PLearn 0.1
|
Implements the word representation layer for the online NNLM. More...
#include <NnlmWordRepresentationLayer.h>
Public Member Functions | |
NnlmWordRepresentationLayer () | |
Default constructor. | |
virtual void | fprop (const Vec &input, Vec &output) const |
given the input, compute the output (possibly resize it appropriately) | |
virtual void | bpropUpdate (const Vec &input, const Vec &output, const Vec &output_gradient) |
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 | forget () |
this version allows to obtain the input gradient as well N.B. | |
virtual string | classname () const |
virtual OptionList & | getOptionList () const |
virtual OptionMap & | getOptionMap () const |
virtual RemoteMethodMap & | getRemoteMethodMap () const |
virtual NnlmWordRepresentationLayer * | 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_ () |
optionally perform some processing after training, or after a series of fprop/bpropUpdate calls to prepare the model for truly out-of-sample operation. | |
static OptionList & | _getOptionList_ () |
static RemoteMethodMap & | _getRemoteMethodMap_ () |
static Object * | _new_instance_for_typemap_ () |
static bool | _isa_ (const Object *o) |
static void | _static_initialize_ () |
static const PPath & | declaringFile () |
Public Attributes | |
int | vocabulary_size |
int | word_representation_size |
int | context_size |
real | start_learning_rate |
learning_rate = start_learning_rate / (1 + decrease_constant*t), where t is the number of updates since the beginning | |
real | decrease_constant |
Static Public Attributes | |
static StaticInitializer | _static_initializer_ |
Protected Member Functions | |
virtual void | resetWeights () |
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. | |
Private Attributes | |
int | step_number |
real | learning_rate |
Mat | weights |
Implements the word representation layer for the online NNLM.
Best explained as a onehot GradNNetLayer repeated 'context size' (ie input_size) times.
Some variables explained: input_size -> 'context size' word_representation_size -> size of the real distributed word representation output_size -> word representation size * input_size virtual_input_size -> input_size * vocabulary_size
Definition at line 66 of file NnlmWordRepresentationLayer.h.
typedef OnlineLearningModule PLearn::NnlmWordRepresentationLayer::inherited [private] |
Reimplemented from PLearn::OnlineLearningModule.
Definition at line 68 of file NnlmWordRepresentationLayer.h.
PLearn::NnlmWordRepresentationLayer::NnlmWordRepresentationLayer | ( | ) |
Default constructor.
Definition at line 51 of file NnlmWordRepresentationLayer.cc.
: OnlineLearningModule(), vocabulary_size( -1 ), word_representation_size( -1 ), context_size( -1 ), start_learning_rate( 0.001 ), decrease_constant( 0 ), step_number( 0 ), learning_rate( 0.0 ) { // ### You may (or not) want to call build_() to finish building the object // ### (doing so assumes the parent classes' build_() have been called too // ### in the parent classes' constructors, something that you must ensure) }
string PLearn::NnlmWordRepresentationLayer::_classname_ | ( | ) | [static] |
optionally perform some processing after training, or after a series of fprop/bpropUpdate calls to prepare the model for truly out-of-sample operation.
THE DEFAULT IMPLEMENTATION PROVIDED IN THE SUPER-CLASS DOES NOT DO ANYTHING. in case bpropUpdate does not do anything, make it known THE DEFAULT IMPLEMENTATION PROVIDED IN THE SUPER-CLASS RETURNS false;
Reimplemented from PLearn::OnlineLearningModule.
Definition at line 49 of file NnlmWordRepresentationLayer.cc.
OptionList & PLearn::NnlmWordRepresentationLayer::_getOptionList_ | ( | ) | [static] |
Reimplemented from PLearn::OnlineLearningModule.
Definition at line 49 of file NnlmWordRepresentationLayer.cc.
RemoteMethodMap & PLearn::NnlmWordRepresentationLayer::_getRemoteMethodMap_ | ( | ) | [static] |
Reimplemented from PLearn::OnlineLearningModule.
Definition at line 49 of file NnlmWordRepresentationLayer.cc.
Reimplemented from PLearn::OnlineLearningModule.
Definition at line 49 of file NnlmWordRepresentationLayer.cc.
Object * PLearn::NnlmWordRepresentationLayer::_new_instance_for_typemap_ | ( | ) | [static] |
Reimplemented from PLearn::Object.
Definition at line 49 of file NnlmWordRepresentationLayer.cc.
