-
Notifications
You must be signed in to change notification settings - Fork 0
/
qbf_data.py
437 lines (363 loc) · 13.7 KB
/
qbf_data.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
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
from cnf_parser import *
# from aag_parser import read_qaiger
from utils import *
import numpy as np
from functools import partial
from torch.utils.data import Dataset
from enum import Enum
import time
import torch
import re
import collections
import os
import random
from IPython.core.debugger import Tracer
_use_shared_memory = False
MAX_VARIABLES = 1000000
MAX_CLAUSES = 5000000
# MAX_VARIABLES = 10000
# MAX_CLAUSES = 50000
GROUND_DIM = 8 # config.ground_dim duplicates this.
IDX_VAR_UNIVERSAL = 0
IDX_VAR_EXISTENTIAL = 1
# IDX_VAR_MISSING = 2
IDX_VAR_DETERMINIZED = 2
IDX_VAR_ACTIVITY = 3
IDX_VAR_POLARITY_POS = 4
IDX_VAR_POLARITY_NEG = 5
IDX_VAR_SET_POS = 6
IDX_VAR_SET_NEG = 7
# external utility function to filter small formulas
def filter_dir(dirname, bound):
a = [join(dirname, f) for f in listdir(dirname)]
rc = []
for fname in a:
if fname.endswith('qdimacs'):
with open(fname,'r') as f:
l = int(f.readline().split()[2])
if l <= bound: rc.append(fname)
return rc
class QbfBase(object):
def __init__(self, qcnf = None, **kwargs):
self.sparse = kwargs['sparse'] if 'sparse' in kwargs else True
self.qcnf = qcnf
self.sp_indices = None
self.sp_vals = None
self.extra_clauses = {}
self.removed_old_clauses = []
# self.get_base_embeddings = lru_cache(max_size=16)(self.get_base_embeddings)
if 'max_variables' in kwargs:
self._max_vars = kwargs['max_variables']
if 'max_clauses' in kwargs:
self._max_clauses = kwargs['max_clauses']
def reset(self):
self.sp_indices = None
self.sp_vals = None
self.extra_clauses = {}
def reload_qdimacs(self, fname):
try:
self.qcnf = qdimacs_to_cnf(fname)
except Exception as e:
print('Got exception in loading {}'.format(fname))
print(e)
self.reset()
@classmethod
def from_qdimacs(cls, fname, **kwargs):
try:
rc = qdimacs_to_cnf(fname)
if rc:
return cls(rc, **kwargs)
except:
print('Error parsing file %s' % fname)
@property
def num_vars(self):
return self.qcnf['maxvar']
@property
def num_existential(self):
a = self.var_types
return len(a[a>0])
@property
def num_universal(self):
return self.num_vars - self.num_existential
@property
def num_clauses(self):
k = self.extra_clauses.keys()
if not k:
return len(self.qcnf['clauses'])
return max(k)+1
@property
def max_vars(self):
try:
return self._max_vars
except:
return self.num_vars
@property
def max_clauses(self):
try:
return self._max_clauses
except:
return self.num_clauses
# This returns a 0-based numpy array of values per variable up to num_vars. 0 in universal, 1 is existential, 2 is missing
@property
def var_types(self):
a = self.qcnf['cvars']
rc = np.zeros(self.num_vars)+2 # Default var type for "missing" is 2
for k in a.keys():
if a[k]['universal']:
rc[k-1] = 0
else:
rc[k-1] = 1
return rc.astype(int)
def get_adj_matrices(self):
sample = self.qcnf
if self.sparse:
return self.get_sparse_adj_matrices()
else:
return self.get_dense_adj_matrices()
def get_sparse_adj_matrices(self):
sample = self.qcnf
if self.sp_indices is None:
clauses = sample['clauses']
indices = []
values = []
for i,c in enumerate(clauses):
if i in self.removed_old_clauses:
continue
for v in c:
val = np.sign(v)
v = abs(v)-1 # We read directly from file, which is 1-based, this makes it into 0-based
indices.append(np.array([i,v]))
values.append(val)
self.sp_indices = np.vstack(indices)
self.sp_vals = np.stack(values)
if not self.extra_clauses:
return self.sp_indices, self.sp_vals
indices = []
values = []
for i, c in self.extra_clauses.items():
for v in c:
val = np.sign(v)
