-
Notifications
You must be signed in to change notification settings - Fork 1
/
Sampler.py
250 lines (216 loc) · 8.89 KB
/
Sampler.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
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
import random
from Utils.JobReader import skill_lst
from math import log
def get_act_pool(state, n_a, relational_lst, pool_size):
cnt = [0] * n_a
for i in range(n_a):
if state[i] == 1:
for u in relational_lst[i]:
cnt[u] += 1
skill_id = list(range(n_a))
skill_id.sort(key=lambda x: cnt[x] + 1e-5 * x, reverse=True)
ret = []
for u in range(n_a):
s = skill_id[u]
if state[s] == 0:
ret.append(s)
if len(ret) == pool_size:
break
return ret
class DistributionPoolSampler(object):
def __init__(self, p, n_a, pool_size, relational_lst):
self.n_a = n_a
self.pool_size = pool_size
self.relational_lst = relational_lst
self.p = p
def judge(self, p, prob, x):
return prob < p[x]
def binary_search(self, p, prob, l, r):
if r - l <= 1:
if self.judge(p, prob, l): return l
return r
m = (l + r) // 2
if self.judge(p, prob, m): return self.binary_search(p, prob, l, m)
return self.binary_search(p, prob, m + 1, r)
def sample(self, state):
pool = get_act_pool(state, self.n_a, self.relational_lst, self.pool_size)
p = [self.p[u] for u in pool]
for i in range(len(p) - 1):
p[i + 1] += p[i]
s = pool[self.binary_search(p, random.random() * p[-1], 0, self.pool_size - 1)]
while state[s] != 0:
s = pool[self.binary_search(p, random.random() * p[-1], 0, self.pool_size - 1)]
return s, None
class EpsilonGreedySampler(object):
def __init__(self, Qa, epsilon, n_a):
self.epsilon = epsilon
self.n_a = n_a
self.Qa = Qa
def sample(self, state, retQ=False):
if random.random() > self.epsilon:
s_ret = random.randint(0, self.n_a - 1)
while state[s_ret] != 0:
s_ret = random.randint(0, self.n_a - 1)
if retQ:
q_ret = self.Qa.estimate_single(state, s_ret)
else:
q_ret = None
else:
q_ret, s_ret = self.Qa.estimate_maxq_action(state)
return s_ret, q_ret
def sample_batch(self, state, retQ=False):
q_ret, s_ret = self.Qa.estimate_maxq_batch(state)
q_ret_batch, s_ret_batch = [], []
for state_now, q, s in zip(state, q_ret, s_ret):
if random.random() > self.epsilon:
s = random.randint(0, self.n_a - 1)
while state_now[s] != 0:
s = random.randint(0, self.n_a - 1)
if retQ:
q = self.Qa.estimate_single(state_now, s)
else:
q = None
q_ret_batch.append(q)
s_ret_batch.append(s)
return s_ret_batch, q_ret_batch
class EpsilonGreedyPoolSampler(object):
def __init__(self, relational_lst, Qa, epsilon, n_a, pool_size):
self.epsilon = epsilon
self.n_a = n_a
self.Qa = Qa
self.relational_lst = relational_lst
self.pool_size = pool_size
def judge(self, p, prob, x):
return prob < p[x]
def binary_search(self, p, prob, l, r):
if r - l <= 1:
if self.judge(p, prob, l): return l
return r
m = (l + r) // 2
if self.judge(p, prob, m): return self.binary_search(p, prob, l, m)
return self.binary_search(p, prob, m + 1, r)
def sample(self, state, pool=None, retQ=False):
if pool is None:
pool = get_act_pool(state, self.n_a, self.relational_lst, self.pool_size)
if random.random() > self.epsilon:
s_ret = pool[random.randint(0, len(pool) - 1)]
while state[s_ret] != 0:
s_ret = pool[random.randint(0, len(pool) - 1)]
if retQ:
q_ret = self.Qa.estimate_single(state, s_ret)
else:
q_ret = None
else:
q_ret, s_ret = self.Qa.estimate_maxq_action(state, pool)
return s_ret, q_ret
class BestStrategyPoolSampler(object):
def __init__(self, relational_lst, Qa, n_a, pool_size):
self.n_a = n_a
self.Qa = Qa
self.relational_lst = relational_lst
self.pool_size = pool_size
def sample(self, state, pool=None):
if pool is None:
