trunk: merging some minor fixes/improvements from sandbox/dan, including speed...

trunk: merging some minor fixes/improvements from sandbox/dan, including speed improvement to lattice rescoring code.

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4426 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

src/gmm/diag-gmm.cc

@@ -4,7 +4,7 @@
 //                      Saarland University (Author: Arnab Ghoshal);
 //                      Georg Stemmer;  Jan Silovsky
 //           2012       Arnab Ghoshal
-//           2013       Johns Hopkins University (author: Daniel Povey)
+//           2013-2014  Johns Hopkins University (author: Daniel Povey)
 
 // See ../../COPYING for clarification regarding multiple authors
 //
@@ -941,4 +941,28 @@ void DiagGmm::Generate(VectorBase<BaseFloat> *output) {
   }
 }
 
+DiagGmm::DiagGmm(const GaussClusterable &gc,
+                 BaseFloat var_floor): valid_gconsts_(false) {
+  Vector<BaseFloat> x (gc.x_stats());
+  Vector<BaseFloat> x2 (gc.x2_stats());
+  BaseFloat count =  gc.count();
+  KALDI_ASSERT(count > 0.0);
+  this->Resize(1, x.Dim());
+  x.Scale(1.0/count);
+  x2.Scale(1.0/count);
+  x2.AddVec2(-1.0, x);  // subtract mean^2.
+  x2.ApplyFloor(var_floor);
+  x2.InvertElements();  // get inv-var.
+  KALDI_ASSERT(x2.Min() > 0);
+  Matrix<BaseFloat> mean(1, x.Dim());
+  mean.Row(0).CopyFromVec(x);
+  Matrix<BaseFloat> inv_var(1, x.Dim());
+  inv_var.Row(0).CopyFromVec(x2);
+  this->SetInvVarsAndMeans(inv_var, mean);
+  Vector<BaseFloat> weights(1);
+  weights(0) = 1.0;
+  this->SetWeights(weights);
+  this->ComputeGconsts();
+}
+
 }  // End namespace kaldi

src/gmm/diag-gmm.h

@@ -4,7 +4,7 @@
 //                      Saarland University (Author: Arnab Ghoshal);
 //                      Georg Stemmer;  Jan Silovsky
 //           2012       Arnab Ghoshal
-//           2013       Johns Hopkins University (author: Daniel Povey)
+//           2013-2014  Johns Hopkins University (author: Daniel Povey)
 
 // See ../../COPYING for clarification regarding multiple authors
 //
@@ -31,6 +31,7 @@
 #include "gmm/model-common.h"
 #include "matrix/matrix-lib.h"
 #include "tree/cluster-utils.h"
+#include "tree/clusterable-classes.h"
 
 namespace kaldi {
 
@@ -51,6 +52,10 @@ class DiagGmm {
     CopyFromDiagGmm(gmm);
   }
 
+  /// Initializer from GaussClusterable initializes the DiagGmm as
+  /// a single Gaussian from tree stats.
+  DiagGmm(const GaussClusterable &gc, BaseFloat var_floor);
+
   /// Copies from DiagGmmNormal; does not resize.
   void CopyFromNormal(const DiagGmmNormal &diag_gmm_normal);
   

