diff --git a/tupa/scripts/visualize_learning_curve.py b/tupa/scripts/visualize_learning_curve.py
index f74b457c..a8055d1a 100644
--- a/tupa/scripts/visualize_learning_curve.py
+++ b/tupa/scripts/visualize_learning_curve.py
@@ -16,29 +16,22 @@
 def load_scores(basename, div="dev"):
     filename = "%s.%s.csv" % (basename, div)
     print("Loading %s scores from '%s'" % (div, filename))
-    offset = 0
     try:
         with open(filename) as f:
-            if "iteration" not in f.readline():
-                offset = -1
+            columns = [i for i, text in enumerate(f.readline().split(",")) if text == "f"]
             scores = np.genfromtxt(f, delimiter=",", invalid_raise=False)
     except ValueError as e:
         raise ValueError("Failed reading '%s'" % filename) from e
     try:
-        return scores[:, [PRIMARY_F1_COLUMN + offset, REMOTE_F1_COLUMN + offset]]
-    except IndexError:
-        try:
-            return scores[:, PRIMARY_F1_COLUMN + offset]
-        except IndexError as e:
-            raise ValueError("Failed reading '%s'" % filename) from e
+        return scores[:, columns]
+    except IndexError as e:
+        raise ValueError("Failed reading '%s'" % filename) from e
 
 
 def visualize(scores, filename, div="dev"):
     plt.plot(range(1, 1 + len(scores)), scores)
     plt.xlabel("epochs")
     plt.ylabel("%s f1" % div)
-    if len(scores.shape) > 1:
-        plt.legend(["primary", "remote"])
     plt.title(filename)
     output_file = "%s.%s.png" % (filename, div)
     plt.savefig(output_file)