diff --git a/src/ner/ner_frontend.cpp b/src/ner/ner_frontend.cpp
index 37ac104d1..77d70f82f 100644
--- a/src/ner/ner_frontend.cpp
+++ b/src/ner/ner_frontend.cpp
@@ -192,9 +192,9 @@ void NamedEntityRecognizerFrontend::train(void) {
   INFO_LOG("report: allocate %d dimensition parameter.", model->space.dim());
 
   int nr_groups = model->space.num_groups();
-  std::vector<int> groupwise_update_counters;
+  std::vector<size_t> update_counts;
   if (train_opt.rare_feature_threshold > 0) {
-    groupwise_update_counters.resize(nr_groups, 0);
+    update_counts.resize(nr_groups, 0);
     INFO_LOG("report: allocate %d update-time counters", nr_groups);
   } else {
     INFO_LOG("report: model truncation is inactived.");
@@ -203,7 +203,6 @@ void NamedEntityRecognizerFrontend::train(void) {
   int best_iteration = -1;
   double best_f_score = -1.;
 
-  std::vector<size_t> update_counts;
   if (train_opt.rare_feature_threshold > 0) {
     update_counts.resize(nr_groups, 0);
     INFO_LOG("report: allocate %d update-time counters", nr_groups);
@@ -232,7 +231,7 @@ void NamedEntityRecognizerFrontend::train(void) {
       updated_features.add(ctx.predict_features, -1.);
 
       learn(train_opt.algorithm, updated_features, 
-        iter*train_dat.size()+1, inst->num_errors(), model);
+        iter*train_dat.size() + i + 1, inst->num_errors(), model);
 
       if (train_opt.rare_feature_threshold > 0) {
         increase_groupwise_update_counts(model, updated_features, update_counts);
diff --git a/src/postagger/postagger_frontend.cpp b/src/postagger/postagger_frontend.cpp
index cc3c2c897..54b4756e2 100644
--- a/src/postagger/postagger_frontend.cpp
+++ b/src/postagger/postagger_frontend.cpp
@@ -146,10 +146,10 @@ void PostaggerFrontend::train(void) {
   INFO_LOG("report: allocate %d dimensition parameter.", model->space.dim());
 
   int nr_groups = model->space.num_groups();
-  std::vector<int> groupwise_update_counters;
+  std::vector<size_t> update_counts;
 
   if (train_opt.rare_feature_threshold > 0) {
-    groupwise_update_counters.resize(nr_groups, 0);
+    update_counts.resize(nr_groups, 0);
     INFO_LOG("report: allocate %d update-time counters", nr_groups);
   } else {
     INFO_LOG("report: model truncation is inactived.");
@@ -158,7 +158,6 @@ void PostaggerFrontend::train(void) {
   int best_iteration = -1;
   double best_p = -1.;
 
-  std::vector<size_t> update_counts;
 
   for (int iter = 0; iter < train_opt.max_iter; ++ iter) {
     INFO_LOG("Training iteraition #%d", (iter + 1));
@@ -179,7 +178,7 @@ void PostaggerFrontend::train(void) {
       updated_features.add(ctx.predict_features, -1.);
 
       learn(train_opt.algorithm, updated_features,
-        iter*train_dat.size() + 1, inst->num_errors(), model);
+        iter*train_dat.size() + i + 1, inst->num_errors(), model);
 
 
       if (train_opt.rare_feature_threshold > 0) {
diff --git a/src/segmentor/segmentor_frontend.cpp b/src/segmentor/segmentor_frontend.cpp
index 43811e404..c040f7b7e 100644
--- a/src/segmentor/segmentor_frontend.cpp
+++ b/src/segmentor/segmentor_frontend.cpp
@@ -321,12 +321,12 @@ void SegmentorFrontend::train(void) {
   INFO_LOG("report: allocate %d dimensition parameter.", model->space.dim());
 
   int nr_groups = model->space.num_groups();
-  std::vector<int> groupwise_update_counters;
+  std::vector<size_t> update_counts;
 
   // If the rare feature threshold is used, allocate memory for the
   // feature group updated time.
   if (train_opt.rare_feature_threshold > 0) {
-    groupwise_update_counters.resize(nr_groups, 0);
+    update_counts.resize(nr_groups, 0);
     INFO_LOG("report: allocate %d update-time counters", nr_groups);
   } else {
     INFO_LOG("report: model truncation is inactived.");
@@ -335,7 +335,6 @@ void SegmentorFrontend::train(void) {
   int best_iteration = -1;
   double best_p = -1., best_r = -1., best_f = -1.;
 
-  std::vector<size_t> update_counts;
   for (size_t iter = 0; iter < train_opt.max_iter; ++ iter) {
     INFO_LOG("Training iteration #%d", (iter + 1));