forked from Facebear-ljx/PROTO
-
Notifications
You must be signed in to change notification settings - Fork 0
/
critic.py
226 lines (169 loc) · 8 KB
/
critic.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
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
from typing import Tuple
import jax.numpy as jnp
import jax
from functools import partial
from common import Batch, InfoDict, Model, Params, PRNGKey
from ensemble import subsample_ensemble
def gumbel_rescale_loss(diff, alpha, args=None):
""" Gumbel loss J: E[e^x - x - 1]. For stability to outliers, we scale the gradients with the max value over a batch
and optionally clip the exponent. This has the effect of training with an adaptive lr.
"""
z = diff/alpha
if args.max_clip is not None:
z = jnp.minimum(z, args.max_clip) # clip max value
max_z = jnp.max(z, axis=0)
max_z = jnp.where(max_z < -1.0, -1.0, max_z)
max_z = jax.lax.stop_gradient(max_z) # Detach the gradients
loss = jnp.exp(z - max_z) - z*jnp.exp(-max_z) - jnp.exp(-max_z) # scale by e^max_z
return loss
def gumbel_log_loss(diff, alpha=1.0):
""" Gumbel loss J: E[e^x - x - 1]. We can calculate the log of Gumbel loss for stability, i.e. Log(J + 1)
log_gumbel_loss: log((e^x - x - 1).mean() + 1)
"""
diff = diff
x = diff/alpha
grad = grad_gumbel(x, alpha)
# use analytic gradients to improve stability
loss = jax.lax.stop_gradient(grad) * x
return loss
def grad_gumbel(x, alpha, clip_max=7):
"""Calculate grads of log gumbel_loss: (e^x - 1)/[(e^x - x - 1).mean() + 1]
We add e^-a to both numerator and denominator to get: (e^(x-a) - e^(-a))/[(e^(x-a) - xe^(-a)).mean()]
"""
# clip inputs to grad in [-10, 10] to improve stability (gradient clipping)
x = jnp.minimum(x, clip_max) # jnp.clip(x, a_min=-10, a_max=10)
# calculate an offset `a` to prevent overflow issues
x_max = jnp.max(x, axis=0)
# choose `a` as max(x_max, -1) as its possible for x_max to be very small and we want the offset to be reasonable
x_max = jnp.where(x_max < -1, -1, x_max)
# keep track of original x
x_orig = x
# offsetted x
x1 = x - x_max
grad = (jnp.exp(x1) - jnp.exp(-x_max)) / \
(jnp.mean(jnp.exp(x1) - x_orig * jnp.exp(-x_max), axis=0, keepdims=True))
return grad
def expectile_loss(diff, expectile=0.8):
weight = jnp.where(diff > 0, expectile, (1 - expectile))
return weight * (diff**2)
def update_v(critic: Model, value: Model, batch: Batch,
expectile: float, loss_temp: float, double: bool, vanilla: bool, key: PRNGKey, args) -> Tuple[Model, InfoDict]:
actions = batch.actions
rng1, rng2 = jax.random.split(key)
obs = batch.observations
acts = batch.actions
qs = critic(obs, acts)
if double:
q = qs.min(axis=0)
else:
q = qs[0, :]
# zero_acts = jnp.zeros_like(batch.actions)
def value_loss_fn(value_params: Params) -> Tuple[jnp.ndarray, InfoDict]:
# detach the action part so the value network is a V function instead of a Q function
v = value.apply({'params': value_params}, obs)
if vanilla:
value_loss = expectile_loss(q - v, expectile).mean()
else:
if args.log_loss:
value_loss = gumbel_log_loss(q - v, alpha=loss_temp, args=args).mean()
else:
value_loss = gumbel_rescale_loss(q - v, alpha=loss_temp, args=args).mean()
if args.bc_pretrain:
value_loss = ((q - v) ** 2).mean()
return value_loss, {
'value_loss': value_loss,
'v': v.mean(),
}
new_value, info = value.apply_gradient(value_loss_fn)
return new_value, info
def update_q(critic: Model, target_value: Model, offline_actor: Model, behavior: Model, batch: Batch,
discount: float, double: bool, key: PRNGKey, loss_temp: float, offline: bool, args) -> Tuple[Model, InfoDict]:
# zero_acts = jnp.zeros_like(batch.actions)
next_v = target_value(batch.next_observations)
target_q = batch.rewards + discount * batch.masks * next_v