pool = get_act_pool(state, self.n_a, self.relational_lst, self.pool_size)
q_ret, s_ret = self.Qa.estimate_maxq_action(state, pool)
return s_ret, q_ret
class BestStrategySampler(object):
def __init__(self, Qa, n_a):
self.n_a = n_a
self.Qa = Qa
def sample(self, state):
q_ret, s_ret = self.Qa.estimate_maxq_action(state)
return s_ret, q_ret
def sample_batch(self, state):
q_ret, s_ret = self.Qa.estimate_maxq_batch(state)
return s_ret, q_ret
class DistributionSampler(object):
def __init__(self, p, n_a):
self.n_a = n_a
self.p = p
for i in range(len(p) - 1):
self.p[i + 1] += self.p[i]
def judge(self, prob, x):
return prob < self.p[x]
def binary_search(self, prob, l, r):
if r - l <= 1:
if self.judge(prob, l): return l
return r
m = (l + r) // 2
if self.judge(prob, m): return self.binary_search(prob, l, m)
return self.binary_search(prob, m + 1, r)
def sample(self, state):
s = self.binary_search(random.random(), 0, self.n_a - 1)
while state[s] != 0:
s = self.binary_search(random.random(), 0, self.n_a - 1)
return s, None
class GreedySampler(object):
def __init__(self, relational_lst, environment, n_a, pool_size, rtype="salary"):
self.n_a = n_a
self.environment = environment
self.jm = environment.job_matcher
self.de = environment.d_estimator
self.relational_lst = relational_lst
self.pool_size = pool_size
self.type = rtype
def sample(self, state):
pool = get_act_pool(state, self.n_a, self.relational_lst, self.pool_size)
if self.type == 'salary':
r_lst = [self.jm.predict_salary(s) for s in pool]
elif self.type == 'easy':
r_lst = [self.de.predict_easy(s) for s in pool]
else:
r_lst = [self.environment.get_reward(salary=self.jm.predict_salary(s), easy=self.de.predict_easy(s)) for s in pool]
s_ret, r_ret = -1, -1000
for s, rnow in zip(pool, r_lst):
if rnow > r_ret:
s_ret = s
r_ret = rnow
return s_ret, r_ret
class GreedyUnionSampler(object):
def __init__(self, relational_lst, environment, Qa, n_a, pool_size, beta):
self.n_a = n_a
self.environment = environment
self.jm = environment.job_matcher
self.de = environment.d_estimator
self.relational_lst = relational_lst
self.pool_size = pool_size
self.Qa = Qa
self.beta = beta
def sample(self, state):
pool = get_act_pool(state, self.n_a, self.relational_lst, self.pool_size)
r_lst = [self.environment.get_reward(salary=self.jm.predict_salary(s), easy=self.de.predict_easy(s)) for s in pool]
s_ret, r_ret = -1, -1000
state_lst = []
pool_lst = []
for s in pool:
state_now = state.copy()
state_now[s] = 1
state_lst.append(state_now)
pool_lst.append(get_act_pool(state_now, self.n_a, self.relational_lst, self.pool_size))
q_ret, _ = self.Qa.estimate_maxq_batch(state_lst, pool_lst)
salary_ret, easy_ret = q_ret
# for s, r, sal, ease in zip(pool, r_lst, salary_ret, easy_ret):
# print(s, r, sal, ease)
q_ret = [rnow + (1 - self.beta) * self.environment.get_reward(salary=sal, easy=easy) for rnow, sal, easy in zip(r_lst, salary_ret, easy_ret)]
for s, rnow in zip(pool, q_ret):
if rnow > r_ret:
s_ret = s
r_ret = rnow
# print(s_ret)
return s_ret, r_ret
def sample_pre(self, state):
pool = get_act_pool(state, self.n_a, self.relational_lst, self.pool_size)
r_lst = [self.environment.get_reward(salary=self.jm.predict_salary(s), easy=self.de.predict_easy(s)) for s in pool]
s_ret, r_ret = -1, -1000
for s, rnow in zip(pool, r_lst):
state[s] = 1
pool_nxt = get_act_pool(state, self.n_a, self.relational_lst, self.pool_size)
q_ret, _ = self.Qa.estimate_maxq_action(state, pool_nxt)
q_ret = self.environment.get_reward(salary=q_ret[0], easy=q_ret[1])
rnow = rnow + (1-self.beta) * q_ret
state[s] = 0
if rnow > r_ret:
s_ret = s
r_ret = rnow
return s_ret, r_ret