starting an mpi version.

git-svn-id: https://ws10smt.googlecode.com/svn/trunk@253 ec762483-ff6d-05da-a07a-a48fb63a330f

1 files changed, 37 insertions, 15 deletions

diff --git a/gi/pyp-topics/src/pyp.hh b/gi/pyp-topics/src/pyp.hh
index 7a520d6a..dc47244b 100644
--- a/gi/pyp-topics/src/pyp.hh
+++ b/gi/pyp-topics/src/pyp.hh
@@ -4,6 +4,7 @@
 #include <math.h>
 #include <map>
 #include <tr1/unordered_map>
+//#include <google/sparse_hash_map>
 
 #include <boost/random/uniform_real.hpp>
 #include <boost/random/variate_generator.hpp>
@@ -11,6 +12,7 @@
 
 #include "log_add.h"
 #include "slice-sampler.h"
+#include "mt19937ar.h"
 
 //
 // Pitman-Yor process with customer and table tracking
@@ -18,12 +20,17 @@
 
 template <typename Dish, typename Hash=std::tr1::hash<Dish> >
 class PYP : protected std::tr1::unordered_map<Dish, int, Hash>
+//class PYP : protected google::sparse_hash_map<Dish, int, Hash>
 {
 public:
   using std::tr1::unordered_map<Dish,int>::const_iterator;
   using std::tr1::unordered_map<Dish,int>::iterator;
   using std::tr1::unordered_map<Dish,int>::begin;
   using std::tr1::unordered_map<Dish,int>::end;
+//  using google::sparse_hash_map<Dish,int>::const_iterator;
+//  using google::sparse_hash_map<Dish,int>::iterator;
+//  using google::sparse_hash_map<Dish,int>::begin;
+//  using google::sparse_hash_map<Dish,int>::end;
 
   PYP(double a, double b, unsigned long seed = 0, Hash hash=Hash());
 
@@ -39,6 +46,7 @@ public:
 
   int num_customers() const { return _total_customers; }
   int num_types() const { return std::tr1::unordered_map<Dish,int>::size(); }
+  //int num_types() const { return google::sparse_hash_map<Dish,int>::size(); }
   bool empty() const { return _total_customers == 0; }
 
   double log_prob(Dish dish, double log_p0) const;
@@ -79,6 +87,7 @@ private:
     std::map<int, int> table_histogram; // num customers at table -> number tables
   };
   typedef std::tr1::unordered_map<Dish, TableCounter, Hash> DishTableType;
+  //typedef google::sparse_hash_map<Dish, TableCounter, Hash> DishTableType;
   DishTableType _dish_tables;
   int _total_customers, _total_tables;
 
@@ -86,11 +95,10 @@ private:
   typedef boost::uniform_real<> uni_dist_type;
   typedef boost::variate_generator<base_generator_type&, uni_dist_type> gen_type;
 
-  uni_dist_type uni_dist;
-  base_generator_type rng; //this gets the seed
-  gen_type rnd; //instantiate: rnd(rng, uni_dist)
+//  uni_dist_type uni_dist;
+//  base_generator_type rng; //this gets the seed
+//  gen_type rnd; //instantiate: rnd(rng, uni_dist)
                 //call: rnd() generates uniform on [0,1)
- 
 
   // Function objects for calculating the parts of the log_prob for 
   // the parameters a and b
@@ -132,12 +140,12 @@ private:
     }
   };
    
-   /* lbetadist() returns the log probability density of x under a Beta(alpha,beta)
+  /* lbetadist() returns the log probability density of x under a Beta(alpha,beta)
    * distribution. - copied from Mark Johnson's gammadist.c
    */
-    static long double lbetadist(long double x, long double alpha, long double beta);
+  static long double lbetadist(long double x, long double alpha, long double beta);
 
-   /* lgammadist() returns the log probability density of x under a Gamma(alpha,beta)
+  /* lgammadist() returns the log probability density of x under a Gamma(alpha,beta)
    * distribution - copied from Mark Johnson's gammadist.c
    */
   static long double lgammadist(long double x, long double alpha, long double beta);
@@ -146,13 +154,15 @@ private:
 
 template <typename Dish, typename Hash>
 PYP<Dish,Hash>::PYP(double a, double b, unsigned long seed, Hash)
-: std::tr1::unordered_map<Dish, int, Hash>(), _a(a), _b(b), 
+: std::tr1::unordered_map<Dish, int, Hash>(10), _a(a), _b(b), 
+//: google::sparse_hash_map<Dish, int, Hash>(10), _a(a), _b(b), 
   _a_beta_a(1), _a_beta_b(1), _b_gamma_s(1), _b_gamma_c(1),
   //_a_beta_a(1), _a_beta_b(1), _b_gamma_s(10), _b_gamma_c(0.1),
-  _total_customers(0), _total_tables(0),
-  uni_dist(0,1), rng(seed == 0 ? (unsigned long)this : seed), rnd(rng, uni_dist)
+  _total_customers(0), _total_tables(0)//,
+  //uni_dist(0,1), rng(seed == 0 ? (unsigned long)this : seed), rnd(rng, uni_dist)
 {
 //  std::cerr << "\t##PYP<Dish,Hash>::PYP(a=" << _a << ",b=" << _b << ")" << std::endl;
+  //set_deleted_key(-std::numeric_limits<Dish>::max());
 }
 
 template <typename Dish, typename Hash>
@@ -235,7 +245,8 @@ PYP<Dish,Hash>::increment(Dish dish, double p0) {
   assert (pshare >= 0.0);
   //assert (pnew > 0.0);
 