StaticInitializer NnlmWordRepresentationLayer::_static_initializer_ & PLearn::NnlmWordRepresentationLayer::_static_initialize_ | ( | ) | [static] |
Reimplemented from PLearn::OnlineLearningModule.
Definition at line 49 of file NnlmWordRepresentationLayer.cc.
void PLearn::NnlmWordRepresentationLayer::bpropUpdate | ( | const Vec & | input, |
const Vec & | output, | ||
const Vec & | output_gradient | ||
) | [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).
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.
Reimplemented from PLearn::OnlineLearningModule.
Definition at line 213 of file NnlmWordRepresentationLayer.cc.
References decrease_constant, i, PLearn::OnlineLearningModule::input_size, j, learning_rate, PLearn::OnlineLearningModule::output_size, PLERROR, PLearn::TVec< T >::size(), start_learning_rate, step_number, vocabulary_size, weights, and word_representation_size.
{ int in_size = input.size(); int out_size = output.size(); int og_size = output_gradient.size(); // size check if( in_size != input_size ) { PLERROR("NnlmWordRepresentationLayer::bpropUpdate: 'input.size()' should be equal\n" " to 'input_size' (%i != %i)\n", in_size, input_size); } if( out_size != output_size ) { PLERROR("NnlmWordRepresentationLayer::bpropUpdate: 'output.size()' should be" " equal\n" " to 'output_size' (%i != %i)\n", out_size, output_size); } if( og_size != output_size ) { PLERROR("NnlmWordRepresentationLayer::bpropUpdate: 'output_gradient.size()'" " should\n" " be equal to 'output_size' (%i != %i)\n", og_size, output_size); } learning_rate = start_learning_rate / ( 1.0 + decrease_constant * step_number); //cout << "NnlmWordRepresentationLayer::bpropUpdate -> output_gradient is " << output_gradient << endl; // magnitude of index check and update for( int i=0; i<input_size; i++) { if( input[i] >= vocabulary_size || input[i] < 0 ) { PLERROR("NnlmWordRepresentationLayer::bpropUpdate: 'input[%i]' should be smaller\n" " than 'vocabulary_size' (%i !< %i)\n", i, input[i], vocabulary_size); } // MISTAKE??????? // MISTAKE??????? /*for(int j=0; j < output_size; j++) { weights( (int) input[i], j%word_representation_size) -= learning_rate * output_gradient[j]; }*/ //cout << "word rep gradient "; for(int j=0; j < word_representation_size; j++) { weights( (int) input[i], j) -= learning_rate * output_gradient[j+i*word_representation_size]; //cout << - learning_rate * output_gradient[j+i*word_representation_size] << " "; } //cout << endl; } step_number++; }
void PLearn::NnlmWordRepresentationLayer::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 146 of file NnlmWordRepresentationLayer.cc.
References PLearn::OnlineLearningModule::build(), and build_().
{ inherited::build(); build_(); }
void PLearn::NnlmWordRepresentationLayer::build_ | ( | ) | [private] |
This does the actual building.
Reimplemented from PLearn::OnlineLearningModule.
Definition at line 110 of file NnlmWordRepresentationLayer.cc.
References context_size, forget(), PLearn::OnlineLearningModule::input_size, PLearn::OnlineLearningModule::output_size, PLERROR, PLearn::TMat< T >::size(), vocabulary_size, weights, and word_representation_size.
Referenced by build().
{ // *** Some variables are connected... // for now we overwrite these input_size = context_size; output_size = context_size * word_representation_size; // *** A few sanity checks if( input_size <= 0 ) { PLERROR("NnlmWordRepresentationLayer::build_: 'input_size' <= 0 (%i).\n" "You should set it to a positive integer.\n", input_size); } else if( word_representation_size * context_size != output_size ) { PLERROR("NnlmWordRepresentationLayer::build_: 'output_size' inconsistent with\n" " 'word_representation_size * input_size': %i != ( %i * %i)\n" , output_size, word_representation_size, input_size); } else if( vocabulary_size <= 0 ) { PLERROR("NnlmWordRepresentationLayer::build_: 'vocabulary_size' <= 0(%i).\n" , vocabulary_size); } // *** Initialize weights if not loaded if( weights.size() == 0 ) { forget(); } }
string PLearn::NnlmWordRepresentationLayer::classname | ( | ) | const [virtual] |
Reimplemented from PLearn::Object.
