dtrain: all pairs

3 files changed, 25 insertions, 11 deletions

diff --git a/training/dtrain/dtrain.cc b/training/dtrain/dtrain.cc
index bbe14547..30c64037 100644
--- a/training/dtrain/dtrain.cc
+++ b/training/dtrain/dtrain.cc
@@ -48,6 +48,7 @@ main(int argc, char** argv)
   const string adadelta_output   = conf["adadelta_output"].as<string>();
   const size_t max_input         = conf["stop_after"].as<size_t>();
   const bool batch               = conf["batch"].as<bool>();
+  const bool all                 = conf["all"].as<bool>();
 
   // setup decoder
   register_feature_functions();
@@ -267,14 +268,14 @@ main(int argc, char** argv)
       num_up += update_structured(sample, updates, margin,
                                   out_up, i);
     else if (all_pairs)
-      num_up += updates_all(sample, updates, max_up, threshold,
+      num_up += updates_all(sample, updates, max_up, margin, threshold, all,
                             out_up, i);
     else if (pro)
       num_up += updates_pro(sample, updates, cut, max_up, threshold,
                             out_up, i);
     else
       num_up += updates_multipartite(sample, updates, cut, margin,
-                                     max_up, threshold, adjust_cut,
+                                     max_up, threshold, adjust_cut, all,
                                      out_up, i);
 
     SparseVector<weight_t> lambdas_copy;
diff --git a/training/dtrain/dtrain.h b/training/dtrain/dtrain.h
index 883e6028..83dcd945 100644
--- a/training/dtrain/dtrain.h
+++ b/training/dtrain/dtrain.h
@@ -71,6 +71,8 @@ dtrain_init(int argc,
      "use top/bottom 10% (default) of k-best as 'good' and 'bad' for pair sampling, 0 to use all pairs")
     ("adjust,A",           po::bool_switch()->default_value(false),
      "adjust cut for optimal pos. in k-best to cut")
+    ("all,A",              po::bool_switch()->default_value(false),
+     "update using all pairs, ignoring margin and threshold")
     ("score,s",            po::value<string>()->default_value("nakov"),
      "per-sentence BLEU (approx.)")
     ("nakov_fix",         po::value<weight_t>()->default_value(1.0),
@@ -106,7 +108,7 @@ dtrain_init(int argc,
      "output raw data (e.g. k-best lists) [to filename]")
     ("stop_after",         po::value<size_t>()->default_value(numeric_limits<size_t>::max()),
      "only look at this number of segments")
-    ("print_weights,P",    po::value<string>()->default_value("EgivenFCoherent SampleCountF CountEF MaxLexFgivenE MaxLexEgivenF IsSingletonF IsSingletonFE Glue WordPenalty PassThrough LanguageModel LanguageModel_OOV"),
+    ("print_weights",      po::value<string>()->default_value("EgivenFCoherent SampleCountF CountEF MaxLexFgivenE MaxLexEgivenF IsSingletonF IsSingletonFE Glue WordPenalty PassThrough LanguageModel LanguageModel_OOV"),
      "list of weights to print after each iteration");
   po::options_description clopts("Command Line Options");
   clopts.add_options()
diff --git a/training/dtrain/update.h b/training/dtrain/update.h
index 405a3f76..edcfe391 100644
--- a/training/dtrain/update.h
+++ b/training/dtrain/update.h
@@ -20,6 +20,7 @@ updates_multipartite(vector<Hyp>* sample,
                      size_t max_up,
                      weight_t threshold,
                      bool adjust,
+                     bool all,
                      WriteFile& output,
                      size_t id)
 {
@@ -44,9 +45,11 @@ updates_multipartite(vector<Hyp>* sample,
   for (size_t i = 0; i < sep_hi; i++) {
     for (size_t j = sep_hi; j < sz; j++) {
       Hyp& first=(*sample)[i], second=(*sample)[j];
-      if ((first.model-second.model)>margin
-           || (first.gold==second.gold)
-           || (threshold && (first.gold-second.gold < threshold)))
+      if (first.gold==second.gold)
+        continue;
+      if (!all
+          && (((first.model-second.model)>margin)
+              || (threshold && (first.gold-second.gold < threshold))))
         continue;
       if (output)
         *output << id << "\t" << first.f-second.f << endl;
@@ -64,9 +67,11 @@ updates_multipartite(vector<Hyp>* sample,
   for (size_t i = sep_hi; i < sep_lo; i++) {
     for (size_t j = sep_lo; j < sz; j++) {
       Hyp& first=(*sample)[i], second=(*sample)[j];
-      if ((first.model-second.model)>margin
-           || (first.gold==second.gold)
-           || (threshold && (first.gold-second.gold < threshold)))
+      if (first.gold==second.gold)
+        continue;
+      if (!all
+           && (((first.model-second.model)>margin)
+               || (threshold && (first.gold-second.gold < threshold))))
         continue;
       if (output)
         *output << id << "\t" << first.f-second.f << endl;
@@ -83,13 +88,16 @@ updates_multipartite(vector<Hyp>* sample,
  * all pairs
  *  only ignore a pair if gold scores are
  *  identical
+ *  FIXME: that's really _all_
  *
  */
 inline size_t
 updates_all(vector<Hyp>* sample,
             SparseVector<weight_t>& updates,
             size_t max_up,
+            weight_t margin,
             weight_t threshold,
+            bool all,
             WriteFile output,
             size_t id)
 {
@@ -102,8 +110,11 @@ updates_all(vector<Hyp>* sample,
   for (size_t i = 0; i < sz-1; i++) {
     for (size_t j = i+1; j < sz; j++) {
       Hyp& first=(*sample)[i], second=(*sample)[j];
-      if ((first.gold == second.gold)
-           || (threshold && (first.gold-second.gold < threshold)))
+      if (first.gold == second.gold)
+        continue;
+      if (!all
+          && (((first.model-second.model)>margin)
+              || (threshold && (first.gold-second.gold < threshold))))
         continue;
       if (output)
         *output << id << "\t" << first.f-second.f << endl;