-
Notifications
You must be signed in to change notification settings - Fork 0
/
model.py
305 lines (244 loc) · 10.9 KB
/
model.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
import torch
from torch.autograd import Variable
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
import utils
import numpy as np
from IPython.core.debugger import Tracer
import pdb
from settings import *
class TestDup(nn.Module):
def __init__(self, param1):
super(TestDup, self).__init__()
self.extra_embedding = nn.Embedding(1, param1, max_norm=1.)
self.layer1 = nn.Linear(param1*3,param1)
@property
def false(self):
return self.extra_embedding(Variable(torch.LongTensor([0]), requires_grad=False))
def forward(self, inputs):
return self.layer1(torch.cat([self.false,self.false,inputs],dim=1))
class TestAdd(nn.Module):
def __init__(self, param1):
super(TestAdd, self).__init__()
self.coefficient = nn.Parameter(torch.Tensor([param1]))
def forward(self, input):
return self.coefficient * input[0] + input[1]
class ResidualCombine(nn.Module):
def __init__(self, input_size, embedding_dim):
super(ResidualCombine, self).__init__()
self.layer1 = nn.Linear(input_size*embedding_dim,embedding_dim)
self.layer2 = nn.Linear(input_size*embedding_dim,embedding_dim)
def forward(self, input):
try:
out = utils.normalize(F.sigmoid(self.layer1(input)) + self.layer2(input))
except Exception as e:
print(e)
Tracer()()
return out
class SimpleCombinator(nn.Module):
def __init__(self, embedding_dim, hidden_dim=12):
super(SimpleCombinator, self).__init__()
self.layer1 = nn.Linear(embedding_dim,hidden_dim)
self.layer2 = nn.Linear(hidden_dim,embedding_dim)
def forward(self, inputs):
out = [self.layer2(F.sigmoid(self.layer1(x))) for x in inputs]
return torch.stack(out,dim=2).sum(dim=2).view(out[0].size())
class GroundCombinator(nn.Module):
def __init__(self, ground_dim, embedding_dim, hidden_dim=12):
super(GroundCombinator, self).__init__()
self.layer1 = nn.Linear(embedding_dim+ground_dim,hidden_dim)
self.layer2 = nn.Linear(hidden_dim,embedding_dim)
def forward(self, ground,state):
return self.layer2(F.sigmoid(self.layer1(torch.cat([ground,state],dim=1))))
class DummyGroundCombinator(nn.Module):
def __init__(self, *args, **kwargs):
super(DummyGroundCombinator, self).__init__()
def forward(self, ground,state):
return state
class SymmetricSumCombine(nn.Module):
def __init__(self, embedding_dim):
super(SymmetricSumCombine, self).__init__()
self.layer1 = nn.Linear(embedding_dim,embedding_dim)
self.layer2 = nn.Linear(embedding_dim,embedding_dim)
def forward(self, inputs):
out = [utils.normalize(F.sigmoid(self.layer1(x)) + self.layer2(x)) for x in inputs]
return torch.stack(out,dim=2).sum(dim=2).view(out[0].size())
# class VariableIteration(nn.Module):
# def __init__(self, embedding_dim, max_clauses, max_variables, num_ground_variables, max_iters):
class InnerIteration(nn.Module):
def __init__(self, get_ground_embeddings, embedding_dim, max_variables, num_ground_variables, split=True, permute=True, **kwargs):
super(InnerIteration, self).__init__()
self.settings = kwargs['settings'] if 'settings' in kwargs.keys() else CnfSettings()
self.comb_type = eval(self.settings['combinator_type'])
self.ground_comb_type = eval(self.settings['ground_combinator_type'])
self.get_ground_embeddings = get_ground_embeddings
self.embedding_dim = embedding_dim
self.max_variables = max_variables
self.split = split
self.permute = permute
self.num_ground_variables = num_ground_variables
self.negation = nn.Linear(embedding_dim, embedding_dim) # add non-linearity?
self.extra_embedding = nn.Embedding(1, embedding_dim, max_norm=1.)
