Way to record the replays of the training

README.md

@@ -147,6 +147,14 @@ After Reaver has finished training, you can look at how it performs by appending
 python -m reaver.run --env MoveToBeacon --agent a2c --test --render 2> stderr.log
 ```
 
+To record replays of the training you can use `--replay` and `--replay_dir`. The `replay` flag needs an integger `n`, after `n` episodes a replay will be saved. The `replay_dir` flag will need a `path` where do you want to save the replay, `Linux OS` will save on ~/StarCraftII/Replays/ + `path(replay_dir)` and `Windows OS` will save on `path(replay_dir)`.
+
+```bash
+python -m reaver.run --env MoveToBeacon --agent a2c --test --render --replay 10 --replay_dir a2c_agent_mtb_replay 2> stderr.log
+```
+**NB!** If you want to use the flag `replay` you will need to install manually it from source as described above.
+
+
 ### Google Colab
 
 A companion [Google Colab notebook](https://colab.research.google.com/drive/1DvyCUdymqgjk85FB5DrTtAwTFbI494x7) 

reaver/envs/sc2.py

@@ -30,13 +30,17 @@ def __init__(
         spatial_dim=16,
         step_mul=8,
         obs_features=None,
+        save_replay_episodes=0,
+        replay_dir=None,
         action_ids=ACTIONS_MINIGAMES
     ):
         super().__init__(map_name, render, reset_done, max_ep_len)
 
         self.step_mul = step_mul
         self.spatial_dim = spatial_dim
         self._env = None
+        self.save_replay_episodes = save_replay_episodes
+        self.replay_dir = replay_dir
 
         # sensible action set for all minigames
         if not action_ids or action_ids in [ACTIONS_MINIGAMES, ACTIONS_MINIGAMES_ALL]:
@@ -79,7 +83,10 @@ def start(self):
                 rgb_screen=None,
                 rgb_minimap=None
             )],
-            step_mul=self.step_mul,)
+            step_mul=self.step_mul,
+            save_replay_episodes=self.save_replay_episodes,
+            replay_dir=self.replay_dir
+        )
 
     def step(self, action):
         try:

reaver/run.py

@@ -32,6 +32,11 @@
                   'Run an agent in test mode: restore flag is set to true and number of envs set to 1'
                   'Loss is calculated, but gradients are not applied.'
                   'Checkpoints, summaries, log files are not updated, but console logger is enabled.')
+flags.DEFINE_integer('replay', 0, "Save a replay after this many episodes. Default of 0 means don't save replays.")
+flags.DEFINE_string('replay_dir', None, 'Directory to save replays. '
+                                        'Linux distros will save on ~/StarCraftII/Replays/ + path(replay_dir)'
+                                        'Windows distros will save on path(replay_dir)')
+
 
 flags.DEFINE_alias('e', 'env')
 flags.DEFINE_alias('a', 'agent')
@@ -42,6 +47,8 @@
 flags.DEFINE_alias('la', 'log_eps_avg')
 flags.DEFINE_alias('n', 'experiment')
 flags.DEFINE_alias('g', 'gin_bindings')
+flags.DEFINE_alias('r', 'replay')
+flags.DEFINE_alias('rd', 'replay_dir')
 
 
 def main(argv):
@@ -75,7 +82,10 @@ def main(argv):
     sess_mgr = rvr.utils.tensorflow.SessionManager(sess, expt.path, args.ckpt_freq, training_enabled=not args.test)
 
     env_cls = rvr.envs.GymEnv if '-v' in args.env else rvr.envs.SC2Env
-    env = env_cls(args.env, args.render, max_ep_len=args.max_ep_len)
+    env = env_cls(args.env, args.render,
+                  max_ep_len=args.max_ep_len,
+                  save_replay_episodes=args.replay,
+                  replay_dir=args.replay_dir)
 
     agent = rvr.agents.registry[args.agent](env.obs_spec(), env.act_spec(), sess_mgr=sess_mgr, n_envs=args.n_envs)
     agent.logger = rvr.utils.StreamLogger(args.n_envs, args.log_freq, args.log_eps_avg, sess_mgr, expt.log_path)