Test example applications and rllib in jenkins tests. (#707)

* Test example applications in Jenkins.

* Fix default upload_dir argument for Algorithm class.

* Fix evolution strategies.

* Comment out policy gradient example which doesn't seem to work.

* Set --env-name for evolution strategies.

python/ray/rllib/a3c/example.py

@@ -17,7 +17,9 @@ if __name__ == "__main__":
     parser.add_argument("--redis-address", default=None, type=str,
                         help="The Redis address of the cluster.")
     parser.add_argument("--num-workers", default=4, type=int,
-                        help="The number of A3C workers to use>")
+                        help="The number of A3C workers to use.")
+    parser.add_argument("--iterations", default=-1, type=int,
+                        help="The number of training iterations to run.")
 
     args = parser.parse_args()
     ray.init(redis_address=args.redis_address, num_cpus=args.num_workers)
@@ -27,6 +29,8 @@ if __name__ == "__main__":
 
     a3c = A3C(args.environment, config)
 
-    while True:
+    iteration = 0
+    while iteration != args.iterations:
+        iteration += 1
         res = a3c.train()
         print("current status: {}".format(res))

python/ray/rllib/common.py

@@ -67,7 +67,7 @@ class Algorithm(object):
     TODO(ekl): support checkpoint / restore of training state.
     """
 
-    def __init__(self, env_name, config, upload_dir="file:///tmp/ray"):
+    def __init__(self, env_name, config, upload_dir=None):
         """Initialize an RLLib algorithm.
 
         Args:
@@ -77,6 +77,7 @@ class Algorithm(object):
             should be placed. Can be local like file:///tmp/ray/ or on S3
             like s3://bucketname/.
         """
+        upload_dir = "file:///tmp/ray" if upload_dir is None else upload_dir
         self.experiment_id = uuid.uuid4()
         self.env_name = env_name
         self.config = config

python/ray/rllib/evolution_strategies/example.py

@@ -21,6 +21,8 @@ if __name__ == "__main__":
                         help="The stepsize to use.")
     parser.add_argument("--redis-address", default=None, type=str,
                         help="The Redis address of the cluster.")
+    parser.add_argument("--iterations", default=-1, type=int,
+                        help="The number of training iterations to run.")
 
     args = parser.parse_args()
     num_workers = args.num_workers
@@ -30,11 +32,13 @@ if __name__ == "__main__":
     ray.init(redis_address=args.redis_address,
              num_workers=(0 if args.redis_address is None else None))
 
-    config = DEFAULT_CONFIG._replace(
-        num_workers=num_workers,
-        stepsize=stepsize)
+    config = DEFAULT_CONFIG.copy()
+    config["num_workers"] = num_workers
+    config["stepsize"] = stepsize
 
     alg = EvolutionStrategies(env_name, config)
-    while True:
+    iteration = 0
+    while iteration != args.iterations:
+        iteration += 1
         result = alg.train()
         print("current status: {}".format(result))

python/ray/rllib/policy_gradient/example.py

@@ -21,12 +21,16 @@ if __name__ == "__main__":
                         help="Run the script inside of tf-dbg.")
     parser.add_argument("--load-checkpoint", default=None, type=str,
                         help="Continue training from a checkpoint.")
+    parser.add_argument("--iterations", default=None, type=int,
+                        help="The number of training iterations to run.")
 
     args = parser.parse_args()
     config = DEFAULT_CONFIG.copy()
     config["use_tf_debugger"] = args.use_tf_debugger
-    if args.load_checkpoint:
+    if args.load_checkpoint is not None:
         config["load_checkpoint"] = args.load_checkpoint
+    if args.iterations is not None:
+        config["max_iterations"] = args.iterations
 
     ray.init(redis_address=args.redis_address)