gryf/coach
1
0
Fork 0
mirror of https://github.com/gryf/coach.git synced 2025-12-18 11:40:18 +01:00
Code Issues Projects Releases Wiki Activity

add redis options to training worker

Browse Source
This commit is contained in:
Zach Dwiel
2018-09-19 11:13:20 -04:00
committed by zach dwiel
parent 04f32a0f02
commit 13d81f65b9
2 changed files with 15 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
rl_coach/graph_managers/graph_manager.py
View File

@@ -511,7 +511,7 @@ class GraphManager(object):
): ):
self.save_checkpoint() self.save_checkpoint()
def _log_save_checkpoint(self): def save_checkpoint(self):
checkpoint_path = os.path.join(self.task_parameters.save_checkpoint_dir, checkpoint_path = os.path.join(self.task_parameters.save_checkpoint_dir,
"{}_Step-{}.ckpt".format( "{}_Step-{}.ckpt".format(
self.checkpoint_id, self.checkpoint_id,
@@ -521,6 +521,9 @@ class GraphManager(object):
else: else:
saved_checkpoint_path = checkpoint_path saved_checkpoint_path = checkpoint_path
# this is required in order for agents to save additional information like a DND for example
[manager.save_checkpoint(self.checkpoint_id) for manager in self.level_managers]
screen.log_dict( screen.log_dict(
OrderedDict([ OrderedDict([
("Saving in path", saved_checkpoint_path), ("Saving in path", saved_checkpoint_path),
@@ -528,12 +531,6 @@ class GraphManager(object):
prefix="Checkpoint" prefix="Checkpoint"
) )
def save_checkpoint(self):
# this is required in order for agents to save additional information like a DND for example
[manager.save_checkpoint(self.checkpoint_id) for manager in self.level_managers]
self._log_save_checkpoint()
self.checkpoint_id += 1 self.checkpoint_id += 1
self.last_checkpoint_saving_time = time.time() self.last_checkpoint_saving_time = time.time()

16
rl_coach/training_worker.py
View File

@@ -29,11 +29,6 @@ def training_worker(graph_manager, checkpoint_dir):
# save randomly initialized graph # save randomly initialized graph
graph_manager.save_checkpoint() graph_manager.save_checkpoint()
# TODO: Q: training steps passed into graph_manager.train ignored?
# TODO: specify training steps between checkpoints (in preset?)
# TODO: replace outer training loop with something general
# TODO: low priority: move evaluate out of this process
heatup(graph_manager) heatup(graph_manager)
# training loop # training loop
@@ -57,10 +52,21 @@ def main():
help='(string) Path to a folder containing a checkpoint to write the model to.', help='(string) Path to a folder containing a checkpoint to write the model to.',
type=str, type=str,
default='/checkpoint') default='/checkpoint')
parser.add_argument('-r', '--redis_ip',
help="(string) IP or host for the redis server",
default='localhost',
type=str)
parser.add_argument('-rp', '--redis_port',
help="(int) Port of the redis server",
default=6379,
type=int)
args = parser.parse_args() args = parser.parse_args()
graph_manager = short_dynamic_import(expand_preset(args.preset), ignore_module_case=True) graph_manager = short_dynamic_import(expand_preset(args.preset), ignore_module_case=True)
graph_manager.agent_params.memory.redis_ip = args.redis_ip
graph_manager.agent_params.memory.redis_port = args.redis_port
training_worker( training_worker(
graph_manager=graph_manager, graph_manager=graph_manager,
checkpoint_dir=args.checkpoint_dir, checkpoint_dir=args.checkpoint_dir,