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::LinearCombinationModule Class Reference

This module outputs a linear combination of input ports. More...

#include <LinearCombinationModule.h>

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

List of all members.

Public Member Functions

 LinearCombinationModule ()
 Default constructor.
void fprop (const TVec< Mat * > &ports_value)
 Perform a fprop step.
virtual void bpropAccUpdate (const TVec< Mat * > &ports_value, const TVec< Mat * > &ports_gradient)
 Perform a back propagation step (also updating parameters according to the provided gradient).
virtual void forget ()
 Reset the parameters to the state they would be BEFORE starting training.
virtual const TVec< string > & getPorts ()
 Return the list of ports in the module.
virtual string classname () const
virtual OptionListgetOptionList () const
virtual OptionMapgetOptionMap () const
virtual RemoteMethodMapgetRemoteMethodMap () const
virtual LinearCombinationModuledeepCopy (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

Vec weights
 ### declare public option fields (such as build options) here
bool adaptive
 whether to adapt the weights, and whether to clear them to 0 upon forget()
real learning_rate
 learning rate if adapting the weights
TVec< string > port_names
 all the input ports, followed by the output port

Static Public Attributes

static StaticInitializer _static_initializer_

Static Protected Member Functions

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

Private Types

typedef OnlineLearningModule inherited

Private Member Functions

void build_ ()
 This does the actual building.

Detailed Description

This module outputs a linear combination of input ports.

All the ports should have the same dimensions. The weights of the linear combination could either be learned or user-defined.

Definition at line 52 of file LinearCombinationModule.h.


Member Typedef Documentation

Reimplemented from PLearn::OnlineLearningModule.

Definition at line 54 of file LinearCombinationModule.h.


Constructor & Destructor Documentation

PLearn::LinearCombinationModule::LinearCombinationModule ( )

Default constructor.

Definition at line 57 of file LinearCombinationModule.cc.

    : adaptive(false), learning_rate(0)
{
}

Member Function Documentation

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

Reimplemented from PLearn::OnlineLearningModule.

Definition at line 55 of file LinearCombinationModule.cc.

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

Reimplemented from PLearn::OnlineLearningModule.

Definition at line 55 of file LinearCombinationModule.cc.

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

Reimplemented from PLearn::OnlineLearningModule.

Definition at line 55 of file LinearCombinationModule.cc.

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

Reimplemented from PLearn::OnlineLearningModule.

Definition at line 55 of file LinearCombinationModule.cc.

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

Reimplemented from PLearn::Object.

Definition at line 55 of file LinearCombinationModule.cc.

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

Reimplemented from PLearn::OnlineLearningModule.

Definition at line 55 of file LinearCombinationModule.cc.

void PLearn::LinearCombinationModule::bpropAccUpdate ( const TVec< Mat * > &  ports_value,
const TVec< Mat * > &  ports_gradient 
) [virtual]

Perform a back propagation step (also updating parameters according to the provided gradient).

The matrices in 'ports_value' must be the same as the ones given in a previous call to 'fprop' (and thus they should in particular contain the result of the fprop computation). However, they are not necessarily the same as the ones given in the LAST call to 'fprop': if there is a need to store an internal module state, this should be done using a specific port to store this state. Each Mat* pointer in the 'ports_gradient' vector can be one of:

  • a full matrix : this is the gradient that is provided to the module, and can be used to compute other ports' gradient.
  • an empty matrix: this is a gradient we want to compute and accumulate into. This matrix must have length 0 and a width equal to the width of the corresponding matrix in the 'ports_value' vector (we can thus accumulate gradients using PLearn's ability to keep intact stored values when resizing a matrix' length).
  • a NULL pointer : this is a gradient that is not available, but does not need to be returned (or even computed). The default version tries to use the standard mini-batch bpropUpdate method, when possible.

Reimplemented from PLearn::OnlineLearningModule.

Definition at line 167 of file LinearCombinationModule.cc.

References adaptive, PLearn::OnlineLearningModule::checkProp(), PLearn::dot(), i, PLearn::TMat< T >::isEmpty(), learning_rate, PLearn::TMat< T >::length(), PLearn::TVec< T >::length(), PLearn::multiplyAcc(), PLASSERT, weights, and PLearn::TMat< T >::width().

