mirror of
https://github.com/gryf/coach.git
synced 2025-12-17 11:10:20 +01:00
7e7006305a
* Integrate coach.py params with distributed Coach. * Minor improvements - Use enums instead of constants. - Reduce code duplication. - Ask experiment name with timeout.
107 lines
3.3 KiB
Python
107 lines
3.3 KiB
Python
"""
|
|
this rollout worker:
|
|
|
|
- restores a model from disk
|
|
- evaluates a predefined number of episodes
|
|
- contributes them to a distributed memory
|
|
- exits
|
|
"""
|
|
|
|
import time
|
|
import os
|
|
import math
|
|
|
|
from rl_coach.base_parameters import TaskParameters, DistributedCoachSynchronizationType
|
|
from rl_coach.core_types import EnvironmentSteps, RunPhase
|
|
from google.protobuf import text_format
|
|
from tensorflow.python.training.checkpoint_state_pb2 import CheckpointState
|
|
|
|
|
|
def has_checkpoint(checkpoint_dir):
|
|
"""
|
|
True if a checkpoint is present in checkpoint_dir
|
|
"""
|
|
if os.path.isdir(checkpoint_dir):
|
|
if len(os.listdir(checkpoint_dir)) > 0:
|
|
return os.path.isfile(os.path.join(checkpoint_dir, "checkpoint"))
|
|
|
|
return False
|
|
|
|
|
|
def wait_for_checkpoint(checkpoint_dir, data_store=None, timeout=10):
|
|
"""
|
|
block until there is a checkpoint in checkpoint_dir
|
|
"""
|
|
for i in range(timeout):
|
|
if data_store:
|
|
data_store.load_from_store()
|
|
|
|
if has_checkpoint(checkpoint_dir):
|
|
return
|
|
time.sleep(10)
|
|
|
|
# one last time
|
|
if has_checkpoint(checkpoint_dir):
|
|
return
|
|
|
|
raise ValueError((
|
|
'Waited {timeout} seconds, but checkpoint never found in '
|
|
'{checkpoint_dir}'
|
|
).format(
|
|
timeout=timeout,
|
|
checkpoint_dir=checkpoint_dir,
|
|
))
|
|
|
|
|
|
def data_store_ckpt_load(data_store):
|
|
while True:
|
|
data_store.load_from_store()
|
|
time.sleep(10)
|
|
|
|
|
|
def get_latest_checkpoint(checkpoint_dir):
|
|
if os.path.exists(os.path.join(checkpoint_dir, 'checkpoint')):
|
|
ckpt = CheckpointState()
|
|
contents = open(os.path.join(checkpoint_dir, 'checkpoint'), 'r').read()
|
|
text_format.Merge(contents, ckpt)
|
|
rel_path = os.path.relpath(ckpt.model_checkpoint_path, checkpoint_dir)
|
|
return int(rel_path.split('_Step')[0])
|
|
|
|
|
|
def rollout_worker(graph_manager, checkpoint_dir, data_store, num_workers):
|
|
"""
|
|
wait for first checkpoint then perform rollouts using the model
|
|
"""
|
|
wait_for_checkpoint(checkpoint_dir)
|
|
|
|
task_parameters = TaskParameters()
|
|
task_parameters.__dict__['checkpoint_restore_dir'] = checkpoint_dir
|
|
time.sleep(30)
|
|
graph_manager.create_graph(task_parameters)
|
|
with graph_manager.phase_context(RunPhase.TRAIN):
|
|
error_compensation = 100
|
|
|
|
last_checkpoint = 0
|
|
|
|
act_steps = math.ceil((graph_manager.agent_params.algorithm.num_consecutive_playing_steps.num_steps + error_compensation)/num_workers)
|
|
|
|
for i in range(int(graph_manager.improve_steps.num_steps/act_steps)):
|
|
|
|
graph_manager.act(EnvironmentSteps(num_steps=act_steps))
|
|
|
|
new_checkpoint = get_latest_checkpoint(checkpoint_dir)
|
|
|
|
if graph_manager.agent_params.algorithm.distributed_coach_synchronization_type == DistributedCoachSynchronizationType.SYNC:
|
|
while new_checkpoint < last_checkpoint + 1:
|
|
if data_store:
|
|
data_store.load_from_store()
|
|
new_checkpoint = get_latest_checkpoint(checkpoint_dir)
|
|
|
|
graph_manager.restore_checkpoint()
|
|
|
|
if graph_manager.agent_params.algorithm.distributed_coach_synchronization_type == DistributedCoachSynchronizationType.ASYNC:
|
|
if new_checkpoint > last_checkpoint:
|
|
graph_manager.restore_checkpoint()
|
|
|
|
last_checkpoint = new_checkpoint
|