-
Notifications
You must be signed in to change notification settings - Fork 0
/
rl_example_2.py
190 lines (161 loc) · 7.04 KB
/
rl_example_2.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
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
#!/usr/bin/env python
# Copyright (c) CTU - All Rights Reserved
# Created on: 3/17/20
# Author: Vladimir Petrik <[email protected]>
#
# Trivial example to test rlpyt library with discrete actions.
import datetime
import os
from rlpyt.samplers.serial.sampler import SerialSampler
from rlpyt.runners.minibatch_rl import MinibatchRl
from rlpyt.spaces.float_box import FloatBox
from rlpyt.spaces.int_box import IntBox
from rlpyt.utils.logging.context import logger_context
from rlpyt.envs.base import *
from rlpyt.algos.pg.ppo import *
from rlpyt.agents.pg.gaussian import *
from rlpyt.agents.pg.categorical import *
from rlpyt.models.mlp import MlpModel
import matplotlib.pyplot as plt
from rlpyt.utils.tensor import infer_leading_dims, restore_leading_dims
import torch.nn.functional as F
plt.switch_backend('Qt5Agg')
plt.rcParams['axes.grid'] = True
plt.rcParams['legend.fancybox'] = True
horizon = 500
class MyEnv(Env):
def __init__(self) -> None:
super().__init__()
self.state = None
self.current_goal = None
self.iter = 0
self.action_discrete_mapping = [
np.array([0.0, 0.0]),
np.array([-0.1, 0.0]),
np.array([0.0, -0.1]),
np.array([0.1, 0.0]),
np.array([0.0, 0.1]),
]
self._action_space = IntBox(low=0, high=len(self.action_discrete_mapping))
self._observation_space = FloatBox(low=-1., high=1., shape=4) # current state and goal
self.goal_space = FloatBox(low=-1., high=1., shape=2)
def get_obs(self):
return np.concatenate([self.state, self.current_goal]).astype(np.float32)
def step(self, action):
self.iter += 1
self.state += self.action_discrete_mapping[action]
dist = np.linalg.norm(self.state - self.current_goal)
rew = np.exp(-0.5 * dist) / self.horizon
return EnvStep(self.get_obs(), rew, self.iter == self.horizon, EnvInfo())
def reset(self):
self.state = np.zeros(2)
self.current_goal = self.goal_space.sample()
self.iter = 0
return self.get_obs()
@property
def horizon(self):
return horizon
class MyModel(torch.nn.Module):
def __init__(self, observation_shape, action_size, hidden_sizes=None, hidden_nonlinearity=torch.nn.Tanh, ):
"""Instantiate neural net modules according to inputs."""
super().__init__()
self._obs_ndim = len(observation_shape)
input_size = int(np.prod(observation_shape))
hidden_sizes = hidden_sizes or [64, 64]
self.pi = MlpModel(input_size=input_size, hidden_sizes=hidden_sizes, output_size=action_size,
nonlinearity=hidden_nonlinearity, )
self.v = MlpModel(input_size=input_size, hidden_sizes=hidden_sizes, output_size=1,
nonlinearity=hidden_nonlinearity, )
def forward(self, observation, prev_action, prev_reward):
lead_dim, T, B, _ = infer_leading_dims(observation, self._obs_ndim)
obs_flat = observation.view(T * B, -1)
pi = F.softmax(self.pi(obs_flat), dim=-1)
v = self.v(obs_flat).squeeze(-1)
pi, v = restore_leading_dims((pi, v), lead_dim, T, B)
return pi, v
class MyAgent(CategoricalPgAgent):
def __init__(self, greedy_eval, ModelCls=MyModel, **kwargs):
super().__init__(ModelCls=ModelCls, **kwargs)
self.greedy_eval = greedy_eval
def make_env_to_model_kwargs(self, env_spaces):
return dict(
observation_shape=env_spaces.observation.shape, action_size=env_spaces.action.n,
)
def step(self, observation, prev_action, prev_reward):
action, agent_info = super().step(observation, prev_action, prev_reward)
if self._mode == "eval" and self.greedy_eval:
action = torch.argmax(agent_info.dist_info.prob, dim=-1)
return action, agent_info
def build_and_train(run_id=0, greedy_eval=False):
sampler = SerialSampler(
EnvCls=MyEnv,
env_kwargs=dict(),
eval_env_kwargs=dict(),
batch_T=horizon,
batch_B=64,
max_decorrelation_steps=0,
eval_n_envs=64,
eval_max_steps=int(1e6),
eval_max_trajectories=64,
)
runner = MinibatchRl(
algo=PPO(entropy_loss_coeff=0., learning_rate=3e-4),
agent=MyAgent(greedy_eval),
sampler=sampler,
n_steps=int(400 * horizon * 64),
log_interval_steps=int(10 * horizon * 64),
)
log_params = dict()
log_dir = "data/rl_example_2/{}".format(datetime.datetime.today().strftime("%Y%m%d_%H%M"))
with logger_context(log_dir, run_id, 'Reacher2D', log_params=log_params, snapshot_mode="last",
use_summary_writer=True, override_prefix=True):
runner.train()
def build_and_test(run_id=0, test_date=None, greedy_eval=False):
log_dir = "data/rl_example_2/"
if test_date is None:
exps = os.listdir(log_dir)
exps.sort(reverse=True)
test_date = exps[0]
print('Using the latest experiment with timestamp: {}'.format(test_date))
params_path = '{}{}/run_{}/params.pkl'.format(log_dir, test_date, run_id)
data = torch.load(params_path)
agent = MyAgent(greedy_eval=greedy_eval, initial_model_state_dict=data['agent_state_dict'], )
env = MyEnv()
agent.initialize(env.spaces)
agent.eval_mode(0)
while True:
observations = []
obs = env.reset()
action = env.action_space.sample()
rew = 0.
for _ in range(horizon):
observations.append(obs)
action, action_info = agent.step(torch.from_numpy(obs).float(),
torch.from_numpy(action).float(),
torch.tensor(rew).float())
action = action.numpy()
# if use_mode:
# action = action_info.dist_info.mean.numpy() # use mean of the distribution
obs, rew, _, _ = env.step(action=action)
observations = np.stack(observations, axis=0)
fig, axes = plt.subplots(1, 1, squeeze=False)
ax = axes[0, 0]
ax.plot(observations[0, 2], observations[0, 3], 'o', label='Goal', color='tab:green')
ax.plot(observations[0, 0], observations[0, 1], 'x', label='Start', color='tab:blue')
ax.plot(observations[:, 0], observations[:, 1], '--k', label='Path')
plt.legend()
ax.set_xlim(-1, 1)
ax.set_ylim(-1, 1)
plt.show()
if __name__ == "__main__":
import argparse
parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument('--run_id', help='run identifier (logging)', type=int, default=0)
parser.add_argument('--test', dest='test', action='store_true')
parser.add_argument('--test_date', type=str, default=None)
parser.add_argument('--greedy_eval', dest='greedy_eval', action='store_true')
args = parser.parse_args()
if args.test:
build_and_test(run_id=args.run_id, test_date=args.test_date, greedy_eval=args.greedy_eval)
else:
build_and_train(run_id=args.run_id, greedy_eval=args.greedy_eval)