train_cmd.py

import os
import json
import glob
from pyfiglet import Figlet
from termcolor import colored
def get_latest_run_id(log_path: str, env_name: str) -> int:
    """
    Returns the latest run number for the given log name and log path,
    by finding the greatest number in the directories.

    :param log_path: path to log folder
    :param env_name:
    :return: latest run number
    """
    max_run_id = 0
    for path in glob.glob(os.path.join(log_path, env_name + "_[0-9]*")):
        run_id = path.split("_")[-1]
        path_without_run_id = path[: -len(run_id) - 1]
        if path_without_run_id.endswith(env_name) and run_id.isdigit() and int(run_id) > max_run_id:
            max_run_id = int(run_id)
    return max_run_id

envid = "BounceBall_Pybullet_env-v0"
algo="sac"





logpath = os.path.join(os.path.dirname(__file__), "logs/%s"%algo)
run_id = get_latest_run_id(logpath, envid) + 1 
projname = "BounceBallEnv"
runname = os.getlogin()+"_"+algo+"_"+str(run_id)

### 打印信息
f=Figlet(font='starwars',width=150)
f2=Figlet(font='slant',width=150)
logo="PeiTian Tech"
print (f.renderText("\n"))
print (colored(f.renderText(logo),"light_green"))
print (f.renderText("\n"))
print (colored(f2.renderText(envid),"light_green"))
print (f.renderText("\n"))
f3=Figlet(font='standard',width=110)
print(colored(f3.renderText(str("Model ID "+runname) ),"light_yellow"))

hyper_opt=False #Flag 超参数优化
nEvalEnv=16  #训练中，测试环境的个数，一般与并行环境数相等
### 训练时需要调整的参数
pretrain = False  # 是否需要用预训练的模型
pretrain_model_id = 51
pretrain_best_model_id = "best_model"   ###选择加载哪个bestmodel- 20230512脚本默认改为训练只有一个最佳model
pretrain_best_model_flag = False     ###True则加载最佳模型，False加载最后模型
if pretrain_best_model_flag:
    model_file = "%s.zip"%pretrain_best_model_id
else:
    model_file = "%s.zip"%envid
train_step=5e6
modelName_prefix=runname
envconfig = {
    "00_modelname": "air4a", # "air4a", "air7l_b"
    "000_initBallMethod": "toss1", # "fall", "toss1", "random","toss2","toss1back"
    "act_opt": 2,
    "01_frame_skip": 40,
    "02_robotobs_timelag": 40,
    "03_ballobs_timelag": 114,
    "04_act_timelag": 50,  
    "05_obs_stack_n": 30,
    "07_rw_coeff_sparse": 1,
    "08_rw_coeff_paddlez": 1,
    "09_rw_coeff_ballvel": 3.5,
    "10_rw_coeff_acc": 0.01,
    "11_rw_coeff_dist": 0.01,
    "12_rw_coeff_baseballdis": 0.005,
    "rand_ballpos": 0.05,
    "rand_force_on_ball": 0.001,
    "rand_bounceforce_on_ball": 0.01,
    "13_rand_obs_ballpos": 0.015,
    "14_rand_obs_ballvel": 0.1,
    "q_alpha": 0.01,
    "q_vel_alpha": 0.01,
    "act_alpha": 1,
    "act_acc_alpha": 1,
    "modelName_prefix":str(modelName_prefix),
    "15_action_noise":0.00025
}

### End 训练时需要调整的参数


pretrain_model_file = os.path.join(os.path.dirname(__file__), "logs/%s"%algo, "%s_%d"%(envid, pretrain_model_id), model_file)
envconfig_str="config:\"%s\""%envconfig

 
if hyper_opt:
    print(">>> Hyperparameters tuning")
    cmd='python train.py --algo %s --env %s -n 50000 -optimize --wandb-project-name %s --wandb-run-name %s --env-kwargs %s --n-trials 100 --n-jobs 1 --sampler tpe --pruner median --n-evaluations 10000 --n-eval-envs %d  --verbose 0 --vec-env subproc -P'%(algo, envid, projname, runname, envconfig_str,nEvalEnv)
else:
    if pretrain:
        print(">>> use pretrained model %s"%pretrain_model_file)
        if not os.path.exists(pretrain_model_file):
            raise IOError("Model %s does not exist"%pretrain_model_file)
        cmd = 'python train.py --algo %s --env %s -i %s --save-model-n 1 --wandb-project-name %s --wandb-run-name %s --env-kwargs %s --n-timesteps %d --vec-env subproc --n-eval-envs %d --eval-freq 70000 -P'%(algo, envid, pretrain_model_file, projname, runname, envconfig_str,train_step,nEvalEnv)
    else:
        cmd = 'python train.py --algo %s --env %s --save-model-n 1 --wandb-project-name %s --wandb-run-name %s --env-kwargs %s --n-timesteps %d --vec-env subproc --n-eval-envs %d --eval-freq 70000 -P'%(algo, envid, projname, runname, envconfig_str,train_step,nEvalEnv)

print(">>> Train command:")
print(cmd)

os.system(cmd)