-  if (rnd() < pnew / (pshare + pnew)) {
+  //if (rnd() < pnew / (pshare + pnew)) {
+  if (mt_genrand_res53() < pnew / (pshare + pnew)) {
     // assign to a new table
     tc.tables += 1;
     tc.table_histogram[1] += 1;
@@ -245,7 +256,8 @@ PYP<Dish,Hash>::increment(Dish dish, double p0) {
   else {
     // randomly assign to an existing table
     // remove constant denominator from inner loop
-    double r = rnd() * (c - _a*t);
+    //double r = rnd() * (c - _a*t);
+    double r = mt_genrand_res53() * (c - _a*t);
     for (std::map<int,int>::iterator
          hit = tc.table_histogram.begin();
          hit != tc.table_histogram.end(); ++hit) {
@@ -266,6 +278,7 @@ PYP<Dish,Hash>::increment(Dish dish, double p0) {
   }
 
   std::tr1::unordered_map<Dish,int,Hash>::operator[](dish) += 1;
+  //google::sparse_hash_map<Dish,int,Hash>::operator[](dish) += 1;
   _total_customers += 1;
 
   return delta;
@@ -276,6 +289,7 @@ int
 PYP<Dish,Hash>::count(Dish dish) const
 {
   typename std::tr1::unordered_map<Dish, int>::const_iterator 
+  //typename google::sparse_hash_map<Dish, int>::const_iterator 
     dcit = find(dish);
   if (dcit != end())
     return dcit->second;
@@ -288,6 +302,7 @@ int
 PYP<Dish,Hash>::decrement(Dish dish)
 {
   typename std::tr1::unordered_map<Dish, int>::iterator dcit = find(dish);
+  //typename google::sparse_hash_map<Dish, int>::iterator dcit = find(dish);
   if (dcit == end()) {
     std::cerr << dish << std::endl;
     assert(false);
@@ -296,6 +311,7 @@ PYP<Dish,Hash>::decrement(Dish dish)
   int delta = 0;
 
   typename std::tr1::unordered_map<Dish, TableCounter>::iterator dtit = _dish_tables.find(dish);
+  //typename google::sparse_hash_map<Dish, TableCounter>::iterator dtit = _dish_tables.find(dish);
   if (dtit == _dish_tables.end()) {
     std::cerr << dish << std::endl;
     assert(false);
@@ -307,7 +323,8 @@ PYP<Dish,Hash>::decrement(Dish dish)
   //std::cerr << "count: " << count(dish) << " ";
   //std::cerr << "tables: " << tc.tables << "\n";
 
-  double r = rnd() * count(dish);
+  //double r = rnd() * count(dish);
+  double r = mt_genrand_res53() * count(dish);
   for (std::map<int,int>::iterator hit = tc.table_histogram.begin();
        hit != tc.table_histogram.end(); ++hit)
   {
@@ -357,6 +374,7 @@ int
 PYP<Dish,Hash>::num_tables(Dish dish) const
 {
   typename std::tr1::unordered_map<Dish, TableCounter, Hash>::const_iterator 
+  //typename google::sparse_hash_map<Dish, TableCounter, Hash>::const_iterator 
     dtit = _dish_tables.find(dish);
 
   //assert(dtit != _dish_tables.end());
@@ -379,6 +397,7 @@ PYP<Dish,Hash>::debug_info(std::ostream& os) const
 {
   int hists = 0, tables = 0;
   for (typename std::tr1::unordered_map<Dish, TableCounter, Hash>::const_iterator 
+  //for (typename google::sparse_hash_map<Dish, TableCounter, Hash>::const_iterator 
        dtit = _dish_tables.begin(); dtit != _dish_tables.end(); ++dtit)
   {
     hists += dtit->second.table_histogram.size();
@@ -409,6 +428,7 @@ void
 PYP<Dish,Hash>::clear()
 {
   this->std::tr1::unordered_map<Dish,int,Hash>::clear();
+  //this->google::sparse_hash_map<Dish,int,Hash>::clear();
   _dish_tables.clear();
   _total_tables = _total_customers = 0;
 }
@@ -509,7 +529,8 @@ PYP<Dish,Hash>::resample_prior_b() {
   int niterations = 10;   // number of resampling iterations
   //std::cerr << "\n## resample_prior_b(), initial a = " << _a << ", b = " << _b << std::endl;
   resample_b_type b_log_prob(_total_customers, _total_tables, _a, _b_gamma_c, _b_gamma_s);
-  _b = slice_sampler1d(b_log_prob, _b, rnd, (double) 0.0, std::numeric_limits<double>::infinity(), 
+  //_b = slice_sampler1d(b_log_prob, _b, rnd, (double) 0.0, std::numeric_limits<double>::infinity(), 
+  _b = slice_sampler1d(b_log_prob, _b, random, (double) 0.0, std::numeric_limits<double>::infinity(), 
                        (double) 0.0, niterations, 100*niterations);
   //std::cerr << "\n## resample_prior_b(), final a = " << _a << ", b = " << _b << std::endl;
 }
@@ -523,7 +544,8 @@ PYP<Dish,Hash>::resample_prior_a() {
   int niterations = 10;
   //std::cerr << "\n## Initial a = " << _a << ", b = " << _b << std::endl;
   resample_a_type a_log_prob(_total_customers, _total_tables, _b, _a_beta_a, _a_beta_b, _dish_tables);
-  _a = slice_sampler1d(a_log_prob, _a, rnd, std::numeric_limits<double>::min(), 
+  //_a = slice_sampler1d(a_log_prob, _a, rnd, std::numeric_limits<double>::min(), 
+  _a = slice_sampler1d(a_log_prob, _a, random, std::numeric_limits<double>::min(), 
                        (double) 1.0, (double) 0.0, niterations, 100*niterations);
 }