Merge ../mosesdecoder into perf_moses2

compile.sh

@@ -4,5 +4,5 @@
 
 set -e -o pipefail
 opt=$(pwd)/opt
-./bjam --with-irstlm=$opt --with-boost=$opt --with-cmph=$opt --with-xmlrpc-c=$opt --with-mm --with-probing-pt -j$(getconf _NPROCESSORS_ONLN) $@
+./bjam --with-irstlm=$opt/irstlm-5.80.08 --with-boost=$opt --with-cmph=$opt --with-xmlrpc-c=$opt --with-mm --with-probing-pt -j$(getconf _NPROCESSORS_ONLN) $@
 

contrib/Makefiles/install-dependencies.gmake

@@ -31,7 +31,7 @@ BUILD_DIR = $(CWD)/opt/build/${URL}
 # you can also specify specific prefixes for different packages:
 XMLRPC_PREFIX ?= ${PREFIX}
 CMPH_PREFIX   ?= ${PREFIX}
-IRSTLM_PREFIX ?= ${PREFIX}
+IRSTLM_PREFIX ?= ${PREFIX}/irstlm-5.80.08
 BOOST_PREFIX  ?= ${PREFIX}
 
 # currently, the full enchilada means xmlrpc-c, cmph, irstlm, boost

moses/TranslationModel/UG/check-coverage3.cc

@@ -13,6 +13,8 @@
 #include <sstream>
 #include "mm/ug_bitext_sampler.h"
 
+#include <boost/program_options.hpp>
+namespace po=boost::program_options;
 using namespace Moses;
 using namespace sapt;
 using namespace std;
@@ -21,6 +23,13 @@ using namespace boost;
 typedef sapt::L2R_Token<sapt::SimpleWordId> Token;
 typedef mmBitext<Token> bitext_t;
 
+size_t topN;
+string docname;
+string reference_file;
+string domain_name;
+string bname, L1, L2;
+string ifile;
+
 struct mycmp 
 {
   bool operator() (pair<string,uint32_t> const& a, 
@@ -30,36 +39,70 @@ struct mycmp
   }
 };
 
+
+
+void interpret_args(int ac, char* av[]);
+
 string 
-basename(string const path, string const suffix)
+basename(string const path)
 {
   size_t p = path.find_last_of("/");
-  size_t k = path.size() - suffix.size();
+  string dot = ".";
-  cout << path << " " << suffix << endl;
+  size_t k = path.find((dot + L1),p+1);
-  cout << path.substr(0,p) << " " << path.substr(k) << endl;
+  if (k == string::npos) k = path.find(dot + L1 + ".gz");
-  return path.substr(p+1, suffix == &path[k] ? k-p-1 : path.size() - p);
+  if (k == string::npos) return path.substr(p+1);
+  return path.substr(p+1, k-p-1);
+}
+
+void 
+print_evidence_list(bitext_t const& B, std::map<uint32_t, uint32_t> const& indoc)
+{
+  typedef std::map<uint32_t, uint32_t>::const_iterator iter;
+  typedef pair<size_t,string> item;
+  vector<item> where; 
+  where.reserve(indoc.size());
+  
+  for (iter d = indoc.begin(); d != indoc.end(); ++d)
+    where.push_back(item(d->second, B.docid2name(d->first)));
+  sort(where.begin(),where.end(),greater<item>());
+  BOOST_FOREACH(item const& doc, where)
+    if (domain_name == doc.second)
+      cout << (boost::format("\t\t%4d ! %s") % doc.first % doc.second) << endl;
+    else
+      cout << (boost::format("\t\t%4d   %s") % doc.first % doc.second) << endl;
 }
 
