PLearn 0.1
|
00001 // -*- C++ -*- 00002 00003 // GaussPartSupervisedDBN.h 00004 // 00005 // Copyright (C) 2006 Pascal Lamblin 00006 // 00007 // Redistribution and use in source and binary forms, with or without 00008 // modification, are permitted provided that the following conditions are met: 00009 // 00010 // 1. Redistributions of source code must retain the above copyright 00011 // notice, this list of conditions and the following disclaimer. 00012 // 00013 // 2. Redistributions in binary form must reproduce the above copyright 00014 // notice, this list of conditions and the following disclaimer in the 00015 // documentation and/or other materials provided with the distribution. 00016 // 00017 // 3. The name of the authors may not be used to endorse or promote 00018 // products derived from this software without specific prior written 00019 // permission. 00020 // 00021 // THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR 00022 // IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 00023 // OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN 00024 // NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 00025 // SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED 00026 // TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 00027 // PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 00028 // LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 00029 // NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 00030 // SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 00031 // 00032 // This file is part of the PLearn library. For more information on the PLearn 00033 // library, go to the PLearn Web site at www.plearn.org 00034 00035 // Authors: Pascal Lamblin 00036 00040 #ifndef GaussPartSupervisedDBN_INC 00041 #define GaussPartSupervisedDBN_INC 00042 00043 #include <plearn_learners/distributions/PDistribution.h> 00044 00045 namespace PLearn { 00046 00047 class RBMLayer; 00048 class RBMMixedLayer; 00049 class RBMMultinomialLayer; 00050 class RBMParameters; 00051 class RBMLLParameters; 00052 class RBMQLParameters; 00053 class RBMJointLLParameters; 00054 class OnlineLearningModule; 00055 00063 class GaussPartSupervisedDBN : public PDistribution 00064 { 00065 typedef PDistribution inherited; 00066 00067 public: 00068 //##### Public Build Options ############################################ 00069 00071 real learning_rate; 00072 00074 Vec supervised_learning_rates; 00075 00077 real fine_tuning_learning_rate; 00078 00080 real initial_momentum; 00081 00083 real final_momentum; 00084 00087 int momentum_switch_time; 00088 00090 real weight_decay; 00091 00097 string initialization_method; 00098 00101 int n_layers; 00102 00105 TVec< PP<RBMLayer> > layers; 00106 00108 PP<RBMLayer> last_layer; 00109 00111 PP<RBMMultinomialLayer> target_layer; 00112 00114 PP<RBMMixedLayer> joint_layer; 00115 00118 TVec< PP<RBMLLParameters> > params; 00119 00122 PP<RBMQLParameters> input_params; 00123 00125 PP<RBMLLParameters> target_params; 00126 00129 PP<RBMJointLLParameters> joint_params; 00130 00133 TVec< PP<OnlineLearningModule> > regressors; 00134 00139 int parallelization_minibatch_size; 00140 00143 bool sum_parallel_contributions; 00144 00147 TVec<int> training_schedule; 00148 00155 string fine_tuning_method; 00156 00157 // bool use_sample_rather_than_expectation_in_positive_phase_statistics; 00158 00170 TVec<int> use_sample_or_expectation; 00171 00172 public: 00173 //##### Public Member Functions ######################################### 00174 00176 // ### Make sure the implementation in the .cc 00177 // ### initializes all fields to reasonable default values. 00178 GaussPartSupervisedDBN(); 00179 00180 00181 //##### PDistribution Member Functions ################################## 00182 00184 virtual real density(const Vec& y) const; 00185 00187 virtual real log_density(const Vec& y) const; 00188 00190 virtual real survival_fn(const Vec& y) const; 00191 00193 virtual real cdf(const Vec& y) const; 00194 00196 virtual void expectation(Vec& mu) const; 00197 00199 virtual void variance(Mat& cov) const; 00200 00203 virtual void generate(Vec& y) const; 00204 00205 //### Override this method if you need it (and if your distribution can 00206 //### handle it. Default version calls PLERROR. 