From 788ac51bb15cea1ef65b6d876f358079b6568f6a Mon Sep 17 00:00:00 2001
From: Guillaume Mollard <guillaume.mollard2@gmail.com>
Date: Mon, 20 May 2019 16:48:20 +0200
Subject: [PATCH] small changes for the observation benchmark, nr_extra=30
---
RailEnvRLLibWrapper.py | 2 +-
experiment_configs/observation_benchmark/config.gin | 2 +-
train_experiment.py | 6 +++---
3 files changed, 5 insertions(+), 5 deletions(-)
diff --git a/RailEnvRLLibWrapper.py b/RailEnvRLLibWrapper.py
index e35a46e..da54ad0 100644
--- a/RailEnvRLLibWrapper.py
+++ b/RailEnvRLLibWrapper.py
@@ -13,7 +13,7 @@ class RailEnvRLLibWrapper(MultiAgentEnv):
# number_of_agents=1,
# obs_builder_object=TreeObsForRailEnv(max_depth=2)):
super(MultiAgentEnv, self).__init__()
- self.rail_generator = config["rail_generator"](nr_start_goal=config['number_of_agents'], min_dist=5,
+ self.rail_generator = config["rail_generator"](nr_start_goal=config['number_of_agents'], min_dist=5, nr_extra=30,
seed=config['seed'] * (1+config.vector_index))
set_seed(config['seed'] * (1+config.vector_index))
self.env = RailEnv(width=config["width"], height=config["height"], rail_generator=self.rail_generator,
diff --git a/experiment_configs/observation_benchmark/config.gin b/experiment_configs/observation_benchmark/config.gin
index 1fdadd2..9f3c072 100644
--- a/experiment_configs/observation_benchmark/config.gin
+++ b/experiment_configs/observation_benchmark/config.gin
@@ -1,6 +1,6 @@
run_experiment.name = "observation_benchmark_results"
run_experiment.num_iterations = 1002
-run_experiment.save_every = 200
+run_experiment.save_every = 100
run_experiment.hidden_sizes = [32, 32]
run_experiment.map_width = 20
diff --git a/train_experiment.py b/train_experiment.py
index d815416..68b4568 100644
--- a/train_experiment.py
+++ b/train_experiment.py
@@ -34,7 +34,7 @@ from ray.rllib.models.preprocessors import TupleFlatteningPreprocessor
ModelCatalog.register_custom_preprocessor("tree_obs_prep", CustomPreprocessor)
ModelCatalog.register_custom_preprocessor("global_obs_prep", TupleFlatteningPreprocessor)
-ray.init()
+ray.init(object_store_memory=150000000000, redis_max_memory=30000000000)
def train(config, reporter):
@@ -101,7 +101,7 @@ def train(config, reporter):
trainer_config["horizon"] = config['horizon']
trainer_config["num_workers"] = 0
- trainer_config["num_cpus_per_worker"] = 8
+ trainer_config["num_cpus_per_worker"] = 10
trainer_config["num_gpus"] = 0.5
trainer_config["num_gpus_per_worker"] = 0.5
trainer_config["num_cpus_for_driver"] = 2
@@ -155,7 +155,7 @@ def run_experiment(name, num_iterations, n_agents, hidden_sizes, save_every,
"seed": seed
},
resources_per_trial={
- "cpu": 10,
+ "cpu": 12,
"gpu": 0.5
},
local_dir=local_dir
--
GitLab