remote filtration

scripts/train.py

@@ -95,16 +95,8 @@ def evaluate(onlyImproved=False):
         pass
 
       dists[i] = min(dist, dists[i]) # track the best distance
-      # filter the results by the distance, to ignore the outliers
-      maxValue = 0.1
-      if dists[i] < maxValue:
-        totalLoss.append(loss)
-        totalDist.append(dist)
-      else:
-        # prevent the big "jumps" in the loss and distance when the model is becoming better
-        # assuming that maxValue is bigger than the corresponding loss
-        totalLoss.append(maxValue)
-        totalDist.append(maxValue)
+      totalLoss.append(loss)
+      totalDist.append(dist)
       continue
     if not onlyImproved:
       print('Mean loss: %.5f | Mean distance: %.5f' % (