

Reimplemented from PLearn::Object.

Definition at line 49 of file RBMMatrixTransposeConnection.cc.

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

Reimplemented from PLearn::Object.

Definition at line 49 of file RBMMatrixTransposeConnection.cc.

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

Reimplemented from PLearn::Object.

Definition at line 49 of file RBMMatrixTransposeConnection.cc.

void PLearn::RBMMatrixTransposeConnection::makeDeepCopyFromShallowCopy ( CopiesMap copies) [virtual]
Vec PLearn::RBMMatrixTransposeConnection::makeParametersPointHere ( const Vec global_parameters) [virtual]

Make the parameters data be sub-vectors of the given global_parameters.

The argument should have size >= nParameters. The result is a Vec that starts just after this object's parameters end, i.e. result = global_parameters.subVec(nParameters(),global_parameters.size()-nParameters()); This allows to easily chain calls of this method on multiple RBMParameters.

Implements PLearn::RBMConnection.

Definition at line 490 of file RBMMatrixTransposeConnection.cc.

References PLASSERT_MSG, rbm_matrix_connection, and weights.

{
    PLASSERT_MSG( rbm_matrix_connection, "RBMMatrixTransposeConnection must be given an rbm_matrix_connection.\n");
    Vec ret = rbm_matrix_connection->makeParametersPointHere(global_parameters);
    weights = rbm_matrix_connection->weights;
    return ret;
}
int PLearn::RBMMatrixTransposeConnection::nParameters ( ) const [virtual]

optionally perform some processing after training, or after a series of fprop/bpropUpdate calls to prepare the model for truly out-of-sample operation.

return the number of parameters

THE DEFAULT IMPLEMENTATION PROVIDED IN THE SUPER-CLASS DOES NOT DO ANYTHING. return the number of parameters

Implements PLearn::RBMConnection.

Definition at line 480 of file RBMMatrixTransposeConnection.cc.

References PLearn::TMat< T >::size(), and weights.

{
    return weights.size();
}

Here is the call graph for this function:

void PLearn::RBMMatrixTransposeConnection::update ( ) [virtual]

Updates parameters according to contrastive divergence gradient.

Implements PLearn::RBMConnection.

Definition at line 158 of file RBMMatrixTransposeConnection.cc.

References clearStats(), PLearn::TMat< T >::data(), i, j, learn_scale, PLearn::RBMConnection::learning_rate, PLearn::TMat< T >::length(), PLearn::TMat< T >::mod(), PLearn::RBMConnection::momentum, PLearn::RBMConnection::neg_count, PLERROR, PLearn::RBMConnection::pos_count, PLearn::TMat< T >::resize(), w, weights, weights_inc, weights_neg_stats, weights_pos_stats, and PLearn::TMat< T >::width().

{
    if( learn_scale )
        PLERROR("In RBMMatrixTransposeConnection::update(): not implemented "
                "for learned scale");
    // updates parameters
    //weights -= learning_rate * (weights_pos_stats/pos_count
    //                              - weights_neg_stats/neg_count)
    real pos_factor = -learning_rate / pos_count;
    real neg_factor = learning_rate / neg_count;

    int l = weights.length();
    int w = weights.width();

    real* w_i = weights.data();
    real* wps_i = weights_pos_stats.data();
    real* wns_i = weights_neg_stats.data();
    int w_mod = weights.mod();
    int wps_mod = weights_pos_stats.mod();
    int wns_mod = weights_neg_stats.mod();

    if( momentum == 0. )
    {
        // no need to use weights_inc
        for( int i=0 ; i<l ; i++, w_i+=w_mod, wps_i+=wps_mod, wns_i+=wns_mod )
            for( int j=0 ; j<w ; j++ )
                w_i[j] += pos_factor * wps_i[j] + neg_factor * wns_i[j];
    }
    else
    {
        // ensure that weights_inc has the right size
        weights_inc.resize( l, w );

        // The update rule becomes:
        // weights_inc = momentum * weights_inc
        //               - learning_rate * (weights_pos_stats/pos_count
        //                                  - weights_neg_stats/neg_count);
        // weights += weights_inc;
        real* winc_i = weights_inc.data();
        int winc_mod = weights_inc.mod();
        for( int i=0 ; i<l ; i++, w_i += w_mod, wps_i += wps_mod,
                             wns_i += wns_mod, winc_i += winc_mod )
            for( int j=0 ; j<w ; j++ )
            {
                winc_i[j] = momentum * winc_i[j]
                    + pos_factor * wps_i[j] + neg_factor * wns_i[j];
                w_i[j] += winc_i[j];
            }
    }

    clearStats();
}

Here is the call graph for this function:

void PLearn::RBMMatrixTransposeConnection::update ( const Vec pos_down_values,
const Vec pos_up_values,
const Vec neg_down_values,
const Vec neg_up_values 
) [virtual]

Updates parameters according to contrastive divergence gradient, not using the statistics but the explicit values passed.

