From 129007276408568ddc3fc1852e0548f8708f4f9f Mon Sep 17 00:00:00 2001
From: Philipp Moritz <pcmoritz@gmail.com>
Date: Thu, 11 Jan 2018 11:09:01 -0800
Subject: [PATCH] [rllib] Expose PPO evaluator resource requirements (#1391)

---
 build.sh                              | 2 +-
 python/ray/rllib/ppo/ppo.py           | 9 +++++++--
 python/ray/rllib/ppo/ppo_evaluator.py | 3 ---
 3 files changed, 8 insertions(+), 6 deletions(-)

diff --git a/build.sh b/build.sh
index b5e5a2b70..47e84e1d2 100755
--- a/build.sh
+++ b/build.sh
@@ -32,7 +32,7 @@ popd
 bash "$ROOT_DIR/src/thirdparty/download_thirdparty.sh"
 bash "$ROOT_DIR/src/thirdparty/build_thirdparty.sh" $PYTHON_EXECUTABLE
 
-# Now build everything.
+# Now we build everything.
 pushd "$ROOT_DIR/python/ray/core"
   # We use these variables to set PKG_CONFIG_PATH, which is important so that
   # in cmake, pkg-config can find plasma.
diff --git a/python/ray/rllib/ppo/ppo.py b/python/ray/rllib/ppo/ppo.py
index 184988f22..841386d2b 100644
--- a/python/ray/rllib/ppo/ppo.py
+++ b/python/ray/rllib/ppo/ppo.py
@@ -14,7 +14,7 @@ import ray
 from ray.tune.result import TrainingResult
 from ray.rllib.agent import Agent
 from ray.rllib.utils import FilterManager
-from ray.rllib.ppo.ppo_evaluator import PPOEvaluator, RemotePPOEvaluator
+from ray.rllib.ppo.ppo_evaluator import PPOEvaluator
 from ray.rllib.ppo.rollout import collect_samples
 
 
@@ -67,6 +67,8 @@ DEFAULT_CONFIG = {
     "min_steps_per_task": 1000,
     # Number of actors used to collect the rollouts
     "num_workers": 5,
+    # Resource requirements for remote actors
+    "worker_resources": {"num_cpus": 1},
     # Dump TensorFlow timeline after this many SGD minibatches
     "full_trace_nth_sgd_batch": -1,
     # Whether to profile data loading
@@ -85,7 +87,8 @@ DEFAULT_CONFIG = {
 
 class PPOAgent(Agent):
     _agent_name = "PPO"
-    _allow_unknown_subkeys = ["model", "tf_session_args", "env_config"]
+    _allow_unknown_subkeys = ["model", "tf_session_args", "env_config",
+                              "worker_resources"]
     _default_config = DEFAULT_CONFIG
 
     def _init(self):
@@ -93,6 +96,8 @@ class PPOAgent(Agent):
         self.kl_coeff = self.config["kl_coeff"]
         self.local_evaluator = PPOEvaluator(
             self.registry, self.env_creator, self.config, self.logdir, False)
+        RemotePPOEvaluator = ray.remote(
+            **self.config["worker_resources"])(PPOEvaluator)
         self.remote_evaluators = [
             RemotePPOEvaluator.remote(
                 self.registry, self.env_creator, self.config, self.logdir,
diff --git a/python/ray/rllib/ppo/ppo_evaluator.py b/python/ray/rllib/ppo/ppo_evaluator.py
index e559b354d..7eb0291e2 100644
--- a/python/ray/rllib/ppo/ppo_evaluator.py
+++ b/python/ray/rllib/ppo/ppo_evaluator.py
@@ -239,6 +239,3 @@ class PPOEvaluator(Evaluator):
             if flush_after:
                 f.clear_buffer()
         return return_filters
-
-
-RemotePPOEvaluator = ray.remote(PPOEvaluator)