 int main(int argc, char* argv[])
 {
-  boost::intrusive_ptr<bitext_t> B(new bitext_t);
+  boost::shared_ptr<bitext_t> B(new bitext_t);
-  B->open(argv[1],argv[2],argv[3]);
+  interpret_args(argc,argv);
-  string line;
+
-  string ifile = argv[4];
+  B->open(bname, L1, L2);
-  string docname = basename(ifile, string(".") + argv[2] + ".gz");
+  string line, refline;
-  id_type docid = B->docname2docid(docname);
+  if (domain_name == "" && ifile != "-")
-  boost::iostreams::filtering_istream in;
+    domain_name = basename(ifile);
+  
+  id_type docid = B->docname2docid(domain_name);
+  boost::iostreams::filtering_istream in, ref;
   ugdiss::open_input_stream(ifile,in);
+  if (reference_file.size()) 
+    ugdiss::open_input_stream(reference_file,ref);
+
   while(getline(in,line))
     {
-      cout << line << " [" << docname << "]" << endl;
+      if (reference_file.size()) getline(ref, refline);
+      cout << string(80,'-') << endl;
+      cout << " [" << domain_name << "]" << endl;
+      cout << line << endl;
+      if (refline.size()) cout << refline << endl;
+      cout << string(80,'-') << endl;
       vector<id_type> snt;
       B->V1->fillIdSeq(line,snt);
       for (size_t i = 0; i < snt.size(); ++i)
 	{
 	  bitext_t::iter m(B->I1.get());
 	  for (size_t k = i; k < snt.size() && m.extend(snt[k]); ++k);
-	  for (size_t num_occurrences = m.ca(); m.size(); m.up())
+	  for (size_t num_occurrences = 0; m.size(); m.up())
 	    {
 	      if (size_t(m.ca()) == num_occurrences) continue;
 	      num_occurrences = m.ca();
@@ -68,40 +111,45 @@ int main(int argc, char* argv[])
 				     sapt::random_sampling);
 	      s();
 	      if (s.stats()->trg.size() == 0) continue;
-	      // if (s.stats()->indoc[docname] > 10) continue;
 	      sapt::pstats::indoc_map_t::const_iterator d
 		= s.stats()->indoc.find(docid);
 	      size_t indoccnt = d != s.stats()->indoc.end() ? d->second : 0;
-	      cout << m.size() << " : " << m.str(B->V1.get()) << " (" 
+	      cout << m.str(B->V1.get()) << " (" 
 		   << s.stats()->trg.size() << " entries; " 
 		   << indoccnt << "/" << s.stats()->good 
-		   << " samples in domain)" << endl;
+		   << " samples in domain; " << num_occurrences
+		   << " occ.)" << endl;
 	      vector<PhrasePair<Token> > ppairs;
 	      PhrasePair<Token>::SortDescendingByJointCount sorter;
 	      expand(m,*B,*s.stats(),ppairs,NULL);
 	      sort(ppairs.begin(),ppairs.end(),sorter);
 	      boost::format fmt("%4d/%d/%d |%s| (%4.2f : %4.2f)"); 
+	      size_t ctr = 0;
+	      bool skipped_some = false;
 	      BOOST_FOREACH(PhrasePair<Token>& ppair, ppairs)
 		{
-		  if (ppair.joint * 100 < ppair.good1) break;
+		  if (++ctr > topN && ppair.indoc.find(docid) == ppair.indoc.end())
+		    {
+		      skipped_some = true;
+		      continue;
+		    }
+		  if (skipped_some) 
+		    {
+		      cout << string(17,' ') << "..." << endl;
+		      skipped_some = false;
+		    }
+		  // if (ppair.joint * 100 < ppair.good1) break;
 		  ppair.good2 = ppair.raw2 * float(ppair.good1)/ppair.raw1;
 		  ppair.good2 = max(ppair.good2, ppair.joint);
 