self.clause_combiner = self.comb_type(embedding_dim)
self.variable_combiner = self.comb_type(embedding_dim)
self.ground_combiner = self.ground_comb_type(self.settings['ground_dim'],embedding_dim)
self.cuda = kwargs['cuda']
self.W_z = nn.Linear(self.embedding_dim,self.embedding_dim,bias=False)
self.U_z = nn.Linear(self.embedding_dim,self.embedding_dim,bias=self.settings['gru_bias'])
self.W_r = nn.Linear(self.embedding_dim,self.embedding_dim,bias=False)
self.U_r = nn.Linear(self.embedding_dim,self.embedding_dim,bias=self.settings['gru_bias'])
self.W = nn.Linear(self.embedding_dim,self.embedding_dim,bias=False)
self.U = nn.Linear(self.embedding_dim,self.embedding_dim,bias=self.settings['gru_bias'])
@property
def false(self):
return self.extra_embedding(Variable(self.settings.LongTensor([0]), requires_grad=False))
@property
def true(self):
return self.negation(self.false)
def prepare_clauses(self, clauses):
if self.permute and False:
rc = torch.cat(utils.permute_seq(clauses),dim=1)
if not self.split:
return rc
else:
org = torch.cat(clauses,1) # split
return torch.cat([org,rc])
else:
# return torch.cat(clauses,dim=1)
return clauses
# i is the index of the special variable (the current one)
def prepare_variables(self, variables, curr_variable):
tmp = variables.pop(curr_variable)
if self.permute and False:
rc = [tmp] + utils.permute_seq(variables)
try:
perm = torch.cat(rc,1)
except RuntimeError:
Tracer()()
if not self.split:
return perm
else:
org = torch.cat([tmp] + variables,1) # splitting batch
return torch.cat([org,perm])
else:
rc = [tmp] + variables
# return torch.cat(rc,1)
return rc
def _forward_clause(self, variables, clause, i):
c_vars = []
for j in range(self.max_variables):
if j<len(clause): # clause is a list of tensors
l=clause[j] # l is a tensored floaty integer
ind = torch.abs(l)-1 # variables in clauses are 1-based and negative if negated
v = torch.stack(variables)[ind.data][0] # tensored variables (to be indexed by tensor which is inside a torch variable..gah)
if (ind==i).data.all():
ind_in_clause = j
if (l < 0).data.all():
v = self.negation(v)
else:
continue
c_vars.append(v)
return self.variable_combiner(self.prepare_variables(c_vars,ind_in_clause))
def gru(self, av, prev_emb):
z = F.sigmoid(self.W_z(av) + self.U_z(prev_emb))
r = F.sigmoid(self.W_r(av) + self.U_r(prev_emb))
h_tilda = F.tanh(self.W(av) + self.U(r*prev_emb))
h = (1-z) * prev_emb + z*h_tilda
return h
def forward(self, variables, formula):
out_embeddings = []
for i,clauses in enumerate(formula):
# print('Clauses for variable %d: %d' % (i+1, len(clauses)))
if clauses:
clause_embeddings = [self._forward_clause(variables,c, i) for c in clauses]
new_var_embedding = self.ground_combiner(self.get_ground_embeddings(i),self.clause_combiner(self.prepare_clauses(clause_embeddings)))
out_embeddings.append(new_var_embedding)
else:
out_embeddings.append(variables[i])
new_vars = self.gru(torch.cat(out_embeddings,dim=0), torch.cat(variables,dim=0))
return torch.chunk(new_vars,len(new_vars))
class Encoder(nn.Module):
def __init__(self, embedding_dim, num_ground_variables, max_iters, **kwargs):
super(Encoder, self).__init__()
self.settings = kwargs['settings'] if 'settings' in kwargs.keys() else CnfSettings()
self.ground_dim = self.settings['ground_dim']
self.embedding_dim = embedding_dim
self.expand_dim_const = Variable(torch.zeros(1,self.embedding_dim - self.ground_dim))
self.max_iters = max_iters
self.num_ground_variables = num_ground_variables
self.embedding = nn.Embedding(num_ground_variables, self.ground_dim, max_norm=1.)