def critic_loss_fn(critic_params: Params) -> Tuple[jnp.ndarray, InfoDict]:
acts = batch.actions
qs = critic.apply({'params': critic_params}, batch.observations, acts)
v = target_value(batch.observations)
def mse_loss(q, q_target, *args):
loss_dict = {}
x = q-q_target
loss = x ** 2
loss_dict['critic_loss'] = loss.mean()
return loss.mean(), loss_dict
critic_loss = mse_loss
if double:
critic_loss, loss_dict = critic_loss(qs, target_q, v, loss_temp)
else:
critic_loss, loss_dict = critic_loss(qs[0, :], target_q, v, loss_temp)
loss_dict.update({
'q1': qs[0, :].mean(),
'q2': qs[1, :].mean()
})
return critic_loss, loss_dict
new_critic, info = critic.apply_gradient(critic_loss_fn)
return new_critic, info
def update_q_online(critic: Model, target_critic: Model, target_online_actor: Model, online_actor: Model, behavior: Model, offline_actor: Model, batch: Batch,
discount: float, double: bool, key: PRNGKey, temp: float, temp_online: float, log_alpha: float, args) -> Tuple[Model, InfoDict]:
rng1, rng2, rng3, rng4, rng5, rng = jax.random.split(key, num=6)
dist = online_actor(batch.next_observations, rngs={"dropout": rng1})
dist_k = target_online_actor(batch.next_observations, rngs={"dropout": rng2})
next_acts = dist.sample(seed=rng3)
if args.sac and args.entropy_backup:
value_reg = log_alpha() * dist.log_prob(next_acts) + temp_online * (dist.log_prob(next_acts) - dist_k.log_prob(next_acts))
else:
value_reg = 0.
if double:
target_qs = target_critic(batch.next_observations, next_acts, rngs={"dropout": rng4})
else:
# random sample a target q value to update (used for REDQ)
target_params = subsample_ensemble(rng5, target_critic.params, num_sample=1, num_qs=2)
target_qs = target_critic.apply({"params": target_params}, batch.next_observations, next_acts, rngs={"dropout": rng4})
target_q = batch.rewards + discount * batch.masks * (target_qs.min(axis=0) - value_reg)
def critic_loss_fn(critic_params: Params) -> Tuple[jnp.ndarray, InfoDict]:
acts = batch.actions
qs = critic.apply({'params': critic_params}, batch.observations, acts)
def mse_loss(q, q_target, *args):
loss_dict = {}
x = q-q_target
loss = x ** 2
loss_dict['critic_loss'] = loss.mean()
return loss.mean(), loss_dict
critic_loss = mse_loss
critic_loss, loss_dict = critic_loss(qs, target_q)
loss_dict.update({
'q1': qs[0, :].mean(),
'q2': qs[1, :].mean()
})
return critic_loss, loss_dict
new_critic, info = critic.apply_gradient(critic_loss_fn)
return new_critic, rng, info
def grad_norm(model, params, obs, action, lambda_=10):
@partial(jax.vmap, in_axes=(0, 0))
@partial(jax.jacrev, argnums=1)
def input_grad_fn(obs, action):
return model.apply({'params': params}, obs, action)
def grad_pen_fn(grad):
# We use gradient penalties inspired from WGAN-LP loss which penalizes grad_norm > 1
penalty = jnp.maximum(jnp.linalg.norm(grad1, axis=-1) - 1, 0)**2
return penalty
grad1, grad2 = input_grad_fn(obs, action)
return grad_pen_fn(grad1), grad_pen_fn(grad2)
def huber_loss(x, delta: float = 1.):
"""Huber loss, similar to L2 loss close to zero, L1 loss away from zero.
See "Robust Estimation of a Location Parameter" by Huber.
(https://projecteuclid.org/download/pdf_1/euclid.aoms/1177703732).
Args:
x: a vector of arbitrary shape.
delta: the bounds for the huber loss transformation, defaults at 1.
Note `grad(huber_loss(x))` is equivalent to `grad(0.5 * clip_gradient(x)**2)`.
Returns:
a vector of same shape of `x`.
"""
# 0.5 * x^2 if |x| <= d
# 0.5 * d^2 + d * (|x| - d) if |x| > d
abs_x = jnp.abs(x)
quadratic = jnp.minimum(abs_x, delta)
# Same as max(abs_x - delta, 0) but avoids potentially doubling gradient.
linear = abs_x - quadratic
return 0.5 * quadratic**2 + delta * linear