-
Notifications
You must be signed in to change notification settings - Fork 0
/
learning_train.py
148 lines (130 loc) · 7.15 KB
/
learning_train.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
import os
import copy
import importlib
import logging
import pickle
import numpy as np
import sys
sys.path.append("..")
from learning_advisor.learning_infer import get_eval_env
from learning_advisor.experiment import Experiment
from learning_advisor.gym_db.common import EnvironmentType
from learning_advisor.learning_utils import swirl_com
from learning_advisor.learning_utils.configuration_parser import ConfigurationParser
from learning_advisor.learning_utils.swirl_com import set_logger
from learning_advisor.learning_utils.workload_generator import WorkloadGenerator
from learning_advisor.learning_utils.workload import Query, Workload
if __name__ == "__main__":
parser = swirl_com.get_parser()
args = parser.parse_args()
logging.info(f"The training mode is `{args.train_mode}`.")
# load the configuration, create the experiment folder.
experiment = Experiment(args)
logging.info(f"The value of `parallel_environments` is `{experiment.exp_config['parallel_environments']}`.")
# only stable_baselines2 supported.
if experiment.exp_config["rl_algorithm"]["stable_baselines_version"] == 2:
from stable_baselines.common.callbacks import EvalCallbackWithTBRunningAverage
from stable_baselines.common.vec_env import DummyVecEnv, SubprocVecEnv, VecNormalize
# <class 'stable_baselines.ppo2.ppo2.PPO2'>
algorithm_class = getattr(
importlib.import_module("learning_advisor.stable_baselines"),
experiment.exp_config["rl_algorithm"]["algorithm"])
else:
raise ValueError
experiment.prepare()
ParallelEnv = SubprocVecEnv if experiment.exp_config["parallel_environments"] > 1 else DummyVecEnv
training_env = ParallelEnv([experiment.make_env(env_id,
environment_type=EnvironmentType.TRAINING,
workloads_in=None,
db_config=experiment.schema.db_config)
for env_id in range(experiment.exp_config["parallel_environments"])])
training_env = VecNormalize(training_env, norm_obs=True, norm_reward=True,
gamma=experiment.exp_config["rl_algorithm"]["gamma"], training=True)
experiment.model_type = algorithm_class
with open(f"{experiment.experiment_folder_path}/experiment_object.pickle", "wb") as handle:
pickle.dump(experiment, handle, protocol=pickle.HIGHEST_PROTOCOL)
model = algorithm_class(policy=experiment.exp_config["rl_algorithm"]["policy"], # MlpPolicy by default
env=training_env,
verbose=2,
seed=experiment.exp_config["random_seed"],
gamma=experiment.exp_config["rl_algorithm"]["gamma"],
tensorboard_log=args.logdir.format(args.exp_id), # "tensor_log",
policy_kwargs=copy.copy(
experiment.exp_config["rl_algorithm"]["model_architecture"]
), # This is necessary because SB modifies the passed dict.
**experiment.exp_config["rl_algorithm"]["args"])
logging.warning(
f"Creating model with NN architecture(value/policy): {experiment.exp_config['rl_algorithm']['model_architecture']}")
experiment.set_model(model)
# get the performance of heuristic algorithms.
logging.disable(logging.INFO)
experiment.compare()
logging.disable(logging.DEBUG)
callback_test_env = VecNormalize(
DummyVecEnv([experiment.make_env(0,
environment_type=EnvironmentType.TESTING,
workloads_in=None,
db_config=experiment.schema.db_config)]),
norm_obs=True,
norm_reward=False,
gamma=experiment.exp_config["rl_algorithm"]["gamma"],
training=False)
test_callback = EvalCallbackWithTBRunningAverage(
n_eval_episodes=experiment.exp_config["workload"]["validation_testing"]["number_of_workloads"],
eval_freq=round(experiment.exp_config["validation_frequency"] / experiment.exp_config["parallel_environments"]),
eval_env=callback_test_env,
verbose=1,
name="test",
deterministic=True,
comparison_performances=experiment.comparison_performances["test"])
callback_validation_env = VecNormalize(
DummyVecEnv([experiment.make_env(0, environment_type=EnvironmentType.VALIDATION,
workloads_in=None,
db_config=experiment.schema.db_config)]),
norm_obs=True,
norm_reward=False,
gamma=experiment.exp_config["rl_algorithm"]["gamma"],
training=False)
validation_callback = EvalCallbackWithTBRunningAverage(
n_eval_episodes=experiment.exp_config["workload"]["validation_testing"]["number_of_workloads"],
eval_freq=round(experiment.exp_config["validation_frequency"] / experiment.exp_config["parallel_environments"]),
eval_env=callback_validation_env,
best_model_save_path=experiment.experiment_folder_path,
verbose=1,
name="validation",
deterministic=True,
comparison_performances=experiment.comparison_performances["validation"])
callbacks = [validation_callback, test_callback]
if len(experiment.multi_validation_wl) > 0:
callback_multi_validation_env = VecNormalize(
DummyVecEnv([experiment.make_env(0, EnvironmentType.VALIDATION,
experiment.multi_validation_wl,
db_config=experiment.schema.db_config)]),
norm_obs=True,
norm_reward=False,
gamma=experiment.exp_config["rl_algorithm"]["gamma"],
training=False,
)
multi_validation_callback = EvalCallbackWithTBRunningAverage(
n_eval_episodes=len(experiment.multi_validation_wl),
eval_freq=round(
experiment.exp_config["validation_frequency"] / experiment.exp_config["parallel_environments"]),
eval_env=callback_multi_validation_env,
best_model_save_path=experiment.experiment_folder_path,
verbose=1,
name="multi_validation",
deterministic=True,
comparison_performances={},
)
callbacks.append(multi_validation_callback)
# set the `training_start_time`.
experiment.record_learning_start_time()
model.learn(total_timesteps=args.timesteps,
callback=callbacks,
tb_log_name=experiment.id) # the name of the run for tensorboard log
experiment.finish_learning_save_model(model.get_env(), # training_env,
validation_callback.moving_average_step * experiment.exp_config[
"parallel_environments"],
validation_callback.best_model_step * experiment.exp_config[
"parallel_environments"])
experiment.finish_evaluation()