-#if 0
+#if 1
 		  cout << "\t" 
 		       << (fmt % ppair.joint % ppair.good1 % ppair.good2
 			   % B->T2->pid2str(B->V2.get(),ppair.p2)
 			   % (float(ppair.joint)/ppair.good1)
 			   % (float(ppair.joint)/ppair.good2)
 			   ) << "\n";
-		  typedef std::map<uint32_t, uint32_t>::const_iterator iter;
+		  print_evidence_list(*B, ppair.indoc);
-		  for (iter d = ppair.indoc.begin(); d != ppair.indoc.end(); ++d)
-		    {
-		      // if (d != ppair.indoc.begin()) cout << "; ";
-		      cout << (boost::format("\t\t%4d %s") % d->second 
-			       % B->docid2name(d->first))  
-			   << endl;
-		    }
 		  cout << endl;
 #else
 		  cout << "\t" 
@@ -126,3 +174,48 @@ int main(int argc, char* argv[])
 	}
     }
 }
+
+void
+interpret_args(int ac, char* av[])
+{
+  po::variables_map vm;
+  po::options_description o("Options");
+  o.add_options()
+
+    ("help,h",  "print this message")
+    ("top,n", po::value<size_t>(&topN)->default_value(5),
+     "max. number of entries to show")
+    ("domain,D", po::value<string>(&domain_name),
+     "domain name (when reading from stdin)")
+    ("reference,r", po::value<string>(&reference_file),
+     "reference file")
+    ;
+
+  po::options_description h("Hidden Options");
+  h.add_options()
+    ("bname", po::value<string>(&bname), "base name of corpus")
+    ("L1", po::value<string>(&L1), "L1 tag")
+    ("L2", po::value<string>(&L2), "L2 tag")
+    ("input", po::value<string>(&ifile), "input file")
+    ;
+
+  h.add(o);
+  po::positional_options_description a;
+  a.add("bname",1);
+  a.add("L1",1);
+  a.add("L2",1);
+  a.add("input",1);
+
+  po::store(po::command_line_parser(ac,av)
+            .options(h)
+            .positional(a)
+            .run(),vm);
+  po::notify(vm);
+  if (vm.count("help"))
+    {
+      std::cout << "\nusage:\n\t" << av[0]
+           << " [options] <model file stem> <L1> <L2> <input file>" << std::endl;
+      std::cout << o << std::endl;
+      exit(0);
+    }
+}

moses/TranslationModel/UG/ptable-lookup-corpus.cc

@@ -29,6 +29,13 @@ int main(int argc, char const* argv[])
   // Only lookup each phrase once
   unordered_set<string> seen;
 
+  string context_weight_spec;
+  params.SetParameter(context_weight_spec,"context-weights",string(""));
+  boost::shared_ptr<ContextScope> scope(new ContextScope);
+  boost::shared_ptr<IOWrapper> none;
+  if (context_weight_spec.size())
+    scope->SetContextWeights(context_weight_spec);
+
   string line;
   while (true) {
     // Input line
@@ -57,7 +64,8 @@ int main(int argc, char const* argv[])
 
         // Setup task for phrase
         boost::shared_ptr<TranslationTask> ttask;
-        ttask = TranslationTask::create(phrase);
+        ttask = TranslationTask::create(phrase, none, scope);
+	
         // Support model combinations (PhraseDictionaryGroup)
         BOOST_FOREACH(PhraseDictionary* p, PhraseDictionary::GetColl()) {
           p->InitializeForInput(ttask);

moses/TranslationModel/UG/sapt_pscore_logcnt.h

@@ -56,7 +56,7 @@ namespace sapt  {
       if (m_specs.find("j") != std::string::npos)
 	(*dest)[i++] = log(pp.joint);
       if (m_specs.find("r2") != std::string::npos)
-	(*dest)[++i] = log(pp.raw2);
+	(*dest)[i] = log(pp.raw2);
     }
   };
 } // namespace sapt

moses/TranslationModel/UG/util/tokenindex.dump.cc

@@ -12,7 +12,7 @@
 #include <iomanip>
 
 using namespace std;
-using namespace ugdiss;
+using namespace sapt;
 int
 main(int argc,char* argv[])
 {

moses/server/TranslationRequest.cpp

@@ -235,6 +235,10 @@ check(std::map<std::string, xmlrpc_c::value> const& param,
 {
   std::map<std::string, xmlrpc_c::value>::const_iterator m = param.find(key);
   if(m == param.end()) return false;
+
+  if (m->second.type() == xmlrpc_c::value::TYPE_BOOLEAN)
+    return xmlrpc_c::value_boolean(m->second);
+
   std::string val = string(xmlrpc_c::value_string(m->second));
   if(val == "true" || val == "True" || val == "TRUE" || val == "1") return true;
   return false;

run-regtests.sh

@@ -13,7 +13,7 @@ eval set -- "$args"
 noserver=false; 
 full=false;
 j=$(getconf _NPROCESSORS_ONLN)
-irstlm=$opt
+irstlm=$opt/irstlm-5.80.08
 boost=$opt
 cmph=$opt
 xmlrpc=--with-xmlrpc-c\=$opt