flush time & bi score

src/segmentor/customized_segmentor.cpp

@@ -222,7 +222,7 @@ CustomizedSegmentor::calculate_scores(Instance * inst, bool use_avg) {
 
   for (int pl = 0; pl < L; ++ pl) {
     for (int l = 0; l < L; ++ l) {
-      int idx = model->space.index(pl, l);
+      int idx = baseline_model->space.index(pl, l);
 
       if(!use_avg) {
         inst->bi_scores[pl][l] = baseline_model->param.dot(idx, false);
@@ -278,6 +278,11 @@ CustomizedSegmentor::train_setup() {
   return true;
 }
 
+int
+CustomizedSegmentor::flush_time(void) {
+  return timestamp + baseline_model->end_time;
+}
+
 void
 CustomizedSegmentor::train_passive_aggressive(int nr_errors) {
   double error = nr_errors;

src/segmentor/customized_segmentor.h

@@ -45,6 +45,9 @@ class CustomizedSegmentor : public Segmentor{
   //!
   void train_averaged_perceptron(void);
 
+  //!
+  int flush_time(void);
+
   //!
   void evaluate(double &p, double &r, double &f);
 

src/segmentor/segmentor.cpp

@@ -644,6 +644,10 @@ Segmentor::train_setup() {
   return true;
 };
 
+int
+Segmentor::flush_time(void) {
+  return timestamp;
+}
 void
 Segmentor::train(void) {
   if (!train_setup()) {
@@ -736,8 +740,9 @@ Segmentor::train(void) {
         }
       }
 
-      model->param.flush( train_dat.size() * (iter + 1) );
-      model->end_time = train_dat.size() * (iter + 1);
+      int t = flush_time();
+      model->param.flush(t);
+      model->end_time = t;
 
       Model * new_model = NULL;
       new_model = erase_rare_features(feature_group_updated_time);

src/segmentor/segmentor.h

@@ -74,6 +74,11 @@ class Segmentor {
    */
   virtual void train_averaged_perceptron();
 
+  /**
+   * Return the flush time after each iteration.
+   */
+  virtual int flush_time(void);
+
   /**
    * The main training process, the training scheme can be summarized as
    *