00211 // virtual void generatePredictorGivenPredicted(Vec& x, const Vec& y); 00212 00214 //### See help in PDistribution.h. 00215 virtual bool setPredictorPredictedSizes(int the_predictor_size, 00216 int the_predicted_size, 00217 bool call_parent = true); 00218 00220 //### See help in PDistribution.h. 00221 virtual void setPredictor(const Vec& predictor, bool call_parent = true) 00222 const; 00223 00224 // ### These methods may be overridden for efficiency purpose: 00225 /* 00226 //### Default version calls setPredictorPredictedSises(0,-1) and generate 00231 virtual void generateJoint(Vec& xy); 00232 00233 //### Default version calls generateJoint and discards y 00238 virtual void generatePredictor(Vec& x); 00239 00240 //### Default version calls generateJoint and discards x 00245 virtual void generatePredicted(Vec& y); 00246 */ 00247 00248 00249 //##### PLearner Member Functions ####################################### 00250 00251 // ### Default version of inputsize returns learner->inputsize() 00252 // ### If this is not appropriate, you should uncomment this and define 00253 // ### it properly in the .cc 00254 // virtual int inputsize() const; 00255 00263 virtual void forget(); 00264 00268 virtual void train(); 00269 00273 virtual void computeCostsFromOutputs(const Vec& input, const Vec& output, 00274 const Vec& target, Vec& costs) const; 00275 00276 virtual TVec<string> getTestCostNames() const; 00277 virtual TVec<string> getTrainCostNames() const; 00278 00280 #if USING_MPI 00281 00282 00283 00284 00285 00286 00287 00288 virtual void test(VMat testset, PP<VecStatsCollector> test_stats, 00289 VMat testoutputs=0, VMat testcosts=0) const; 00290 #endif 00291 00292 00293 //##### PLearn::Object Protocol ######################################### 00294 00295 // Declares other standard object methods. 00296 // ### If your class is not instantiatable (it has pure virtual methods) 00297 // ### you should replace this by PLEARN_DECLARE_ABSTRACT_OBJECT_METHODS 00298 PLEARN_DECLARE_OBJECT(GaussPartSupervisedDBN); 00299 00300 // Simply calls inherited::build() then build_() 00301 virtual void build(); 00302 00304 // (PLEASE IMPLEMENT IN .cc) 00305 virtual void makeDeepCopyFromShallowCopy(CopiesMap& copies); 00306 00307 protected: 00308 //##### Protected Options ############################################### 00309 00310 // ### Declare protected option fields (such as learned parameters) here 00311 // ... 00312 00314 mutable TVec< Vec > activation_gradients; 00315 00317 mutable TVec< Vec > expectation_gradients; 00318 00320 mutable Vec output_gradient; 00321 00322 00323 protected: 00324 //##### Protected Member Functions ###################################### 00325 00326 virtual void contrastiveDivergenceStep( 00327 const PP<RBMLayer>& down_layer, 00328 const PP<RBMParameters>& parameters, 00329 const PP<RBMLayer>& up_layer ); 00330 00331 virtual real supervisedContrastiveDivergenceStep( 00332 const PP<RBMLayer>& down_layer, 00333 const PP<RBMParameters>& parameters, 00334 const PP<RBMLayer>& up_layer, 00335 const Vec& target, 00336 int index ); 00337 00338 virtual real greedyStep( const Vec& predictor, int params_index ); 00339 virtual real jointGreedyStep( const Vec& input ); 00340 virtual void fineTuneByGradientDescent( const Vec& input, 00341 const Vec& train_costs ); 00342 00344 static void declareOptions(OptionList& ol); 00345 00346 private: 00347 //##### Private Member Functions ######################################## 00348 00350 void build_(); 00351 00353 void build_layers(); 00354 00356 void build_params(); 00357 00359 void build_regressors(); 00360 00361 #if USING_MPI 00362 void shareParamsMPI(); 00363 #endif 00364 00365 private: 00366 //##### Private Data Members ############################################ 00367 00368 // The rest of the private stuff goes here 00369 00370 #if USING_MPI 00371 00372 Vec global_params; 00375 Vec previous_global_params; 00376 #endif 00377 }; 00378 00379 // Declares a few other classes and functions related to this class 00380 DECLARE_OBJECT_PTR(GaussPartSupervisedDBN); 00381 00382 } // end of namespace PLearn 00383 00384 #endif 00385 00386 00387 /* 00388 Local Variables: 00389 mode:c++ 00390 c-basic-offset:4 00391 c-file-style:"stroustrup" 00392 c-file-offsets:((innamespace . 0)(inline-open . 0)) 00393 indent-tabs-mode:nil 00394 fill-column:79 00395 End: 00396 */ 00397 // vim: filetype=cpp:expandtab:shiftwidth=4:tabstop=8:softtabstop=4:encoding=utf-8:textwidth=79 :