Definition at line 49 of file NnlmWordRepresentationLayer.cc.
void PLearn::NnlmWordRepresentationLayer::declareOptions | ( | OptionList & | ol | ) | [static, protected] |
Declares the class options.
Reimplemented from PLearn::OnlineLearningModule.
Definition at line 66 of file NnlmWordRepresentationLayer.cc.
References PLearn::OptionBase::buildoption, context_size, PLearn::declareOption(), PLearn::OnlineLearningModule::declareOptions(), decrease_constant, PLearn::OptionBase::learntoption, start_learning_rate, step_number, vocabulary_size, weights, and word_representation_size.
{ declareOption(ol, "vocabulary_size", &NnlmWordRepresentationLayer::vocabulary_size, OptionBase::buildoption, "size of vocabulary used - defines the virtual input size"); declareOption(ol, "word_representation_size", &NnlmWordRepresentationLayer::word_representation_size, OptionBase::buildoption, "size of the real distributed word representation"); declareOption(ol, "context_size", &NnlmWordRepresentationLayer::context_size, OptionBase::buildoption, "size of word context"); declareOption(ol, "start_learning_rate", &NnlmWordRepresentationLayer::start_learning_rate, OptionBase::buildoption, "Learning-rate of stochastic gradient optimization"); declareOption(ol, "decrease_constant", &NnlmWordRepresentationLayer::decrease_constant, OptionBase::buildoption, "Decrease constant of stochastic gradient optimization"); // * Learnt declareOption(ol, "step_number", &NnlmWordRepresentationLayer::step_number, OptionBase::learntoption, "The step number, incremented after each update."); declareOption(ol, "weights", &NnlmWordRepresentationLayer::weights, OptionBase::learntoption, "Input weights of the neurons (one row per neuron, no bias)."); // Now call the parent class' declareOptions inherited::declareOptions(ol); }
static const PPath& PLearn::NnlmWordRepresentationLayer::declaringFile | ( | ) | [inline, static] |
Reimplemented from PLearn::OnlineLearningModule.
Definition at line 157 of file NnlmWordRepresentationLayer.h.
:
//##### Protected Options ###############################################
NnlmWordRepresentationLayer * PLearn::NnlmWordRepresentationLayer::deepCopy | ( | CopiesMap & | copies | ) | const [virtual] |
Reimplemented from PLearn::OnlineLearningModule.
Definition at line 49 of file NnlmWordRepresentationLayer.cc.
void PLearn::NnlmWordRepresentationLayer::forget | ( | ) | [virtual] |
this version allows to obtain the input gradient as well N.B.
reset the parameters to the state they would be BEFORE starting training.
THE DEFAULT IMPLEMENTATION IN SUPER-CLASS JUST RAISES A PLERROR. Similar to bpropUpdate, but adapt based also on the estimation of the diagonal of the Hessian matrix, and propagates this back. If these methods are defined, you can use them INSTEAD of bpropUpdate(...) N.B. A DEFAULT IMPLEMENTATION IS PROVIDED IN THE SUPER-CLASS, WHICH JUST CALLS bbpropUpdate(input, output, input_gradient, output_gradient, out_hess, in_hess) AND IGNORES INPUT HESSIAN AND INPUT GRADIENT. this version allows to obtain the input gradient and diag_hessian N.B. A DEFAULT IMPLEMENTATION IS PROVIDED IN THE SUPER-CLASS, WHICH RAISES A PLERROR. reset the parameters to the state they would be BEFORE starting training. Note that this method is necessarily called from build().
Note that this method is necessarily called from build().
Implements PLearn::OnlineLearningModule.
Definition at line 286 of file NnlmWordRepresentationLayer.cc.
References PLearn::OnlineLearningModule::random_gen, resetWeights(), step_number, and weights.
Referenced by build_().
{ // *** Weights resetWeights(); // TODO add an option for the seed if( !random_gen ) { random_gen = new PRandom( 1 ); } //real r = 1.0 / sqrt(input_size); //random_gen->fill_random_uniform(weights,-r,r); random_gen->fill_random_uniform(weights,-1.0,1.0); // *** step_number = 0; }
given the input, compute the output (possibly resize it appropriately)
given the input, compute the output (possibly resize it appropriately) In our case, we just do a lookup in the weights matrix, for each word in the context.
Reimplemented from PLearn::OnlineLearningModule.
Definition at line 166 of file NnlmWordRepresentationLayer.cc.
