-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathpermutationgraph.py
243 lines (207 loc) · 8.16 KB
/
permutationgraph.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
import numpy as np
def get_edges(args):
edges = np.zeros([len(args), len(args)])
for i in range(len(args)):
for j in range(i+1, len(args)):
if args[i] > args[j]:
edges[args[j]][args[i]] = 1
return edges
class QueryLearner():
def __init__(self, objective_ins, sorted_docs, intra) -> None:
self.sorted_docs = sorted_docs
self.g = intra
self.objective = objective_ins
self.n = len(self.sorted_docs)
def _swap_if(self, docs, i, visited, changed):
# print([i, docs, docs[i:i+2]])
d1, d2 = sorted(docs[i:i+2])
if self.g[self.sorted_docs[d1]] != self.g[self.sorted_docs[d2]] and visited[d1][d2] == 0:
if self.verbose:
print([i, d1, d2, self.probs_mat[d1][d2], visited[d1][d2]])
visited[d1][d2] = 1
if np.random.binomial(1, self.probs_mat[d1][d2]):
if self.verbose:
print('x')
tmp = docs[i]
docs[i] = docs[i+1]
docs[i+1] = tmp
changed.append(i)
return True
return False
def _permute(self):
docs = np.arange(self.n)
visited = np.zeros_like(self.probs_mat)
if self.verbose:
print('-------------')
changed = list(range(self.n))
inversion_cnt = 0
while changed:
candid_set = set()
for i in changed:
if i - 1 >= 0:
candid_set.add(i-1)
if i + 1 < len(docs) - 1:
candid_set.add(i+1)
# print(candid_set)
candid_set = np.random.permutation(list(candid_set))
# print(candid_set)
changed = []
skip = set()
for i in candid_set:
if i in skip:
continue
if self._swap_if(docs, i, visited, changed):
skip.add(i+1)
skip.add(i-1)
# pass
inversion_cnt += len(changed)
# break
return docs, inversion_cnt
def fit(self, epochs, lr, verbose=False):
self.verbose = verbose
val = self.objective.eval(self.sorted_docs)
vals = [val]
reordered = True
if verbose:
print([val, self.sorted_docs])
min_val = val
min_val_docs = self.sorted_docs
vals_list = [val]
for epoch in range(epochs):
if reordered:
self.probs_mat = (0.5**(1./2.)) * np.ones([self.n, self.n])
reordered = False
docs, cnt = self._permute()
# print(cnt)
if cnt == 0:
continue
new_val = self.objective.eval(self.sorted_docs[docs])
vals_list.append(new_val)
if new_val < min_val:
min_val = new_val
min_val_docs = self.sorted_docs[docs]
# if new_val < min_val:
if new_val < val:
# self.sorted_docs = self.sorted_docs[docs]
# reordered = True
# min_val = new_val
diff = max((new_val - val) / np.array(vals_list).mean(), -1.)
edges = get_edges(docs)
self.probs_mat -= (edges) * diff * lr
self.probs_mat[self.probs_mat < 0] = 0.01
self.probs_mat[self.probs_mat > 1] = 0.99
# elif new_val > min_val:
elif new_val > val:
# diff = min((new_val - min_val) / np.array(vals_list).mean(), 1.)
diff = min((new_val - val) / np.array(vals_list).mean(), 1.)
edges = get_edges(docs)
self.probs_mat -= (edges) * diff * lr
self.probs_mat[self.probs_mat < 0] = 0.01
self.probs_mat[self.probs_mat > 1] = 0.99
if verbose:
if min_val == new_val:
print([new_val, self.sorted_docs] + self.objective.get_info(self.sorted_docs))
else:
print([min_val, '<', new_val, self.sorted_docs[docs]] + self.objective.get_info(self.sorted_docs[docs]))
print(self.probs_mat)
# print(edges)
# probs_mat -= (edges) * diff * lr
vals.append(new_val)
val = new_val
self.sorted_docs = min_val_docs
return vals
class BatchLearner():
def __init__(self, objective_ins, sorted_docs, intra, inter) -> None:
self.sorted_docs = sorted_docs
self.g = intra
self.batch_numbers = inter
self.objective = objective_ins
self.n = len(sorted_docs)
def _swap_if(self, docs, i, visited, changed):
# print([i, docs, docs[i:i+2]])
d1, d2 = sorted(docs[i:i+2])
if self.g[self.batch_numbers[d1] + self.sorted_docs[d1]] != self.g[self.batch_numbers[d2] + self.sorted_docs[d2]] and \
self.batch_numbers[d1] == self.batch_numbers[d2] and \
visited[d1][d2] == 0:
visited[d1][d2] = 1
if np.random.binomial(1, self.probs_mat[d1][d2]):
tmp = docs[i]
docs[i] = docs[i+1]
docs[i+1] = tmp
changed.append(i)
return True
return False
def _permute(self):
docs = np.arange(self.n)
visited = np.zeros_like(self.probs_mat)
changed = list(range(self.n))
inversion_cnt = 0
while changed:
candid_set = set()
for i in changed:
if i - 1 >= 0:
candid_set.add(i-1)
if i + 1 < len(docs) - 1:
candid_set.add(i+1)
# print(candid_set)
candid_set = np.random.permutation(list(candid_set))
# print(candid_set)
changed = []
skip = set()
for i in candid_set:
if i in skip:
continue
if self._swap_if(docs, i, visited, changed):
skip.add(i+1)
skip.add(i-1)
# pass
inversion_cnt += len(changed)
return docs, inversion_cnt
def fit(self, epochs, lr, verbose=False):
val = self.objective.eval(self.sorted_docs)
vals = [val]
reordered = True
if verbose:
print(self.sorted_docs)
print(self.batch_numbers)
print([val, self.sorted_docs])
min_val = val
for epoch in range(epochs):
if reordered:
self.probs_mat = 0.5 * np.ones([self.n, self.n])
reordered = False
docs, cnt = self._permute()
# print(cnt)
if cnt == 0:
continue
new_val = self.objective.eval(self.sorted_docs[docs])
if new_val < min_val:
self.sorted_docs = self.sorted_docs[docs]
reordered = True
min_val = new_val
else:
diff = new_val - min_val
edges = get_edges(docs)
self.probs_mat -= (edges) * diff * lr
self.probs_mat[self.probs_mat < 0] = 0.01
self.probs_mat[self.probs_mat > 1] = 0.99
if verbose:
if min_val == new_val:
print([new_val, self.sorted_docs] + self.objective.get_info(self.sorted_docs))
else:
print([min_val, '<', new_val, self.sorted_docs[docs]] + self.objective.get_info(self.sorted_docs[docs]))
# print(probs_mat)
# print(edges)
vals.append(new_val)
val = new_val
return vals
def get_group_counts(g, dlr):
groups = np.unique(g)
gcnt = [[] for _ in range(len(groups))]
for qid in range(dlr.shape[0] - 1):
s, e = dlr[qid:qid+2]
for i, group in enumerate(groups):
gcnt[i].append(len(np.where(g[s:e] == group)[0]))
for i, group in enumerate(groups):
gcnt[i] = np.array(gcnt[i])
return groups, gcnt