Reimplemented from PLearn::RBMConnection.

Definition at line 213 of file RBMMatrixTransposeConnection.cc.

References PLearn::TVec< T >::data(), PLearn::TMat< T >::data(), i, j, learn_scale, PLearn::RBMConnection::learning_rate, PLearn::TVec< T >::length(), PLearn::TMat< T >::length(), PLearn::TMat< T >::mod(), PLearn::RBMConnection::momentum, PLASSERT, PLASSERT_MSG, PLERROR, rbm_matrix_connection, PLearn::TMat< T >::resize(), w, weights, weights_inc, and PLearn::TMat< T >::width().

{
    if( learn_scale )
        PLERROR("In RBMMatrixTransposeConnection::update(): not implemented "
                "for learned scale");

    PLASSERT_MSG( rbm_matrix_connection, "RBMMatrixTransposeConnection must be given an rbm_matrix_connection.\n");
    // 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]);

    int l = weights.length();
    int w = weights.width();
    PLASSERT( pos_up_values.length() == l );
    PLASSERT( neg_up_values.length() == l );
    PLASSERT( pos_down_values.length() == w );
    PLASSERT( neg_down_values.length() == w );

    real* w_i = weights.data();
    real* puv_i = pos_up_values.data();
    real* nuv_i = neg_up_values.data();
    real* pdv = pos_down_values.data();
    real* ndv = neg_down_values.data();
    int w_mod = weights.mod();

    if( momentum == 0. )
    {
        for( int i=0 ; i<l ; i++, w_i += w_mod, puv_i++, nuv_i++ )
            for( int j=0 ; j<w ; j++ )
                w_i[j] += learning_rate * (*nuv_i * ndv[j] - *puv_i * pdv[j]);
    }
    else
    {
        // ensure that weights_inc has the right size
        weights_inc.resize( l, w );

        // The update rule becomes:
        // weights_inc = momentum * weights_inc
        //               - learning_rate * ( h_0 v_0' - h_1 v_1' );
        // weights += weights_inc;

        real* winc_i = weights_inc.data();
        int winc_mod = weights_inc.mod();
        for( int i=0 ; i<l ; i++, w_i += w_mod, winc_i += winc_mod,
                             puv_i++, nuv_i++ )
            for( int j=0 ; j<w ; j++ )
            {
                winc_i[j] = momentum * winc_i[j]
                    + learning_rate * (*nuv_i * ndv[j] - *puv_i * pdv[j]);
                w_i[j] += winc_i[j];
            }
    }
}

Here is the call graph for this function:


Member Data Documentation

Reimplemented from PLearn::RBMConnection.

Definition at line 190 of file RBMMatrixTransposeConnection.h.

Indication that the scale of the weight matrix should be learned.

Definition at line 69 of file RBMMatrixTransposeConnection.h.

Referenced by bpropUpdate(), computeProduct(), computeProducts(), declareOptions(), forget(), and update().

Learned scale for weight matrix.

Definition at line 74 of file RBMMatrixTransposeConnection.h.

Referenced by bpropUpdate(), computeProduct(), computeProducts(), declareOptions(), and forget().

Matrix containing unit-to-unit weights ( $output_size \times input_size$)

Definition at line 63 of file RBMMatrixTransposeConnection.h.

Referenced by bpropUpdate(), build_(), computeProduct(), computeProducts(), makeDeepCopyFromShallowCopy(), makeParametersPointHere(), nParameters(), and update().

Used if momentum != 0.

Definition at line 83 of file RBMMatrixTransposeConnection.h.

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

Accumulates negative contribution to the weights' gradient.

Definition at line 80 of file RBMMatrixTransposeConnection.h.

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

Accumulates positive contribution to the weights' gradient.

Definition at line 77 of file RBMMatrixTransposeConnection.h.

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


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