References i, PLearn::OnlineLearningModule::input_size, PLearn::OnlineLearningModule::output_size, PLERROR, PLearn::TVec< T >::size(), PLearn::TVec< T >::subVec(), vocabulary_size, weights, and word_representation_size.
{ // TODO only do these in debug // *** Sanity checks // check the input holds input_size hot indexes int in_size = input.size(); if( in_size != input_size ) { PLERROR("NnlmWordRepresentationLayer::fprop: 'input.size()' should be equal\n" " to 'input_size' (%i != %i)\n", in_size, input_size); } // int out_size = output.size(); if( out_size != output_size ) { PLERROR("NnlmWordRepresentationLayer::fprop: 'output.size()' should be equal\n" " to 'output_size' (%i != %i)\n", out_size, output_size); } // magnitude of index check for( int i=0; i<input_size; i++) { if( input[i] >= vocabulary_size || input[i] < 0 ) { PLERROR("NnlmWordRepresentationLayer::fprop: 'input[%i]' should be smaller\n" " than 'vocabulary_size' (%i !< %i)\n", i, input[i], vocabulary_size); } output.subVec( i*word_representation_size, word_representation_size ) << weights( (int) input[i] ); } }
OptionList & PLearn::NnlmWordRepresentationLayer::getOptionList | ( | ) | const [virtual] |
Reimplemented from PLearn::Object.
Definition at line 49 of file NnlmWordRepresentationLayer.cc.
OptionMap & PLearn::NnlmWordRepresentationLayer::getOptionMap | ( | ) | const [virtual] |
Reimplemented from PLearn::Object.
Definition at line 49 of file NnlmWordRepresentationLayer.cc.
RemoteMethodMap & PLearn::NnlmWordRepresentationLayer::getRemoteMethodMap | ( | ) | const [virtual] |
Reimplemented from PLearn::Object.
Definition at line 49 of file NnlmWordRepresentationLayer.cc.
void PLearn::NnlmWordRepresentationLayer::makeDeepCopyFromShallowCopy | ( | CopiesMap & | copies | ) | [virtual] |
Transforms a shallow copy into a deep copy.
Reimplemented from PLearn::OnlineLearningModule.
Definition at line 153 of file NnlmWordRepresentationLayer.cc.
References PLearn::deepCopyField(), PLearn::OnlineLearningModule::makeDeepCopyFromShallowCopy(), and weights.
{ inherited::makeDeepCopyFromShallowCopy(copies); deepCopyField(weights, copies); // ### Remove this line when you have fully implemented this method. //PLERROR("NnlmWordRepresentationLayer::makeDeepCopyFromShallowCopy not fully (correctly) implemented yet!"); }
void PLearn::NnlmWordRepresentationLayer::resetWeights | ( | ) | [protected, virtual] |
Definition at line 359 of file NnlmWordRepresentationLayer.cc.
References PLearn::TMat< T >::fill(), PLearn::TMat< T >::resize(), vocabulary_size, weights, and word_representation_size.
Referenced by forget().
{ weights.resize( vocabulary_size, word_representation_size ); weights.fill( 0 ); }
Reimplemented from PLearn::OnlineLearningModule.
Definition at line 157 of file NnlmWordRepresentationLayer.h.
Definition at line 76 of file NnlmWordRepresentationLayer.h.
Referenced by build_(), and declareOptions().
Definition at line 81 of file NnlmWordRepresentationLayer.h.
Referenced by bpropUpdate(), and declareOptions().
Definition at line 185 of file NnlmWordRepresentationLayer.h.
Referenced by bpropUpdate().
learning_rate = start_learning_rate / (1 + decrease_constant*t), where t is the number of updates since the beginning
Definition at line 80 of file NnlmWordRepresentationLayer.h.
Referenced by bpropUpdate(), and declareOptions().
Definition at line 184 of file NnlmWordRepresentationLayer.h.
Referenced by bpropUpdate(), declareOptions(), and forget().
Definition at line 74 of file NnlmWordRepresentationLayer.h.
Referenced by bpropUpdate(), build_(), declareOptions(), fprop(), and resetWeights().
Definition at line 187 of file NnlmWordRepresentationLayer.h.
Referenced by bpropUpdate(), build_(), declareOptions(), forget(), fprop(), makeDeepCopyFromShallowCopy(), and resetWeights().
Definition at line 75 of file NnlmWordRepresentationLayer.h.
Referenced by bpropUpdate(), build_(), declareOptions(), fprop(), and resetWeights().