Nitro-VLDB25/eval_boost_efficiency.py at master · IntelliSys-Lab/Nitro-VLDB25 · GitHub

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
import numpy as np
import copy
import torch
import time
import csv
import collections
import glob
import logging
import ray
from env import Environment
import config
import utils
import numpy as np


def eval_boost_efficiency(
    scheduler_name,
    is_serverless,
    algo_name,
    env_name,
    ckpt_path,
    json_path,
    num_rollout_workers,
    num_envs_per_worker,
):
    # Set up environment and load checkpoint
    env = Environment(
        scheduler_name=scheduler_name,
        algo_name=algo_name,
        env_name=env_name,
        target_reward=config.envs[env_name]["max_reward"],
        budget=config.envs[env_name]["budget"],
        stop_min_round=config.stop_min_round,
        stop_max_round=config.stop_max_round,
        stop_num_results=config.stop_num_results,
        stop_cv=config.stop_cv,
        stop_grace_period=config.stop_grace_period,
        is_serverless=is_serverless,
    )

    # Start training
    state, mask, info = env.reset()
    env.load(ckpt_path)
    boost_round_id = utils.json_load(json_path)['round_id']

    round_id = boost_round_id

    action = {}
    action["num_envs_per_worker"] = num_envs_per_worker
    action["num_rollout_workers"] = num_rollout_workers

    next_state, next_mask, reward, done, info = env.step(
        round_id=round_id,
        action=action
    )

    env.trainer.stop()

    return info["episode_reward"]


if __name__ == '__main__':
    ckpt_scheduler_name = "serverful_baseline"
    # is_serverless = False
    is_serverless = True

    print("")
    print("**********")
    print("**********")
    print("**********")
    print("")
    ray.init(
        log_to_driver=False,
        configure_logging=True,
        logging_level=logging.ERROR
    )

    for algo_name in config.serverful_algos:
        for env_name in config.envs.keys():
            # for ckpt_filename in glob.glob("{}/{}~{}~{}~*/".format(config.ckpt_path, ckpt_scheduler_name, env_name, algo_name)):
            #     eval_round_id = ckpt_filename.split('~')[-1].replace('/', '')
            for eval_round_id in config.envs[env_name]["eval_boost_efficiency"]:
                scheduler_name = "eval_boost_efficiency_{}".format(eval_round_id)
                ckpt_path = "{}/{}~{}~{}~{}".format(config.ckpt_path, ckpt_scheduler_name, env_name, algo_name, eval_round_id)
                json_path = "{}/{}~{}~{}~{}.json".format(config.ckpt_path, ckpt_scheduler_name, env_name, algo_name, eval_round_id)

                csv_efficiency = [
                    [
                        "num_envs_per_worker",
                        "reward",
                    ]
                ]

                for [num_rollout_workers, num_envs_per_worker] in config.Nitro_boost_candidates:
                    for eval_time in range(config.Nitro_boost_eval_time):
                        episode_reward = eval_boost_efficiency(
                            scheduler_name=scheduler_name,
                            is_serverless=is_serverless,
                            algo_name=algo_name,
                            env_name=env_name,
                            ckpt_path=ckpt_path,
                            json_path=json_path,
                            num_rollout_workers=num_rollout_workers,
                            num_envs_per_worker=num_envs_per_worker,
                        )

                        for reward in episode_reward:
                            csv_efficiency.append(
                                [
                                    int(num_rollout_workers / config.num_rollout_workers_serverful * num_envs_per_worker),
                                    reward,
                                ]
                            )

                    print("")
                    print("******************")
                    print("******************")
                    print("******************")
                    print("")
                    print("Running {}, algo {}, env {}".format(scheduler_name, algo_name, env_name))
                    print("round_id: {}".format(eval_round_id))
                    print("eval_reward_mean: {}".format(np.mean(episode_reward)))

                utils.export_csv(
                    scheduler_name=scheduler_name,
                    env_name=env_name,
                    algo_name=algo_name,
                    csv_name="",
                    csv_file=csv_efficiency
                )

    ray.shutdown()
    print("")
    print("**********")
    print("**********")
    print("**********")