v = abs(v)-1 # We read directly from file, which is 1-based, this makes it into 0-based
indices.append(np.array([i,v]))
values.append(val)
# Tracer()()
return np.concatenate([self.sp_indices,np.asarray(indices)]), np.concatenate([self.sp_vals, np.asarray(values)])
def get_dense_adj_matrices(self):
sample = self.qcnf
clauses = sample['clauses']
new_all_clauses = []
rc = np.zeros([self.max_clauses, self.max_vars])
for i in range(self.num_clauses):
for j in clauses[i]:
t = abs(j)-1
rc[i][t]=sign(j)
return rc
def get_base_embeddings(self):
embs = np.zeros([self.num_vars,GROUND_DIM])
for i in (IDX_VAR_UNIVERSAL, IDX_VAR_EXISTENTIAL):
embs[:,i][np.where(self.var_types==i)]=1
return embs
def get_clabels(self):
rc = np.ones(self.num_clauses)
rc[:len(self.qcnf['clauses'])]=0
return rc
def add_clause(self,clause, clause_id):
assert(clause_id not in self.extra_clauses.keys())
self.extra_clauses[clause_id]=clause
def remove_clause(self, clause_id):
if not (clause_id in self.extra_clauses.keys()):
self.removed_old_clauses.append(clause_id)
else:
del self.extra_clauses[clause_id]
@property
def label(self):
return 0 if 'UNSAT' in self.qcnf['fname'].upper() else 1
def as_tensor_dict(self):
rc = {'sparse': torch.Tensor([int(self.sparse)])}
if self.sparse:
rc_i, rc_v = self.get_sparse_adj_matrices()
sp_ind_pos = torch.from_numpy(rc_i[np.where(rc_v>0)])
sp_ind_neg = torch.from_numpy(rc_i[np.where(rc_v<0)])
sp_val_pos = torch.ones(len(sp_ind_pos))
sp_val_neg = torch.ones(len(sp_ind_neg))
rc['sp_v2c_pos'] = torch.sparse.FloatTensor(sp_ind_pos.t(),sp_val_pos,torch.Size([self.max_clauses,self.max_vars]))
rc['sp_v2c_neg'] = torch.sparse.FloatTensor(sp_ind_neg.t(),sp_val_neg,torch.Size([self.max_clauses,self.max_vars]))
rc['v2c'] = torch.from_numpy(self.get_dense_adj_matrices())
return rc
def as_np_dict(self):
rc = {}
rc_i, rc_v = self.get_sparse_adj_matrices()
# rc['sp_ind_pos'] = rc_i[np.where(rc_v>0)]
# rc['sp_ind_neg'] = rc_i[np.where(rc_v<0)]
rc['sp_indices'] = rc_i
rc['sp_vals'] = rc_v
rc['var_types'] = self.var_types
rc['num_vars'] = self.num_vars
rc['num_clauses'] = self.num_clauses
rc['ground'] = self.get_base_embeddings()
rc['clabels'] = self.get_clabels()
rc['label'] = self.label
return rc
f2qbf = lambda x: QbfBase.from_qdimacs(x)
class QbfDataset(Dataset):
def __init__(self, fnames=None, max_variables=MAX_VARIABLES, max_clauses=MAX_CLAUSES):
self.samples = ([], []) # UNSAT, SAT
self.max_vars = max_variables
self.max_clauses = max_clauses
if fnames:
if type(fnames) is list:
self.load_files(fnames)
else:
self.load_files([fnames])
def load_dir(self, directory):
self.load_files([join(directory, f) for f in listdir(directory)])
def load_files(self, files):
only_files = [x for x in files if os.path.isfile(x)]
only_dirs = [x for x in files if os.path.isdir(x)]
for x in only_dirs:
self.load_dir(x)
rc = map(f2qbf,only_files)
rc = [x for x in rc if x and x.num_vars <= self.max_vars and x.num_clauses < self.max_clauses\
and x.num_clauses > 0 and x.num_vars > 0]
for x in rc:
self.samples[x.label].append(x)
try:
del self.__weights_vector
except:
pass
return len(rc)
@property
def num_sat(self):
return len(self.samples[1])
@property
def num_unsat(self):
return len(self.samples[0])
@property
def weights_vector(self):
try:
return self.__weights_vector
except:
pass
rc = []
a =[[1/x]*x for x in [self.num_unsat, self.num_sat]]
a = np.concatenate(a) / 2
self.__weights_vector = a
return a
def load_file(self,fname):
if os.path.isdir(fname):
self.load_dir(fname)
else:
self.load_files([fname])
def get_files_list(self):
return [x.qcnf['fname'] for x in self.samples[0]] + [x.qcnf['fname'] for x in self.samples[1]]
def __len__(self):
return self.num_unsat + self.num_sat
