wassname/ray
1
0
Fork 0
mirror of https://github.com/wassname/ray.git synced 2026-09-09 11:32:43 +08:00
Code Issues Packages Projects Releases Wiki Activity

[rllib] Switch to tune.run() instead of run_experiments() (#4515)

Browse Source
This commit is contained in:
Eric Liang
2019-03-30 14:07:50 -07:00
committed by GitHub
parent 5efb21e1d0
commit fce0062380
13 changed files with 174 additions and 208 deletions
Download Patch File Download Diff File Expand all files Collapse all files
doc/source/rllib-training.rst
+36 -44
View File
@@ -145,21 +145,19 @@ All RLlib agents are compatible with the `Tune API <tune-usage.html>`__. This en
.. code-block:: python
import ray
import ray.tune as tune
from ray import tune
ray.init()
tune.run_experiments({
"my_experiment": {
"run": "PPO",
tune.run(
"PPO",
stop={"episode_reward_mean": 200},
config={
"env": "CartPole-v0",
"stop": {"episode_reward_mean": 200},
"config": {
"num_gpus": 0,
"num_workers": 1,
"lr": tune.grid_search([0.01, 0.001, 0.0001]),
},
"num_gpus": 0,
"num_workers": 1,
"lr": tune.grid_search([0.01, 0.001, 0.0001]),
},
})
)
Tune will schedule the trials to run in parallel on your Ray cluster:
@@ -258,20 +256,18 @@ You can provide callback functions to be called at points during policy evaluati
info["agent"].__name__, info["result"]["episodes_this_iter"]))
ray.init()
trials = tune.run_experiments({
"test": {
trials = tune.run(
"PG",
config={
"env": "CartPole-v0",
"run": "PG",
"config": {
"callbacks": {
"on_episode_start": tune.function(on_episode_start),
"on_episode_step": tune.function(on_episode_step),
"on_episode_end": tune.function(on_episode_end),
"on_train_result": tune.function(on_train_result),
},
"callbacks": {
"on_episode_start": tune.function(on_episode_start),
"on_episode_step": tune.function(on_episode_step),
"on_episode_end": tune.function(on_episode_end),
"on_train_result": tune.function(on_train_result),
},
}
})
},
)
Custom metrics can be accessed and visualized like any other training result:
@@ -306,20 +302,18 @@ Approach 1: Use the Agent API and update the environment between calls to ``trai
lambda env: env.set_phase(phase)))
ray.init()
tune.run_experiments({
"curriculum": {
"run": train,
"config": {
"num_gpus": 0,
"num_workers": 2,
},
"resources_per_trial": {
"cpu": 1,
"gpu": lambda spec: spec.config.num_gpus,
"extra_cpu": lambda spec: spec.config.num_workers,
},
tune.run(
train,
config={
"num_gpus": 0,
"num_workers": 2,
},
})
resources_per_trial={
"cpu": 1,
"gpu": lambda spec: spec.config.num_gpus,
"extra_cpu": lambda spec: spec.config.num_workers,
},
)
Approach 2: Use the callbacks API to update the environment on new training results:
@@ -342,17 +336,15 @@ Approach 2: Use the callbacks API to update the environment on new training resu
lambda env: env.set_phase(phase)))
ray.init()
tune.run_experiments({
"curriculum": {
"run": "PPO",
tune.run(
"PPO",
config={
"env": YourEnv,
"config": {
"callbacks": {
"on_train_result": tune.function(on_train_result),
},
"callbacks": {
"on_train_result": tune.function(on_train_result),
},
},
})
)
Debugging
---------