-
Notifications
You must be signed in to change notification settings - Fork 2
/
env_wrapper.py
192 lines (165 loc) · 6.04 KB
/
env_wrapper.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
import numpy as np
import os
import torch
import torch.optim as optim
from torch.autograd import Variable
DEFAULT_QUALITY = 1
M_IN_K = 1000.0
BUFFER_NORM_FACTOR = 10.0
# DB_NORM_FACTOR = 100.0
VIDEO_BIT_RATE = [300, 750, 1200, 1850, 2850, 4300] # Kbps
S_INFO = 6
S_LEN = 8
A_DIM = 6
CHUNK_TIL_VIDEO_END_CAP = 48.0
M_IN_K = 1000.0
REBUF_PENALTY_LOG = 2.66 # 1 sec rebuffering -> 3 Mbps
REBUF_PENALTY_LIN = 4.3
SMOOTH_PENALTY = 1
DEFAULT_QUALITY = 1 # default video quality without agent
USE_CUDA = torch.cuda.is_available()
dtype = torch.cuda.FloatTensor if torch.cuda.is_available() else torch.FloatTensor
dlongtype = torch.cuda.LongTensor if torch.cuda.is_available() else torch.LongTensor
class VirtualPlayer:
def __init__(self, args, env, log_file):
self.env = env
self.args = args
self.task_list = env.task_list
## get the information of virtual players (personality)
# s_info, s_len, c_len, total_chunk_num, bitrate_versions, \
# quality_penalty, rebuffer_penalty, smooth_penalty_p, smooth_penalty_n \
# = env.get_env_info()
# Video information
self.s_info, self.s_len, self.total_chunk_num, self.quality_p, self.smooth_p = (
S_INFO,
S_LEN,
CHUNK_TIL_VIDEO_END_CAP,
1,
SMOOTH_PENALTY,
)
self.bitrate_versions = VIDEO_BIT_RATE
self.rebuff_p = REBUF_PENALTY_LIN if args.lin else REBUF_PENALTY_LOG
self.br_dim = len(self.bitrate_versions)
# QoE reward scaling
self.scaling_lb = -4 * self.rebuff_p
self.scaling_r = self.rebuff_p
# define the state for rl agent
self.state = np.zeros((self.s_info, self.s_len))
# information of emulating the video playing
self.last_bit_rate = DEFAULT_QUALITY
self.time_stamp = 0.0
self.end_flag = True
self.video_chunk_remain = self.total_chunk_num
# log files, recoding the video playing
self.log_file = log_file
# information of action mask
self.past_errors = []
self.past_bandwidth_ests = []
self.video_chunk_sizes = env.get_video_size()
def step(self, bit_rate):
# execute a step forward
(
delay,
sleep_time,
buffer_size,
rebuf,
video_chunk_size,
next_video_chunk_sizes,
end_of_video,
video_chunk_remain,
) = self.env.get_video_chunk(bit_rate)
# compute and record the reward of current chunk
self.time_stamp += delay # in ms
self.time_stamp += sleep_time # in ms
self.video_chunk_remain = video_chunk_remain
# compute reward of Quality of experience
if self.args.lin:
# -- lin scale reward --
reward = (
self.bitrate_versions[bit_rate] / M_IN_K
- self.rebuff_p * rebuf
- self.smooth_p
* np.abs(
self.bitrate_versions[bit_rate]
- self.bitrate_versions[self.last_bit_rate]
)
/ M_IN_K
)
else:
# -- log scale reward --
log_bit_rate = np.log(
self.bitrate_versions[bit_rate] / float(self.bitrate_versions[0])
)
log_last_bit_rate = np.log(
self.bitrate_versions[self.last_bit_rate]
/ float(self.bitrate_versions[0])
)
reward = (
log_bit_rate
- self.rebuff_p * rebuf
- self.smooth_p * np.abs(log_bit_rate - log_last_bit_rate)
)
rew_ = float(max(reward, self.scaling_lb) / self.scaling_r)
# reward_norm = self.reward_filter(rew_)
reward_norm = rew_
self.last_bit_rate = bit_rate
# -------------- logging -----------------
# log time_stamp, bit_rate, buffer_size, reward
self.log_file.write(
str(self.time_stamp)
+ "\t"
+ str(self.bitrate_versions[bit_rate])
+ "\t"
+
# str(np.sum(self.action_mask)) + '\t' +
str(buffer_size)
+ "\t"
+ str(rebuf)
+ "\t"
+ str(video_chunk_size)
+ "\t"
+ str(delay)
+ "\t"
+ str(reward)
+ "\n"
)
self.log_file.flush()
## dequeue history record
self.state = np.roll(self.state, -1, axis=1)
# this should be S_INFO number of terms
self.state[0, -1] = self.bitrate_versions[bit_rate] / float(
np.max(self.bitrate_versions)
) # last quality
self.state[1, -1] = float(buffer_size / BUFFER_NORM_FACTOR) # 10 sec
self.state[2, -1] = (
float(video_chunk_size) / float(delay) / M_IN_K
) # kilo byte / ms
self.state[3, -1] = float(delay) / M_IN_K / BUFFER_NORM_FACTOR # 10 sec
self.state[4, : self.br_dim] = (
np.array(next_video_chunk_sizes) / M_IN_K / M_IN_K
) # mega byte
self.state[5, -1] = np.minimum(
video_chunk_remain, self.total_chunk_num
) / float(self.total_chunk_num)
state_ = np.array([self.state])
state_ = torch.from_numpy(state_).type(dtype)
self.end_flag = end_of_video
if self.end_flag:
self.reset_play()
return state_, reward_norm, end_of_video
def set_task(self, idx):
self.env.set_task(idx)
def reset_play(self):
self.state = np.zeros((self.s_info, self.s_len))
self.last_bit_rate = DEFAULT_QUALITY
self.video_chunk_remain = self.total_chunk_num
self.time_stamp = 0.0
self.past_bandwidth_ests = []
self.past_errors = []
self.log_file.write("\n")
self.log_file.flush()
def clean_file_cache(self, file_name, max_file_size=4.096e7):
file_size = os.stat(file_name).st_size
if file_size > max_file_size:
self.log_file.seek(0)
self.log_file.truncate()