def __getitem__(self, idx):
if idx < self.num_unsat:
return self.samples[0][idx].as_np_dict()
else:
return self.samples[1][idx-self.num_unsat].as_np_dict()
# Obselete, used only in qbf_train.py
# def qbf_collate(batch):
# rc = {}
# # Get max var/clauses for this batch
# v_size = max([b['num_vars'] for b in batch])
# c_size = max([b['num_clauses'] for b in batch])
# # adjacency matrix indices in one huge matrix
# rc_i = np.concatenate([b['sp_indices'] + np.asarray([i*c_size,i*v_size]) for i,b in enumerate(batch)], 0)
# rc_v = np.concatenate([b['sp_vals'] for b in batch], 0)
# # make var_types into ground embeddings
# all_embs = []
# for i,b in enumerate(batch):
# embs = b['ground']
# l = len(embs)
# embs = np.concatenate([embs,np.zeros([v_size-l,GROUND_DIM])])
# all_embs.append(embs)
# # break into pos/neg
# sp_ind_pos = torch.from_numpy(rc_i[np.where(rc_v>0)])
# sp_ind_neg = torch.from_numpy(rc_i[np.where(rc_v<0)])
# sp_val_pos = torch.ones(len(sp_ind_pos))
# sp_val_neg = torch.ones(len(sp_ind_neg))
# rc['sp_v2c_pos'] = torch.sparse.FloatTensor(sp_ind_pos.t(),sp_val_pos,torch.Size([c_size*len(batch),v_size*len(batch)]))
# rc['sp_v2c_neg'] = torch.sparse.FloatTensor(sp_ind_neg.t(),sp_val_neg,torch.Size([c_size*len(batch),v_size*len(batch)]))
# rc['ground'] = torch.from_numpy(np.stack(all_embs))
# rc['label'] = torch.Tensor([x['label'] for x in batch]).long()
# return rc
class AagBase(object):
"""
Wrapper object for an `aag` circuit (read from a `.qaiger` file).
"""
def __init__(self, aag = None, **kwargs):
self.sparse = kwargs['sparse'] if 'sparse' in kwargs else True
self.aag = aag
self.sp_indices = None
self.sp_vals = None
self.extra_clauses = {}
self.removed_old_clauses = []
# self.get_base_embeddings = lru_cache(max_size=16)(self.get_base_embeddings)
if 'max_variables' in kwargs:
self._max_vars = kwargs['max_variables']
if 'max_clauses' in kwargs:
self._max_clauses = kwargs['max_clauses']
def reset(self):
self.sp_indices = None
self.sp_vals = None
self.extra_clauses = {}
@property
def num_vars(self):
return self.aag['maxvar']
@property
def num_and_gates(self):
return self.aag['num_and_gates']
def load_qaiger(self, filename):
self.aag = read_qaiger(filename)
self.reset()
def get_adj_matrices(self):
sample = self.aag
if self.sparse:
return self.get_sparse_adj_matrices(sample)
else:
return self.get_dense_adj_matrices(sample)
def get_sparse_adj_matrix(self):
"""
NOTE: I SUBTRACTED 1 FROM EACH NODE NUMBER
"""
sample = self.aag
indices = []
values = []
n = self.num_vars
# for i, ag in enumerate(sample['and_gates']):
# for l in ag[1:]:
# val = 1 if l % 2 == 0 else -1
# indices.append( [int(ag[0]/2), int(l/2)] )
# values.append(val)
# return [indices, np.array(values)]
indices0 = []
indices1 = []
for i, ag in enumerate(sample['and_gates']):
for l in ag[1:]:
val = 1 if l % 2 == 0 else -1
# NOTE: I SUBTRACTED 1 FROM EACH NODE NUMBER
indices0.append(int(ag[0]/2) - 1)
indices1.append(int(l/2) - 1)
values.append(val)
i = torch.LongTensor([indices0, indices1])
v = torch.LongTensor(values)
return torch.sparse_coo_tensor(indices = i, values = v, size=[n,n])
# def get_base_embeddings(self):
# embs = np.zeros([self.num_vars,GROUND_DIM])
# for i in (IDX_VAR_UNIVERSAL, IDX_VAR_EXISTENTIAL):
# embs[:,i][np.where(self.var_types==i)]=1
# return embs
# def get_alabels(self):
# rc = np.ones(self.num_and_gates)
# rc[:self.num_and_gates]=0
# return rc
def get_geometric_data(self):
"""
torch_geometric.data Data() class to store aag graph.
https://pytorch-geometric.readthedocs.io/en/latest/modules/data.html
"""
edge_index = self.get_sparse_adj_matrix()
#num_edges =self.num_vars ** 2
#edge_attr = torch.zeros([num_edges, 1]) # all original clauses