Merge branch 'master' of ssh://github.com/moses-smt/mosesdecoder

contrib/relent-filter/src/LatticeMBRGrid.cpp

@@ -51,6 +51,7 @@ POSSIBILITY OF SUCH DAMAGE.
 #include "Manager.h"
 #include "StaticData.h"
 
+#include "util/exception.hh"
 
 using namespace std;
 using namespace Moses;
@@ -68,7 +69,7 @@ public:
   /** Add a parameter with key, command line argument, and default value */
   void addParam(gridkey key, const string& arg, float defaultValue) {
     m_args[arg] = key;
-    CHECK(m_grid.find(key) == m_grid.end());
+    UTIL_THROW_IF2(m_grid.find(key) != m_grid.end(), "Duplicate parameter " << arg);
     m_grid[key].push_back(defaultValue);
   }
 

contrib/server/client_multimodel.py

@@ -16,12 +16,21 @@ else:
     import xmlrpc.client as xmlrpclib
 
 
-def translate(input_object, server, weights=None):
+def translate(input_object, server, weights=None, model_name=None):
+    """translate each sentence in an input_object (list, file-like object or other object that iterates over sentences)
+       server is a xmlrpclib.ServerProxy
+       model_name is the name of the PhraseDictionaryMultiModel(Counts) feature function that the weights should be applied to. It is defined in the moses.ini
+       weights is a list of floats (one float per model, or one float per model per feature)
+    """
 
     for line in input_object:
         params = {}
         params['text'] = line
         if weights:
+            if not model_name:
+                sys.stderr.write("Error: if you define weights, you need to specify the feature to which the weights are to be applied (e.g. PhraseDictionaryMultiModel0)\n")
+                sys.exit(1)
+            params['model_name'] = model_name
             params['lambda'] = weights
 
         print server.translate(params)
@@ -77,4 +86,4 @@ if __name__ == '__main__':
     phrase_pairs = read_phrase_pairs(gzip.open('/path/to/moses-regression-tests/models/multimodel/extract.sorted.gz'))
     weights = optimize(phrase_pairs, server, 'PhraseDictionaryMultiModelCounts0')
 
-    translate(sys.stdin, server, weights)
+    translate(sys.stdin, server, weights, 'PhraseDictionaryMultiModelCounts0')

contrib/server/mosesserver.cpp

@@ -1,4 +1,3 @@
-#include "util/check.hh"
 #include <stdexcept>
 #include <iostream>
 #include <vector>
@@ -237,17 +236,19 @@ public:
         }
     }
 
+    si = params.find("model_name");
+    if (si != params.end() && multiModelWeights.size() > 0) {
+        const string model_name = xmlrpc_c::value_string(si->second);
+        PhraseDictionaryMultiModel* pdmm = (PhraseDictionaryMultiModel*) FindPhraseDictionary(model_name);
+        pdmm->SetTemporaryMultiModelWeightsVector(multiModelWeights);
+    }
+
     const StaticData &staticData = StaticData::Instance();
 
     if (addGraphInfo) {
       (const_cast<StaticData&>(staticData)).SetOutputSearchGraph(true);
     }
 
-    if (multiModelWeights.size() > 0) {
-      PhraseDictionaryMultiModel* pdmm = (PhraseDictionaryMultiModel*) PhraseDictionary::GetColl()[0]; //TODO: only works if multimodel is first phrase table
-      pdmm->SetTemporaryMultiModelWeightsVector(multiModelWeights);
-    }
-
     stringstream out, graphInfo, transCollOpts;
     map<string, xmlrpc_c::value> retData;
 
