-
Notifications
You must be signed in to change notification settings - Fork 6
/
learner.py
169 lines (138 loc) · 6.4 KB
/
learner.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
"""Implementations of algorithms for continuous control."""
from typing import Optional, Sequence, Tuple
import jax
import jax.numpy as jnp
import numpy as np
import optax
import policy
import value_net
from common import Batch, InfoDict, Model, PRNGKey
from actor import update_actor
from critic import update_q, update_v
def target_update(critic: Model, target_critic: Model, tau: float) -> Model:
new_target_params = jax.tree_util.tree_map(
lambda p, tp: p * tau + tp * (1 - tau), critic.params,
target_critic.params)
return target_critic.replace(params=new_target_params)
@jax.jit
def _update_jit_sql(
rng: PRNGKey, actor: Model, critic: Model,
value: Model, target_critic: Model, batch: Batch, discount: float, tau: float,
alpha: float
) -> Tuple[PRNGKey, Model, Model, Model, Model, Model, InfoDict]:
new_value, value_info = update_v(target_critic, value, batch, alpha, alg='SQL')
key, rng = jax.random.split(rng)
new_actor, actor_info = update_actor(key, actor, target_critic,
new_value, batch, alpha, alg='SQL')
new_critic, critic_info = update_q(critic, new_value, batch, discount)
new_target_critic = target_update(new_critic, target_critic, tau)
return rng, new_actor, new_critic, new_value, new_target_critic, {
**critic_info,
**value_info,
**actor_info
}
@jax.jit
def _update_jit_eql(
rng: PRNGKey, actor: Model, critic: Model,
value: Model, target_critic: Model, batch: Batch, discount: float, tau: float,
alpha: float
) -> Tuple[PRNGKey, Model, Model, Model, Model, Model, InfoDict]:
new_value, value_info = update_v(target_critic, value, batch, alpha, alg='EQL')
key, rng = jax.random.split(rng)
new_actor, actor_info = update_actor(key, actor, target_critic,
new_value, batch, alpha, alg='EQL')
new_critic, critic_info = update_q(critic, new_value, batch, discount)
new_target_critic = target_update(new_critic, target_critic, tau)
return rng, new_actor, new_critic, new_value, new_target_critic, {
**critic_info,
**value_info,
**actor_info
}
class Learner(object):
def __init__(self,
seed: int,
observations: jnp.ndarray,
actions: jnp.ndarray,
actor_lr: float = 3e-4,
value_lr: float = 3e-4,
critic_lr: float = 3e-4,
hidden_dims: Sequence[int] = (256, 256),
discount: float = 0.99,
tau: float = 0.005,
alpha: float = 0.1,
dropout_rate: Optional[float] = None,
value_dropout_rate: Optional[float] = None,
layernorm: bool = False,
max_steps: Optional[int] = None,
max_clip: Optional[int] = None,
mix_dataset: Optional[str] = None,
alg: Optional[str] = None,
opt_decay_schedule: str = "cosine"):
"""
An implementation of the version of Soft-Actor-Critic described in https://arxiv.org/abs/1801.01290
"""
# self.expectile = expectile
self.tau = tau
self.discount = discount
self.alpha = alpha
self.max_clip = max_clip
self.alg = alg
rng = jax.random.PRNGKey(seed)
rng, actor_key, critic_key, value_key = jax.random.split(rng, 4)
action_dim = actions.shape[-1]
actor_def = policy.NormalTanhPolicy(hidden_dims,
action_dim,
log_std_scale=1e-3,
log_std_min=-5.0,
dropout_rate=dropout_rate,
state_dependent_std=False,
tanh_squash_distribution=False)
if opt_decay_schedule == "cosine":
schedule_fn = optax.cosine_decay_schedule(-actor_lr, max_steps)
optimiser = optax.chain(optax.scale_by_adam(),
optax.scale_by_schedule(schedule_fn))
else:
optimiser = optax.adam(learning_rate=actor_lr)
actor = Model.create(actor_def,
inputs=[actor_key, observations],
tx=optimiser)
critic_def = value_net.DoubleCritic(hidden_dims)
critic = Model.create(critic_def,
inputs=[critic_key, observations, actions],
tx=optax.adam(learning_rate=critic_lr))
value_def = value_net.ValueCritic(hidden_dims, layer_norm=layernorm, dropout_rate=value_dropout_rate)
value = Model.create(value_def,
inputs=[value_key, observations],
tx=optax.adam(learning_rate=value_lr))
target_critic = Model.create(
critic_def, inputs=[critic_key, observations, actions])
self.actor = actor
self.critic = critic
self.value = value
self.target_critic = target_critic
self.rng = rng
def sample_actions(self,
observations: np.ndarray,
temperature: float = 1.0) -> jnp.ndarray:
rng, actions = policy.sample_actions(self.rng, self.actor.apply_fn,
self.actor.params, observations,
temperature)
self.rng = rng
actions = np.asarray(actions)
return np.clip(actions, -1, 1)
def update(self, batch: Batch) -> InfoDict:
# type <class 'str'> is not a valid JAX type.
if self.alg == 'SQL':
new_rng, new_actor, new_critic, new_value, new_target_critic, info = _update_jit_sql(
self.rng, self.actor, self.critic, self.value, self.target_critic,
batch, self.discount, self.tau, self.alpha)
elif self.alg == 'EQL':
new_rng, new_actor, new_critic, new_value, new_target_critic, info = _update_jit_eql(
self.rng, self.actor, self.critic, self.value, self.target_critic,
batch, self.discount, self.tau, self.alpha)
self.rng = new_rng
self.actor = new_actor
self.critic = new_critic
self.value = new_value
self.target_critic = new_target_critic
return info