{
    int n_ports = weights.length() + 1;
    PLASSERT( ports_value.length() == n_ports && ports_gradient.length() == n_ports);

    const TVec<Mat*>& input_grad = ports_gradient;
    Mat* output_grad = ports_gradient[n_ports-1];
    if (output_grad && !output_grad->isEmpty())
    {
        int mbs = output_grad->length();
        int width = output_grad->width();
        for (int i=0;i<n_ports-1;i++)
        {
            if (input_grad[i])
            {
                PLASSERT(input_grad[i]->isEmpty() &&
                         input_grad[i]->width() == width);
                input_grad[i]->resize(mbs,width);
                multiplyAcc(*input_grad[i],*output_grad,weights[i]);
            }
            if (adaptive && learning_rate > 0)
            {
                Mat* input_i = ports_value[i];
                PLASSERT(input_i);
                weights[i] -= learning_rate * dot(*output_grad,*input_i);
            }
        }
    }

    // Ensure all required gradients have been computed.
    checkProp(ports_gradient);
}

Here is the call graph for this function:

void PLearn::LinearCombinationModule::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::OnlineLearningModule.

Definition at line 113 of file LinearCombinationModule.cc.

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

Here is the call graph for this function:

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

This does the actual building.

Reimplemented from PLearn::OnlineLearningModule.

Definition at line 83 of file LinearCombinationModule.cc.

References adaptive, PLearn::TVec< T >::fill(), i, learning_rate, PLearn::TVec< T >::length(), PLASSERT, PLCHECK, PLWARNING, port_names, PLearn::TVec< T >::resize(), PLearn::tostring(), and weights.

Referenced by build().

{
    PLASSERT(weights.length()==0 || port_names.length()==0 ||
             weights.length() + 1 ==port_names.length());
    int n_ports=0;
    if (weights.length()!=0 && port_names.length()==0)
        // weights provided but not ports: give default port names
    {
        n_ports = weights.length() + 1;
        port_names.resize(n_ports);
        for (int i=0;i<n_ports-1;i++)
            port_names[i]="in_" + tostring(i+1);
        port_names[n_ports-1]="output";
    }
    if (weights.length()==0 && port_names.length()!=0)
        // ports provided but not weights: initialize weights to 0
    {
        n_ports = port_names.length();
        weights.resize(n_ports - 1);
        weights.fill(0);
        if (!adaptive)
            PLWARNING("LinearCombinationModule::build: non-adaptive weights set to 0! the module will always output 0.");
    }

    PLCHECK( learning_rate >= 0 );
}

Here is the call graph for this function:

Here is the caller graph for this function:

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

Reimplemented from PLearn::Object.

Definition at line 55 of file LinearCombinationModule.cc.

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

Declares the class options.

Reimplemented from PLearn::OnlineLearningModule.

Definition at line 62 of file LinearCombinationModule.cc.

References adaptive, PLearn::OptionBase::buildoption, PLearn::declareOption(), PLearn::OnlineLearningModule::declareOptions(), learning_rate, and weights.

{
    declareOption(ol, "weights", &LinearCombinationModule::weights,
                  OptionBase::buildoption,
                  "the weights of the linear combination: a vector with one element per input port\n");

    declareOption(ol, "adaptive", &LinearCombinationModule::adaptive,
                  OptionBase::buildoption,
                  "whether to adapt the weights, if true they are cleared upon initialization (forget()).\n");

    declareOption(ol, "learning_rate", &LinearCombinationModule::learning_rate,
                  OptionBase::buildoption,
                  "Learning rate to adapt the weights by online gradient descent.\n");

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

Here is the call graph for this function:

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

Reimplemented from PLearn::OnlineLearningModule.

Definition at line 251 of file LinearCombinationModule.h.

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

Reimplemented from PLearn::OnlineLearningModule.

Definition at line 55 of file LinearCombinationModule.cc.

void PLearn::LinearCombinationModule::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 204 of file LinearCombinationModule.cc.

