PLearn 0.1
|
00001 // -*- C++ -*- 00002 00003 // StructuralLearner.h 00004 // 00005 // Copyright (C) 2006 Pierre-Antoine Manzagol 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 /* ******************************************************* 00036 * $Id: .pyskeleton_header 544 2003-09-01 00:05:31Z plearner $ 00037 ******************************************************* */ 00038 00039 // Authors: Pierre-Antoine Manzagol 00040 00044 #ifndef StructuralLearner_INC 00045 #define StructuralLearner_INC 00046 00047 #include <plearn_learners/generic/PLearner.h> 00048 00049 namespace PLearn { 00050 00054 class StructuralLearner : public PLearner 00055 { 00056 typedef PLearner inherited; 00057 00058 public: 00059 //##### Public Build Options ############################################ 00060 00061 real start_learning_rate, decrease_constant; 00062 VMat auxiliary_task_train_set; 00063 real lambda; 00064 real abstention_threshold; 00065 real epsilon; 00066 int index_O; 00067 int nhidden; 00068 int n_auxiliary_wordproblems; 00069 bool separate_features; 00070 int max_stage; 00071 bool use_thetas_for_output_weights; 00072 bool use_thetas_for_hidden_weights; 00073 00074 public: 00075 //##### Public Member Functions ######################################### 00076 00078 StructuralLearner(); 00079 00080 00081 //##### PLearner Member Functions ####################################### 00082 00085 virtual int outputsize() const; 00086 00090 virtual void forget(); 00091 00095 virtual void train(); 00096 00098 virtual void computeOutput(const Vec& input, Vec& output) const; 00099 00101 virtual void computeCostsFromOutputs(const Vec& input, const Vec& output, 00102 const Vec& target, Vec& costs) const; 00103 00106 virtual TVec<std::string> getTestCostNames() const; 00107 00110 virtual TVec<std::string> getTrainCostNames() const; 00111 00112 virtual void computeOutputWithFeatures(TVec<TVec<unsigned int> >& feats, Vec& output, bool use_theta=true, int begin_class = -1, int end_class = -1) const; 00113 00114 void computeFeatures(const Vec& input, const Vec& target, int data_set, int 00115 index, TVec< TVec<unsigned int> >& theFeatureGroups, char featureMask = 0x1F) const; 00116 00117 void updateFeatures(const Vec& input, const Vec& target, TVec< TVec<unsigned int> >& theFeatureGroups, char 00118 featureMask = 0x1F) const; 00119 00120 00121 //virtual void updateDynamicFeatures(hash_map<int, TVec<bool> > token_prediction, int token, int prediction); 00122 00123 virtual void test(VMat testset, PP<VecStatsCollector> test_stats, VMat testoutputs=0, VMat testcosts=0) const; 00124 00125 // *** SUBCLASS WRITING: *** 00126 // While in general not necessary, in case of particular needs 00127 // (efficiency concerns for ex) you may also want to overload 00128 // some of the following methods: 00129 // virtual void computeOutputAndCosts(const Vec& input, const Vec& target, Vec& output, Vec& costs) const; 00130 // virtual void computeCostsOnly(const Vec& input, const Vec& target, Vec& costs) const; 00131 // virtual int nTestCosts() const; 00132 // virtual int nTrainCosts() const; 00133 // virtual void resetInternalState(); 00134 // virtual bool isStatefulLearner() const; 00135 00136 00137 //##### PLearn::Object Protocol ######################################### 00138 00139 // Declares other standard object methods. 00140 // ### If your class is not instantiatable (it has pure virtual methods) 00141 // ### you should replace this by PLEARN_DECLARE_ABSTRACT_OBJECT_METHODS 00142 PLEARN_DECLARE_OBJECT(StructuralLearner); 00143 00144 // Simply calls inherited::build() then build_() 00145 virtual void build(); 00146 00148 // (PLEASE IMPLEMENT IN .cc) 00149 virtual void makeDeepCopyFromShallowCopy(CopiesMap& copies); 00150 00151 protected: 00152 //##### Protected Options ############################################### 00153 00154 mutable Mat activations_gradient; 00155 mutable Vec good_class_softmax_gradient; 00156 mutable Vec bad_class_softmax_gradient; 00157 mutable TVec<Mat> vs_times_thetas; 00158 00159 //int ninputs_onehot; // 00160 00161 // For the model 00162 mutable TVec<Mat> ws, vs, thetas; 00163 mutable TVec<Mat> whids, vhids, thetahids; 00164 00165 // Features for an example 00166 mutable TVec< TVec<unsigned int> > feats; 00167 mutable unsigned int *current_features; 00168 00169 //hash_map<int,TVec<bool> > token_prediction_train; 00170 //TVec<hash_map<int,TVec<bool> > > token_prediction_test; 00171 00172 mutable Mat thetas_times_x; 00173 mutable Mat thetahids_times_x; 00174 mutable Mat activations; 00175 00176 // Bag of words features, over window of chunks, precomputed for 00177 // the training set 00178 //TVec< TVec<unsigned int> > bag_of_words_over_chunks; 00179 00180 // For examples 00181 mutable Vec input, target, before_softmax, output, costs; 00182 mutable real weight; 00183 real learning_rate; 00184 00185 // Temporary files for computeFeatures 00186 mutable TVec<unsigned int> currentFeatureGroup; 00187 mutable bool tag; 00188 mutable int size; 00189 mutable unsigned int fl; 00190 mutable std::string symbol; 00191 00192 // Feature dimensions 00193 mutable TVec<unsigned int> fls; 00194 00195 // Indices of auxiliary examples effectively used and their target ("most frequent word"-tag encoded, ie between 0 and 999) 00196 TMat< unsigned int > auxiliary_indices_current; 00197 TMat< unsigned int > auxiliary_indices_left; 00198 00199 // Viterbi table 00200 mutable TMat< pair<real,int> > viterbi_table; 00201 Vec preds; 00202 00203 mutable std::map<int, int> plcw_bigram_mapping; // maps "previous label - current word" bigrams seen in train_set to an index 00204 00205 real best_error,current_error; 00206 00207 protected: 00208 //##### Protected Member Functions ###################################### 00209 00211 // (PLEASE IMPLEMENT IN .cc) 00212 static void declareOptions(OptionList& ol); 00213 00217 virtual void initializeParams(bool set_seed = true); 00218 00221 void initWordProblemsStructures(); 00222 00224 void initPreviousLabelCurrentWordBigramMapping(); 00225 00226 void buildTasksParameters(int nout, TVec<unsigned int> feat_lengths); 00227 void buildThetaParameters(TVec<unsigned int> feat_lengths); 00228 00229 00230 private: 00231 //##### Private Member Functions ######################################## 00232 00234 // (PLEASE IMPLEMENT IN .cc) 00235 void build_(); 00236 00237 private: 00238 //##### Private Data Members ############################################ 00239 00240 // The rest of the private stuff goes here 00241 }; 00242 00243 // Declares a few other classes and functions related to this class 00244 DECLARE_OBJECT_PTR(StructuralLearner); 00245 00246 } // end of namespace PLearn 00247 00248 #endif 00249 00250 00251 /* 00252 Local Variables: 00253 mode:c++ 00254 c-basic-offset:4 00255 c-file-style:"stroustrup" 00256 c-file-offsets:((innamespace . 0)(inline-open . 0)) 00257 indent-tabs-mode:nil 00258 fill-column:79 00259 End: 00260 */ 00261 // vim: filetype=cpp:expandtab:shiftwidth=4:tabstop=8:softtabstop=4:encoding=utf-8:textwidth=79 :