Will continue after lunch.

fastlib/trunk/contrib/dongryel/thesis_research/mlpack/mixed_logit_dcm/mixed_logit_dcm.h

@@ -17,23 +17,11 @@ namespace mlpack {
 namespace mixed_logit_dcm {
 template<typename IncomingTableType>
 class MixedLogitDCM {
-  private:
-
-    double Normalization_(const core::table::DensePoint &beta);
-
   public:
     typedef IncomingTableType TableType;
 
   public:
 
-    int num_dimensions() const;
-
-    double Evaluate(const arma::vec &iterate) const;
-
-    void Gradient(const arma::vec &iterate, arma::vec *gradient_out) const;
-
-    void Hessian(const arma::vec &iterate, arma::mat *hessian_out) const;
-
     TableType *attribute_table();
 
     void Init(

fastlib/trunk/contrib/dongryel/thesis_research/mlpack/mixed_logit_dcm/mixed_logit_dcm_dev.h

@@ -12,28 +12,6 @@
 namespace mlpack {
 namespace mixed_logit_dcm {
 
-template<typename TableType>
-int MixedLogitDCM<TableType>::num_dimensions() const {
-  return table_.num_parameters();
-}
-
-template<typename TableType>
-double MixedLogitDCM<TableType>::Evaluate(const arma::vec &iterate) const {
-  //return table_.SimulatedLogLikelihood();
-}
-
-template<typename TableType>
-void MixedLogitDCM<TableType>::Gradient(
-  const arma::vec &iterate, arma::vec *gradient_out) const {
-  //table_.SimulatedLoglikelihoodGradient(gradient_out);
-}
-
-template<typename TableType>
-void MixedLogitDCM<TableType>::Hessian(
-  const arma::vec &iterate, arma::mat *hessian_out) const {
-  //table_.SimulatedLoglikelihoodHessian(hessian_out);
-}
-
 template<typename TableType>
 void MixedLogitDCM<TableType>::Init(
   mlpack::mixed_logit_dcm::MixedLogitDCMArguments <
@@ -60,10 +38,10 @@ void MixedLogitDCM<TableType>::Compute(
     static_cast<int>(
       table_.num_people() *
       arguments_in.initial_dataset_sample_rate_);
-  std::vector<int> num_integration_samples(
-    num_data_samples, std::max(
+  int num_integration_samples =
+    std::max(
       static_cast<int>(arguments_in.initial_integration_sample_rate_ * R_MAX),
-      36));
+      36);
 
   // Compute the initial simulated log-likelihood, the gradient, and
   // the Hessian.
@@ -73,26 +51,20 @@ void MixedLogitDCM<TableType>::Compute(
   //table_.SimulatedLoglikelihoodGradient(&current_gradient);
   //table_.SimulatedLoglikelihoodHessian(&current_hessian);
 
-  // Initialize the starting optimization parameter $\theta_0$.
-  arma::vec theta;
+  // Initialize the starting optimization parameter $\theta_0$ and its
+  // associated sampling information.
+  typedef mlpack::mixed_logit_dcm::DCMTable<TableType> DCMTableType;
+  std::pair <
+  arma::vec ,
+       mlpack::mixed_logit_dcm::MixedLogitDCMSampling<DCMTableType> >
+       iterate_sampling_pair;
+  iterate_sampling_pair.first.set_size(
+    arguments_in.distribution_->num_parameters());
+  iterate_sampling_pair.first.zeros();
+  iterate_sampling_pair.second.Init(
+    &table_, num_data_samples, num_integration_samples);
+
 
-  // The trust region optimizer.
-  typedef MixedLogitDCM<TableType> FunctionType;
-  core::optimization::TrustRegion<FunctionType> trust_region;
-  if(arguments_in.trust_region_search_method_ == "cauchy") {
-    trust_region.Init(
-      *this, core::optimization::TrustRegionSearchMethod::CAUCHY);
-  }
-  else if(arguments_in.trust_region_search_method_ == "dogleg") {
-    trust_region.Init(
-      *this, core::optimization::TrustRegionSearchMethod::DOGLEG);
-  }
-  else {
-    trust_region.Init(
-      *this, core::optimization::TrustRegionSearchMethod::STEIHAUG);
-  }
-  trust_region.set_max_radius(arma::norm(current_gradient, 2));
-  trust_region.Optimize(-1, &theta);
 }
 
 template<typename TableType>

fastlib/trunk/contrib/dongryel/thesis_research/mlpack/mixed_logit_dcm/mixed_logit_dcm_sampling.h

@@ -55,7 +55,7 @@ class MixedLogitDCMSampling {
      */
     int num_active_people_;
 
-    std::vector<int> num_integration_samples_;
+    arma::ivec num_integration_samples_;
 
   private:
 
@@ -99,6 +99,20 @@ class MixedLogitDCMSampling {
       }
     }
 
+    void BuildSamples_() {
+      for(int i = 0; i < num_active_people_; i++) {
+
+        // Get the index of the active person.
+        int person_index = dcm_table_->shuffled_indices_for_person(i);
+
+        for(int j = simulated_choice_probabilities_[j].num_samples();
+            j < num_integration_samples_[person_index]; j++) {
+
+          // Draw a beta from the parameter theta.
+        }
+      }
+    }
+
   public:
 
     double simulated_choice_probability(int person_index) const {
@@ -271,11 +285,16 @@ class MixedLogitDCMSampling {
     void Init(
       DCMTableType *dcm_table_in,
       int num_active_people_in,
-      const std::vector<int> &num_integration_samples_in) {
+      int initial_num_integration_samples_in) {
 
       dcm_table_ = dcm_table_in;
       num_active_people_ = num_active_people_in;
-      num_integration_samples_ = num_integration_samples_in;
+      num_integration_samples_.zeros(dcm_table_->num_people());
+      for(int i = 0; i < num_active_people_; i++) {
+        int person_index = dcm_table_->shuffled_indices_for_person(i);
+        num_integration_samples_[person_index] =
+          initial_num_integration_samples_in;
+      }
 
       // This vector maintains the running simulated choice
       // probabilities per person.
@@ -295,13 +314,14 @@ class MixedLogitDCMSampling {
       simulated_loglikelihood_hessians_.resize(dcm_table_->num_people());
 
       // Build up the samples.
-
+      BuildSamples_();
     }
 
     void AddActivePeople(int num_additional_people) {
       num_active_people_ += num_additional_people;
 
       // Build up additional samples for the new people.
+      BuildSamples_();
     }
 
     const std::vector <