References adaptive, PLearn::TVec< T >::clear(), and weights.

{
    if (adaptive)
        weights.clear();
}

Here is the call graph for this function:

void PLearn::LinearCombinationModule::fprop ( const TVec< Mat * > &  ports_value) [virtual]

Perform a fprop step.

Each Mat* pointer in the 'ports_value' vector can be one of:

  • a full matrix: this is data that is provided to the module, and can be used to compute other ports' values
  • an empty matrix: this is what we want to compute
  • a NULL pointer: this is data that is not available, but whose value does not need to be returned (or even computed) The default version will either:
  • call the mini-batch versions of standard fprop if 'ports_value' has size 2, with the first value being provided (and the second being the desired output)
  • crash otherwise

Reimplemented from PLearn::OnlineLearningModule.

Definition at line 134 of file LinearCombinationModule.cc.

References PLearn::OnlineLearningModule::checkProp(), PLearn::TMat< T >::clear(), i, PLearn::TMat< T >::isEmpty(), PLearn::TVec< T >::length(), PLearn::multiplyAcc(), PLASSERT, PLERROR, PLearn::TMat< T >::resize(), and weights.

{
    int n_ports = weights.length() + 1;
    if ( n_ports < 2 )
        // has build completed? there should be at least one input port + the output port
        PLERROR("LinearCombinationModule should have at least 2 ports (one input port and one output port)\n");
    PLASSERT( ports_value.length() == n_ports ); // is the input coherent with expected nPorts

    const TVec<Mat*>& inputs = ports_value;
    Mat* output = ports_value[n_ports-1];
    if (output) {
        PLASSERT( output->isEmpty() );
        PLASSERT( inputs[0] );
        int mbs = inputs[0]->length();
        int width = inputs[0]->width();
        output->resize(mbs, width);
        output->clear();
        for (int i=0;i<n_ports-1;i++) {
            Mat* input_i = inputs[i];
            if (!input_i || input_i->isEmpty())
                PLERROR("In LinearCombinationModule::fprop - The %d-th input "
                        "port is missing or empty", i);
            multiplyAcc(*output, *input_i, weights[i]);
        }
    }

    // Ensure all required ports have been computed.
    checkProp(ports_value);
}

Here is the call graph for this function:

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

Reimplemented from PLearn::Object.

Definition at line 55 of file LinearCombinationModule.cc.

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

Reimplemented from PLearn::Object.

Definition at line 55 of file LinearCombinationModule.cc.

const TVec< string > & PLearn::LinearCombinationModule::getPorts ( ) [virtual]

Return the list of ports in the module.

Reimplemented from PLearn::OnlineLearningModule.

Definition at line 242 of file LinearCombinationModule.cc.

References port_names.

                                                      {
    return port_names;
}
RemoteMethodMap & PLearn::LinearCombinationModule::getRemoteMethodMap ( ) const [virtual]

Reimplemented from PLearn::Object.

Definition at line 55 of file LinearCombinationModule.cc.

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

Transforms a shallow copy into a deep copy.

Reimplemented from PLearn::OnlineLearningModule.

Definition at line 123 of file LinearCombinationModule.cc.

References PLearn::deepCopyField(), PLearn::OnlineLearningModule::makeDeepCopyFromShallowCopy(), port_names, and weights.

Here is the call graph for this function:


Member Data Documentation

Reimplemented from PLearn::OnlineLearningModule.

Definition at line 251 of file LinearCombinationModule.h.

whether to adapt the weights, and whether to clear them to 0 upon forget()

Definition at line 64 of file LinearCombinationModule.h.

Referenced by bpropAccUpdate(), build_(), declareOptions(), and forget().

learning rate if adapting the weights

Definition at line 66 of file LinearCombinationModule.h.

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

all the input ports, followed by the output port

Definition at line 69 of file LinearCombinationModule.h.

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

### declare public option fields (such as build options) here

the weights of the linear combination: one per input port

Definition at line 62 of file LinearCombinationModule.h.

Referenced by bpropAccUpdate(), build_(), declareOptions(), forget(), fprop(), and makeDeepCopyFromShallowCopy().


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