From abb1eefdc23f197b7ea7a0e54363a56408a86c61 Mon Sep 17 00:00:00 2001
From: Sven Mika <sven@anyscale.io>
Date: Fri, 11 Dec 2020 22:43:30 +0100
Subject: [PATCH] [RLlib] Issue 12483: Discrete observation space error:
 "ValueError: ('Observation ({}) outside given space ..." when doing
 Trainer.compute_action. (#12787)

---
 rllib/evaluation/worker_set.py | 14 +++++++-------
 rllib/models/preprocessors.py  |  6 +++---
 2 files changed, 10 insertions(+), 10 deletions(-)

diff --git a/rllib/evaluation/worker_set.py b/rllib/evaluation/worker_set.py
index b6a4131ff..80cf617bb 100644
--- a/rllib/evaluation/worker_set.py
+++ b/rllib/evaluation/worker_set.py
@@ -74,7 +74,7 @@ class WorkerSet:
             self.add_workers(num_workers)
 
             # If num_workers > 0, get the action_spaces and observation_spaces
-            # to not be forced to create an Env on the driver.
+            # to not be forced to create an Env on the local worker.
             if self._remote_workers:
                 remote_spaces = ray.get(self.remote_workers(
                 )[0].foreach_policy.remote(
@@ -263,13 +263,13 @@ class WorkerSet:
         elif config["input"] == "sampler":
             input_creator = (lambda ioctx: ioctx.default_sampler_input())
         elif isinstance(config["input"], dict):
-            input_creator = (lambda ioctx: ShuffledInput(
-                MixedInput(config["input"], ioctx), config[
-                    "shuffle_buffer_size"]))
+            input_creator = (
+                lambda ioctx: ShuffledInput(MixedInput(config["input"], ioctx),
+                                            config["shuffle_buffer_size"]))
         else:
-            input_creator = (lambda ioctx: ShuffledInput(
-                JsonReader(config["input"], ioctx), config[
-                    "shuffle_buffer_size"]))
+            input_creator = (
+                lambda ioctx: ShuffledInput(JsonReader(config["input"], ioctx),
+                                            config["shuffle_buffer_size"]))
 
         if isinstance(config["output"], FunctionType):
             output_creator = config["output"]
diff --git a/rllib/models/preprocessors.py b/rllib/models/preprocessors.py
index 51a6f5442..c1b7803c4 100644
--- a/rllib/models/preprocessors.py
+++ b/rllib/models/preprocessors.py
@@ -80,9 +80,9 @@ class Preprocessor:
         obs_space = gym.spaces.Box(-1., 1., self.shape, dtype=np.float32)
         # Stash the unwrapped space so that we can unwrap dict and tuple spaces
         # automatically in model.py
-        if (isinstance(self, TupleFlatteningPreprocessor)
-                or isinstance(self, DictFlatteningPreprocessor)
-                or isinstance(self, RepeatedValuesPreprocessor)):
+        classes = (DictFlatteningPreprocessor, OneHotPreprocessor,
+                   RepeatedValuesPreprocessor, TupleFlatteningPreprocessor)
+        if isinstance(self, classes):
             obs_space.original_space = self._obs_space
         return obs_space