-
Notifications
You must be signed in to change notification settings - Fork 0
/
Formula.py
163 lines (122 loc) · 5.12 KB
/
Formula.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
from typing import Tuple
import torch
from prettytable import PrettyTable
import random
class Formula(torch.nn.Module):
def __init__(self, sub_formulas):
super().__init__()
if sub_formulas is not None:
self.sub_formulas = sub_formulas
self.predicates = list(set([p for sf in self.sub_formulas for p in sf.predicates]))
self.input_tensor = None
def function(self, truth_values):
pass
def boost_function(self, truth_values, delta):
pass
def get_name(self, parenthesis=False):
pass
def print_table(self): # TODO: fix
header = []
for sf in self.sub_formulas:
header.append(sf.get_name(parenthesis=True))
header.append(self.get_name())
pt = PrettyTable(header)
# TODO: Does not give any inputs
results = self.forward()
pt.add_rows(torch.concat([self.input_tensor, results], 1).numpy())
print(pt)
def __str__(self):
s = self.get_name() + '\n'
# s += str(self.input_tensor)
return s
def forward(self, truth_values):
inputs = []
for sf in self.sub_formulas:
inputs.append(sf.forward(truth_values))
if len(inputs) > 1:
self.input_tensor = torch.concat(inputs, 1)
else:
self.input_tensor = inputs[0]
return self.function(self.input_tensor)
def backward(self, delta, randomized=False):
deltas = self.boost_function(self.input_tensor, delta)
if randomized:
deltas = deltas * torch.rand(deltas.shape)
for sf, d in zip(self.sub_formulas, deltas.t()):
sf.backward(torch.unsqueeze(d, 0).t())
def get_delta_tensor(self, truth_values, method='mean'):
indices = []
deltas = []
for p in self.predicates:
i, d = p.aggregate_deltas(method)
p.reset_deltas()
indices.append(i)
deltas.append(d)
delta_tensor = torch.zeros_like(truth_values)
delta_tensor[..., indices] = torch.concat(deltas, 1).type(torch.float)
return delta_tensor
def reset_deltas(self):
for p in self.predicates:
p.reset_deltas()
def satisfaction(self, truth_values):
s = self.forward(truth_values)
self.reset_deltas()
return s
class Predicate(Formula):
def __init__(self, name, index):
super().__init__(None)
self.name = name
self.index = index
self.deltas = []
self.predicates = [self]
def forward(self, truth_values):
return torch.unsqueeze(truth_values[:, self.index], 1)
def backward(self, delta, randomized=False): # TODO: implement the usage of randomized
self.deltas.append(delta)
def reset_deltas(self):
self.deltas = []
def aggregate_deltas(self, method='mean') -> Tuple[int, torch.Tensor]:
if method == 'most_clauses':
deltas = torch.concat(self.deltas, 1)
positive = torch.sum(deltas > 0., 1, keepdim=True) - torch.sum(deltas <= 0., 1, keepdim=True) >= 0
max, _ = torch.max(deltas, 1, keepdim=True)
min, _ = torch.min(deltas, 1, keepdim=True)
return self.index, torch.where(positive, max, min)
if method == 'mean':
deltas = torch.concat(self.deltas, 1)
return self.index, torch.nan_to_num(
torch.sum(deltas, 1, keepdim=True) / torch.sum(deltas != 0.0, 1, keepdim=True))
if method == 'max':
deltas = torch.concat(self.deltas, 1)
abs_deltas = deltas.abs()
i = torch.argmax(abs_deltas, 1, keepdim=True)
return self.index, torch.gather(deltas, 1, i)
if method == 'min':
deltas = torch.concat(self.deltas, 1)
deltas = torch.where(deltas == 0, 100., deltas.double()).float()
abs_deltas = deltas.abs()
i = torch.argmin(abs_deltas, 1, keepdim=True)
final_deltas = torch.gather(deltas, 1, i)
return self.index, torch.where(final_deltas == 100., 0.0, final_deltas.double()).float()
if method == 'randomized_direction':
deltas = torch.concat(self.deltas, 1)
abs_deltas = deltas.abs()
if random.getrandbits(1):
deltas_no_zeros = torch.where(abs_deltas == 0, 100., abs_deltas.double()).float()
i = torch.argmin(deltas_no_zeros, 1, keepdim=True)
final_deltas = torch.gather(deltas, 1, i)
return self.index, torch.where(final_deltas == 100., 0.0, final_deltas.double()).float()
else:
i = torch.argmax(abs_deltas, 1, keepdim=True)
return self.index, torch.gather(deltas, 1, i)
def get_name(self, parenthesis=False):
return self.name
class NOT(Formula):
def __init__(self, sub_formula):
super().__init__([sub_formula])
def function(self, truth_values):
return 1 - truth_values
def boost_function(self, truth_values, delta):
return - delta
def get_name(self, parenthesis=False):
return 'NOT(' + self.sub_formulas[0].get_name() + ')'