src/gmmbin/gmm-init-model.cc

@@ -71,11 +71,9 @@ void InitAmGmm(const BuildTreeStatsType &stats,
   for (size_t i = 0; i < summed_stats.size(); i++) {
     GaussClusterable *c =
         static_cast<GaussClusterable*>(summed_stats[i] != NULL ? summed_stats[i] : avg_stats);
-    DiagGmm gmm;
-    Vector<BaseFloat> x (c->x_stats());
-    Vector<BaseFloat> x2 (c->x2_stats());
+    DiagGmm gmm(*c, var_floor);
+    am_gmm->AddPdf(gmm);
     BaseFloat count = c->count();
-    gmm.Resize(1, x.Dim());
     if (count < 100) {
       std::vector<int32> bad_pdfs(1, i), bad_phones;
       GetPhonesForPdfs(trans_model, bad_pdfs, &bad_phones);
@@ -85,24 +83,6 @@ void InitAmGmm(const BuildTreeStatsType &stats,
       KALDI_WARN << "Very small count for state " << i << ": " 
                  << count << "; corresponding phone list: " << ss.str();
     }
-    x.Scale(1.0/count);
-    x2.Scale(1.0/count);
-    x2.AddVec2(-1.0, x);  // subtract mean^2.
-    x2.ApplyFloor(var_floor);
-    x2.InvertElements();  // get inv-var.
-    KALDI_ASSERT(x2.Min() > 0);
-
-    Matrix<BaseFloat> mean(1, x.Dim());
-    mean.Row(0).CopyFromVec(x);
-    Matrix<BaseFloat> inv_var(1, x.Dim());
-    inv_var.Row(0).CopyFromVec(x2);
-
-    gmm.SetInvVarsAndMeans(inv_var, mean);
-    Vector<BaseFloat> weights(1);
-    weights(0) = 1.0;
-    gmm.SetWeights(weights);
-    gmm.ComputeGconsts();
-    am_gmm->AddPdf(gmm);
   }
   DeletePointers(&summed_stats);
   delete avg_stats;
@@ -139,6 +119,7 @@ void InitAmGmmFromOld(const BuildTreeStatsType &stats,
                       int32 P,  // central-position
                       const std::string &old_tree_rxfilename,
                       const std::string &old_model_rxfilename,
+                      BaseFloat var_floor,
                       AmDiagGmm *am_gmm) {
 
   AmDiagGmm old_am_gmm;
@@ -161,15 +142,30 @@ void InitAmGmmFromOld(const BuildTreeStatsType &stats,
   std::vector<BuildTreeStatsType> split_stats;
   SplitStatsByMap(stats, to_pdf_map, &split_stats);
   // Make sure each leaf has stats.
-  for (size_t i = 0; i < split_stats.size(); i++)
-    KALDI_ASSERT(! split_stats[i].empty() && "Tree has leaves with no stats."
-                 "  Modify your roots file as necessary to fix this.");
-  KALDI_ASSERT(static_cast<int32>(split_stats.size()-1) == to_pdf_map.MaxResult()
-               && "Tree may have final leaf with no stats.  "
-               "Modify your roots file as necessary to fix this.");
+  for (size_t i = 0; i < split_stats.size(); i++) {
+    if (split_stats[i].empty()) {
+      KALDI_WARN << "Leaf " << i << " of new tree has no stats.";
+    }
+  }
+  if (static_cast<int32>(split_stats.size()) != to_pdf_map.MaxResult() + 1) {
+    KALDI_ASSERT(static_cast<int32>(split_stats.size()) <
+                 to_pdf_map.MaxResult() + 1);
+    KALDI_WARN << "Tree may have final leaf with no stats.";
+    split_stats.resize(to_pdf_map.MaxResult() + 1);
+    // avoid indexing errors later.
+  }
 
   int32 oldN = old_tree.ContextWidth(), oldP = old_tree.CentralPosition();
 
+  // avg_stats will be used for leaves that have no stats.
+  Clusterable *avg_stats = SumStats(stats);
+  GaussClusterable *avg_stats_gc = dynamic_cast<GaussClusterable*>(avg_stats);
+  KALDI_ASSERT(avg_stats_gc != NULL && "Empty stats input.");
+  DiagGmm avg_gmm(*avg_stats_gc, var_floor);
+  delete avg_stats;
+  avg_stats = NULL;
+  avg_stats_gc = NULL;
+  
   const EventMap &old_map = old_tree.ToPdfMap();
 
