File size: 2,290 Bytes
079c32c |
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 |
import torch
from copy import deepcopy
from dizoo.mujoco.config.hopper_td3_data_generation_config import main_config, create_config
from ding.entry import serial_pipeline_offline, collect_demo_data, eval, serial_pipeline
def train_td3_bc(args):
from dizoo.mujoco.config.hopper_td3_bc_config import main_config, create_config
main_config.exp_name = 'td3_bc'
main_config.policy.collect.data_path = './td3/expert_demos.hdf5'
main_config.policy.collect.data_type = 'hdf5'
config = deepcopy([main_config, create_config])
serial_pipeline_offline(config, seed=args.seed)
def eval_ckpt(args):
main_config.exp_name = 'td3'
main_config.policy.learn.learner.load_path = './td3/ckpt/ckpt_best.pth.tar'
main_config.policy.learn.learner.hook.load_ckpt_before_run = './td3/ckpt/ckpt_best.pth.tar'
state_dict = torch.load(main_config.policy.learn.learner.load_path, map_location='cpu')
config = deepcopy([main_config, create_config])
eval(config, seed=args.seed, load_path=main_config.policy.learn.learner.hook.load_ckpt_before_run)
# eval(config, seed=args.seed, state_dict=state_dict)
def generate(args):
main_config.exp_name = 'td3'
main_config.policy.learn.learner.load_path = './td3/ckpt/ckpt_best.pth.tar'
main_config.policy.collect.save_path = './td3/expert.pkl'
main_config.policy.collect.data_type = 'hdf5'
config = deepcopy([main_config, create_config])
state_dict = torch.load(main_config.policy.learn.learner.load_path, map_location='cpu')
collect_demo_data(
config,
collect_count=main_config.policy.other.replay_buffer.replay_buffer_size,
seed=args.seed,
expert_data_path=main_config.policy.collect.save_path,
state_dict=state_dict
)
def train_expert(args):
from dizoo.mujoco.config.hopper_td3_config import main_config, create_config
main_config.exp_name = 'td3'
config = deepcopy([main_config, create_config])
serial_pipeline(config, seed=args.seed, max_iterations=int(1e6))
if __name__ == "__main__":
import argparse
parser = argparse.ArgumentParser()
parser.add_argument('--seed', '-s', type=int, default=0)
args = parser.parse_args()
train_expert(args)
eval_ckpt(args)
generate(args)
train_td3_bc(args)
|