From 8eed3231400614e6c34a20434a5df9f740a08da8 Mon Sep 17 00:00:00 2001
From: Vansil <36865898+Vansil@users.noreply.github.com>
Date: Thu, 13 Jun 2019 22:15:19 +0200
Subject: [PATCH] Track validation F1 score of best model

---
 train.py | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/train.py b/train.py
index c9ec85e..482ea73 100755
--- a/train.py
+++ b/train.py
@@ -211,10 +211,9 @@ def main(args):
     val_ppl, val_f1 = eval(val_data, model, 
                            samples = args.mc_samples, count_eos_ppl = args.count_eos_ppl)
     print('--------------------------------')
-    if val_f1 > best_val_f1:
-      best_val_f1 = val_f1
     if val_ppl < best_val_ppl:
       best_val_ppl = val_ppl
+      best_val_f1 = val_f1
       checkpoint = {
         'args': args.__dict__,
         'model': model.cpu(),