   KALDI_ASSERT(am_gmm->NumPdfs() == 0);
@@ -188,7 +184,6 @@ void InitAmGmmFromOld(const BuildTreeStatsType &stats,
     // that align to this "new" pdf... we'll use it to work out the old pdf-id
     // that's "closest" in stats overlap to this new pdf ("pdf").
     std::map<int32, BaseFloat> oldpdf_to_count;
-    KALDI_ASSERT(!my_stats.empty());  // would be code error; checked already.
     for (size_t i = 0; i < my_stats.size(); i++) {
       EventType evec = my_stats[i].first;
       EventAnswerType ans;
@@ -199,7 +194,6 @@ void InitAmGmmFromOld(const BuildTreeStatsType &stats,
       if (oldpdf_to_count.count(ans) == 0) oldpdf_to_count[ans] = stats_count;
       else oldpdf_to_count[ans] += stats_count;
     }
-    KALDI_ASSERT(!oldpdf_to_count.empty());
     BaseFloat max_count = 0; int32 max_old_pdf = -1;
     for (std::map<int32, BaseFloat>::const_iterator iter = oldpdf_to_count.begin();
         iter != oldpdf_to_count.end();
@@ -209,11 +203,15 @@ void InitAmGmmFromOld(const BuildTreeStatsType &stats,
         max_old_pdf = iter->first;
       }
     }
-    KALDI_ASSERT(max_count != 0 && max_old_pdf != -1);
-
+    if (max_count == 0) {  // no overlap - probably a leaf with no stats at all.
+      KALDI_WARN << "Leaf " << pdf << " of new tree being initialized with "
+                 << "globally averaged stats.";
+      am_gmm->AddPdf(avg_gmm);
+    } else {
       am_gmm->AddPdf(old_am_gmm.GetPdf(max_old_pdf));  // Here is where we copy the relevant old PDF.
     }
   }
+}
 
 
 
@@ -289,6 +287,7 @@ int main(int argc, char *argv[]) {
                        ctx_dep.CentralPosition(),
                        old_tree_filename,
                        old_model_filename,
+                       var_floor,
                        &am_gmm);
     }
 

src/latbin/lattice-compose.cc

@@ -45,9 +45,13 @@ int main(int argc, char *argv[]) {
         " or: lattice-compose ark:1.lats G.fst ark:composed.lats\n";
     
     ParseOptions po(usage);
+
+    int32 num_states_cache = 50000;
     int32 phi_label = fst::kNoLabel; // == -1
     po.Register("phi-label", &phi_label, "If >0, the label on backoff arcs of the LM");
-    
+    po.Register("num-states-cache", &num_states_cache,
+                "Number of states we cache when mapping LM FST to lattice type. "
+                "More -> more memory but faster.");
     po.Read(argc, argv);
 
     if (po.NumArgs() != 3) {
@@ -81,9 +85,10 @@ int main(int argc, char *argv[]) {
       if (phi_label > 0)
         PropagateFinal(phi_label, fst2);
 
+      fst::CacheOptions cache_opts(true, num_states_cache);
       fst::StdToLatticeMapper<BaseFloat> mapper;
       fst::MapFst<StdArc, LatticeArc, fst::StdToLatticeMapper<BaseFloat> >
-          mapped_fst2(*fst2, mapper);      
+          mapped_fst2(*fst2, mapper, cache_opts);
       for (; !lattice_reader1.Done(); lattice_reader1.Next()) {
         std::string key = lattice_reader1.Key();
         KALDI_VLOG(1) << "Processing lattice for key " << key;

src/latbin/lattice-lmrescore.cc

@@ -43,8 +43,12 @@ int main(int argc, char *argv[]) {
       
     ParseOptions po(usage);
     BaseFloat lm_scale = 1.0;
+    int32 num_states_cache = 50000;
     
     po.Register("lm-scale", &lm_scale, "Scaling factor for language model costs; frequently 1.0 or -1.0");
+    po.Register("num-states-cache", &num_states_cache,
+                "Number of states we cache when mapping LM FST to lattice type. "
+                "More -> more memory but faster.");
     
     po.Read(argc, argv);
 
@@ -69,9 +73,10 @@ int main(int argc, char *argv[]) {
     // mapped_fst is the LM fst interpreted using the LatticeWeight semiring,
     // with all the cost on the first member of the pair (since it's a graph
     // weight).
+    fst::CacheOptions cache_opts(true, num_states_cache);    
     fst::StdToLatticeMapper<BaseFloat> mapper;
     fst::MapFst<StdArc, LatticeArc, fst::StdToLatticeMapper<BaseFloat> >
-        lm_fst(*std_lm_fst, mapper);
+        lm_fst(*std_lm_fst, mapper, cache_opts);
     delete std_lm_fst;
     
     // The next fifteen or so lines are a kind of optimization and

src/online2bin/ivector-randomize.cc

-// bin/copy-matrix.cc
+// online2bin/ivector-randomize.cc
 
 // Copyright 2014  Johns Hopkins University (author: Daniel Povey)