[rllib] Move repeat field to asv script (#2367)

.gitignore

@@ -29,6 +29,7 @@
 
 # Python byte code files
 *.pyc
+python/.eggs
 
 # Backup files
 *.bak

python/ray/rllib/optimizers/multi_gpu_optimizer.py

@@ -52,7 +52,6 @@ class LocalMultiGPUOptimizer(PolicyOptimizer):
         self.update_weights_timer = TimerStat()
 
         print("LocalMultiGPUOptimizer devices", self.devices)
-        print("LocalMultiGPUOptimizer batch size", self.batch_size)
 
         assert set(self.local_evaluator.policy_map.keys()) == {"default"}, \
             "Multi-agent is not supported"

python/ray/rllib/tuned_examples/regression_tests/cartpole-a3c-pytorch.yaml

@@ -1,7 +1,6 @@
 cartpole-a3c:
     env: CartPole-v0
     run: A3C
-    repeat: 3
     stop:
         episode_reward_mean: 200
         time_total_s: 600

python/ray/rllib/tuned_examples/regression_tests/cartpole-a3c.yaml

@@ -1,7 +1,6 @@
 cartpole-a3c:
     env: CartPole-v0
     run: A3C
-    repeat: 3
     stop:
         episode_reward_mean: 200
         time_total_s: 600

python/ray/rllib/tuned_examples/regression_tests/cartpole-dqn.yaml

@@ -1,7 +1,6 @@
 cartpole-dqn:
     env: CartPole-v0
     run: DQN
-    repeat: 3
     stop:
         episode_reward_mean: 200
         time_total_s: 600

python/ray/rllib/tuned_examples/regression_tests/cartpole-pg.yaml

@@ -1,7 +1,6 @@
 cartpole-pg:
     env: CartPole-v0
     run: PG
-    repeat: 3
     stop:
         episode_reward_mean: 200
         time_total_s: 300

python/ray/rllib/tuned_examples/regression_tests/cartpole-ppo.yaml

@@ -1,7 +1,6 @@
 cartpole-ppo:
     env: CartPole-v0
     run: PPO
-    repeat: 3
     stop:
         episode_reward_mean: 200
         time_total_s: 300

python/ray/rllib/tuned_examples/regression_tests/pendulum-ddpg.yaml

@@ -1,7 +1,6 @@
 pendulum-ddpg:
     env: Pendulum-v0
     run: DDPG
-    repeat: 3
     stop:
         episode_reward_mean: -160
         time_total_s: 900

python/ray/rllib/tuned_examples/regression_tests/pendulum-ppo.yaml

@@ -1,7 +1,6 @@
 pendulum-ppo:
     env: Pendulum-v0
     run: PPO
-    repeat: 3
     stop:
         episode_reward_mean: -160
         # expect -140 within 300-500k steps

python/ray/rllib/tuned_examples/regression_tests/regression_test.py

@@ -22,6 +22,8 @@ CONFIG_DIR = os.path.dirname(os.path.abspath(__file__))
 def _evaulate_config(filename):
     with open(os.path.join(CONFIG_DIR, filename)) as f:
         experiments = yaml.load(f)
+        for _, config in experiments.items():
+            config["repeat"] = 3
     ray.init()
     trials = tune.run_experiments(experiments)
     results = defaultdict(list)