Merge pull request #204 from YuriCat/fix/lazy_memory_limit

(Idea) fix: lazy memory limit

handyrl/train.py

@@ -489,16 +489,16 @@ def feed_episodes(self, episodes):
                 self.generation_results[model_id] = n + 1, r + outcome, r2 + outcome ** 2
 
         # store generated episodes
-        mem = psutil.virtual_memory()
-        mem_used_ratio = mem.used / mem.total
-        mem_ok = mem_used_ratio <= 0.95
-        maximum_episodes = self.args['maximum_episodes'] if mem_ok else len(self.trainer.episodes)
+        self.trainer.episodes.extend([e for e in episodes if e is not None])
+
+        mem_percent = psutil.virtual_memory().percent
+        mem_ok = mem_percent <= 95
+        maximum_episodes = self.args['maximum_episodes'] if mem_ok else int(len(self.trainer.episodes) * 95 / mem_percent)
 
         if not mem_ok and 'memory_over' not in self.flags:
-            warnings.warn("memory usage %.1f%% with buffer size %d" % (mem_used_ratio * 100, len(self.trainer.episodes)))
+            warnings.warn("memory usage %.1f%% with buffer size %d" % (mem_percent, len(self.trainer.episodes)))
             self.flags.add('memory_over')
 
-        self.trainer.episodes.extend([e for e in episodes if e is not None])
         while len(self.trainer.episodes) > maximum_episodes:
             self.trainer.episodes.popleft()