[rllib] Default to truncate_episodes and add some more config validators (#2967)

* update

* link it

* warn about truncation

* fix

* Update rllib-training.rst

* deprecate tests failing

python/ray/rllib/agents/ppo/ppo.py

@@ -48,7 +48,7 @@ DEFAULT_CONFIG = with_common_config({
     # Whether to allocate CPUs for workers (if > 0).
     "num_cpus_per_worker": 1,
     # Whether to rollout "complete_episodes" or "truncate_episodes"
-    "batch_mode": "complete_episodes",
+    "batch_mode": "truncate_episodes",
     # Which observation filter to apply to the observation
     "observation_filter": "MeanStdFilter",
     # Use the sync samples optimizer instead of the multi-gpu one
@@ -80,17 +80,7 @@ class PPOAgent(Agent):
             extra_gpu=cf["num_gpus_per_worker"] * cf["num_workers"])
 
     def _init(self):
-        waste_ratio = (
-            self.config["sample_batch_size"] * self.config["num_workers"] /
-            self.config["train_batch_size"])
-        if waste_ratio > 1:
-            msg = ("sample_batch_size * num_workers >> train_batch_size. "
-                   "This means that many steps will be discarded. Consider "
-                   "reducing sample_batch_size, or increase train_batch_size.")
-            if waste_ratio > 1.5:
-                raise ValueError(msg)
-            else:
-                print("Warning: " + msg)
+        self._validate_config()
         self.local_evaluator = self.make_local_evaluator(
             self.env_creator, self._policy_graph)
         self.remote_evaluators = self.make_remote_evaluators(
@@ -114,6 +104,28 @@ class PPOAgent(Agent):
                     "standardize_fields": ["advantages"],
                 })
 
+    def _validate_config(self):
+        waste_ratio = (
+            self.config["sample_batch_size"] * self.config["num_workers"] /
+            self.config["train_batch_size"])
+        if waste_ratio > 1:
+            msg = ("sample_batch_size * num_workers >> train_batch_size. "
+                   "This means that many steps will be discarded. Consider "
+                   "reducing sample_batch_size, or increase train_batch_size.")
+            if waste_ratio > 1.5:
+                raise ValueError(msg)
+            else:
+                print("Warning: " + msg)
+        if self.config["sgd_minibatch_size"] > self.config["train_batch_size"]:
+            raise ValueError(
+                "Minibatch size {} must be <= train batch size {}.".format(
+                    self.config["sgd_minibatch_size"],
+                    self.config["train_batch_size"]))
+        if (self.config["batch_mode"] == "truncate_episodes"
+                and not self.config["use_gae"]):
+            raise ValueError(
+                "Episode truncation is not supported without a value function")
+
     def _train(self):
         prev_steps = self.optimizer.num_steps_sampled
         fetches = self.optimizer.step()

python/ray/rllib/tuned_examples/hopper-ppo.yaml

@@ -10,3 +10,4 @@ hopper-ppo:
         train_batch_size: 160000
         num_workers: 64
         num_gpus: 4
+        batch_mode: complete_episodes

python/ray/rllib/tuned_examples/humanoid-ppo-gae.yaml

@@ -17,3 +17,4 @@ humanoid-ppo-gae:
             free_log_std: true
         num_workers: 64
         num_gpus: 4
+        batch_mode: complete_episodes

python/ray/rllib/tuned_examples/humanoid-ppo.yaml

@@ -15,3 +15,4 @@ humanoid-ppo:
         use_gae: false
         num_workers: 64
         num_gpus: 4
+        batch_mode: complete_episodes

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

@@ -13,4 +13,4 @@ pendulum-ppo:
         num_sgd_iter: 10
         model:
             fcnet_hiddens: [64, 64]
-            squash_to_range: True
+        batch_mode: complete_episodes

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

@@ -6,3 +6,4 @@ cartpole-ppo:
         time_total_s: 300
     config:
         num_workers: 1
+        batch_mode: complete_episodes

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

@@ -15,3 +15,4 @@ pendulum-ppo:
         num_sgd_iter: 10
         model:
             fcnet_hiddens: [64, 64]
+        batch_mode: complete_episodes

python/ray/rllib/tuned_examples/walker2d-ppo.yaml

@@ -9,3 +9,4 @@ walker2d-v1-ppo:
         train_batch_size: 320000
         num_workers: 64
         num_gpus: 4
+        batch_mode: complete_episodes