@@ -534,7 +535,6 @@ int main(int argc, char** argv)
   } else {
     myAbyssServer.run();
   }
-  // xmlrpc_c::serverAbyss.run() never returns
-  CHECK(false);
-  return 0;
+  std::cerr << "xmlrpc_c::serverAbyss.run() returned but should not." << std::endl;
+  return 1;
 }

moses/TranslationModel/PhraseDictionaryMultiModel.cpp

@@ -37,9 +37,6 @@ PhraseDictionaryMultiModel::PhraseDictionaryMultiModel(const std::string &line)
   }
 
   size_t numWeights = m_numScoreComponents;
-  if (m_mode == "interpolate") {
-    numWeights--;
-  }
   UTIL_THROW_IF2(m_pdStr.size() != m_multimodelweights.size() &
 		  m_pdStr.size()*numWeights != m_multimodelweights.size(),
 		  "Number of scores and weights are not equal");
@@ -95,9 +92,7 @@ const TargetPhraseCollection *PhraseDictionaryMultiModel::GetTargetPhraseCollect
   std::vector<std::vector<float> > multimodelweights;
 
   if (m_mode == "interpolate") {
-    //interpolation of phrase penalty is skipped, and fixed-value (2.718) is used instead. results will be screwed up if phrase penalty is not last feature
-    size_t numWeights = m_numScoreComponents-1;
-    multimodelweights = getWeights(numWeights, true);
+    multimodelweights = getWeights(m_numScoreComponents, true);
   }
 
   std::map<std::string,multiModelStatistics*>* allStats = new(std::map<std::string,multiModelStatistics*>);
@@ -181,13 +176,10 @@ TargetPhraseCollection* PhraseDictionaryMultiModel::CreateTargetPhraseCollection
 
     Scores scoreVector(m_numScoreComponents);
 
-    for(size_t i = 0; i < m_numScoreComponents-1; ++i) {
+    for(size_t i = 0; i < m_numScoreComponents; ++i) {
       scoreVector[i] = TransformScore(std::inner_product(statistics->p[i].begin(), statistics->p[i].end(), multimodelweights[i].begin(), 0.0));
     }
 
-    //assuming that last value is phrase penalty
-    scoreVector[m_numScoreComponents-1] = 1.0;
-
     statistics->targetPhrase->GetScoreBreakdown().Assign(this, scoreVector);
 
     //correct future cost estimates and total score
@@ -377,10 +369,6 @@ vector<float> PhraseDictionaryMultiModel::MinimizePerplexity(vector<pair<string,
   CleanUpAfterSentenceProcessing(sentence); // free memory used by compact phrase tables
 
   size_t numWeights = m_numScoreComponents;
-  if (m_mode == "interpolate") {
-    //interpolation of phrase penalty is skipped, and fixed-value (2.718) is used instead. results will be screwed up if phrase penalty is not last feature
-    numWeights = m_numScoreComponents-1;
-  }
 
   vector<float> ret (m_numModels*numWeights);
   for (size_t iFeature=0; iFeature < numWeights; iFeature++) {

moses/TranslationModel/PhraseDictionaryMultiModelCounts.cpp

@@ -243,12 +243,11 @@ TargetPhraseCollection* PhraseDictionaryMultiModelCounts::CreateTargetPhraseColl
       double lexst = ComputeWeightedLexicalTranslation(static_cast<const Phrase&>(*statistics->targetPhrase), src, alignedToS, m_lexTable_e2f, multimodelweights[1], false );
       double lexts = ComputeWeightedLexicalTranslation(src, static_cast<const Phrase&>(*statistics->targetPhrase), alignedToT, m_lexTable_f2e, multimodelweights[3], true );
 
-      Scores scoreVector(5);
+      Scores scoreVector(4);
       scoreVector[0] = FloorScore(TransformScore(m_combineFunction(statistics->fst, statistics->ft, multimodelweights[0])));
       scoreVector[1] = FloorScore(TransformScore(lexst));
       scoreVector[2] = FloorScore(TransformScore(m_combineFunction(statistics->fst, fs, multimodelweights[2])));
       scoreVector[3] = FloorScore(TransformScore(lexts));
-      scoreVector[4] = FloorScore(TransformScore(2.718));
 
       statistics->targetPhrase->GetScoreBreakdown().Assign(this, scoreVector);