self.tseitin_embedding = nn.Embedding(1, self.ground_dim, max_norm=1.)
self.inner_iteration = InnerIteration(self.get_ground_embeddings, embedding_dim, num_ground_variables=num_ground_variables, **kwargs)
self.use_ground = self.settings['use_ground']
# input is one training sample (a formula), we'll permute it a bit at every iteration and possibly split to create a batch
def expand_ground_to_state(self,v):
return torch.cat([v,self.expand_dim_const],dim=1)
@property
def tseitin(self):
return self.tseitin_embedding(Variable(self.settings.LongTensor([0])))
def get_ground_embeddings(self,i):
if i<self.num_ground_variables and self.use_ground:
return self.embedding(Variable(self.settings.LongTensor([i])))
else:
return self.tseitin
def forward(self, input):
variables = []
for i in range(len(input)):
v = self.expand_ground_to_state(self.get_ground_embeddings(i))
variables.append(v)
for i in range(self.max_iters):
# print('Starting iteration %d' % i)
variables = self.inner_iteration(variables, input)
# We add loss on each variable embedding to encourage different elements in the batch to stay close.
aux_losses = Variable(torch.zeros(len(variables)))
return variables, aux_losses
class EqClassifier(nn.Module):
def __init__(self, num_classes, **kwargs):
super(EqClassifier, self).__init__()
self.num_classes = num_classes
self.settings = kwargs['settings'] if 'settings' in kwargs.keys() else CnfSettings()
self.encoder = Encoder(**kwargs)
self.softmax_layer = nn.Linear(self.encoder.embedding_dim,num_classes)
def forward(self, input, output_ind):
embeddings, aux_losses = self.encoder(input)
neg = (output_ind.data<0).all()
idx = torch.abs(output_ind).data[0]-1
out = embeddings[idx]
if neg:
out = self.encoder.inner_iteration.negation(out)
return self.softmax_layer(out), aux_losses # variables are 1-based
# return F.relu(self.softmax_layer(embeddings[output_ind.data[0]-1])), aux_losses # variables are 1-based
class GraphLevelClassifier(nn.Module):
def __init__(self, num_classes, **kwargs):
super(GraphLevelClassifier, self).__init__()
self.encoder = Encoder(**kwargs)
self.i_mat = nn.Linear(self.encoder.embedding_dim,self.encoder.embedding_dim)
self.j_mat = nn.Linear(self.encoder.embedding_dim,self.encoder.embedding_dim)
self.num_classes = num_classes
self.settings = kwargs['settings'] if 'settings' in kwargs.keys() else CnfSettings()
self.softmax_layer = nn.Linear(self.encoder.embedding_dim,num_classes)
def forward(self, input, output_ind):
embeddings, aux_losses = self.encoder(input)
out = F.tanh(sum([F.sigmoid(self.i_mat(x)) * F.tanh(self.j_mat(x)) for x in embeddings]))
return self.softmax_layer(out), aux_losses # variables are 1-based
# return F.relu(self.softmax_layer(embeddings[output_ind.data[0]-1])), aux_losses # variables are 1-based
class SiameseClassifier(nn.Module):
def __init__(self, **kwargs):
super(SiameseClassifier, self).__init__()
self.settings = kwargs['settings'] if 'settings' in kwargs.keys() else CnfSettings()
self.encoder = Encoder(**kwargs)
def forward(self, inputs, output_ind):
left, right = inputs
left_idx, right_idx = output_ind
l_embeddings, _ = self.encoder(left)
r_embeddings, _ = self.encoder(right)
embeddings = [l_embeddings, r_embeddings]
for i,x in zip([0,1],output_ind):
neg = (x.data<0).all()
idx = torch.abs(x).data[0]-1
out = embeddings[i][idx]
if neg:
out = self.encoder.inner_iteration.negation(out)
embs.append(out)
return tuple(embs)