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
|
# CodeQ: an online programming tutor.
# Copyright (C) 2016,2017 UL FRI
#
# This program is free software: you can redistribute it and/or modify it under
# the terms of the GNU Affero General Public License as published by the Free
# Software Foundation, either version 3 of the License, or (at your option) any
# later version.
#
# This program is distributed in the hope that it will be useful, but WITHOUT
# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
# FOR A PARTICULAR PURPOSE. See the GNU Affero General Public License for more
# details.
#
# You should have received a copy of the GNU Affero General Public License
# along with this program. If not, see <http://www.gnu.org/licenses/>.
import collections
from itertools import chain, combinations, product
import pickle
import random
import sys
from nltk import ParentedTree, Tree
from prolog.util import parse as prolog_parse
# construct pattern to match the structure of nodes given by [include],
# supports variables and literals
def pattern(node, include):
if not isinstance(node, Tree):
return None
label = node.label()
if any(n is node for n in include):
if label == 'literal':
return '"{}"'.format(node[0].val)
if label == 'variable':
return '{}'.format(label)
return None
if label == 'functor':
return '({} "{}")'.format(label, node[0].val)
subpats = [pattern(child, include) for child in node]
pat = None
if any(subpats):
if label == 'and':
if subpats[1]:
pat = subpats[1]
if subpats[0]:
if pat:
pat = subpats[0] + ' ' + pat
else:
pat = subpats[0]
elif label == 'args':
pat = label
for i, subpat in enumerate(subpats):
if subpat:
pat += ' {}'.format(subpat)
pat = '(' + pat + ')'
elif label == 'unop':
pat = '(' + label + ' ' + node[0].val + ' ' + subpats[1] + ')'
elif label == 'binop':
pat = label
if subpats[0]:
pat += ' {}'.format(subpats[0])
pat += ' "{}"'.format(node[1].val)
if subpats[2]:
pat += ' {}'.format(subpats[2])
pat = '(' + pat + ')'
elif label == 'clause':
pat = label
for i, subpat in enumerate(subpats):
if subpat:
pat += ' {}'.format(subpats[i])
return '(' + pat + ')'
elif label == 'compound':
if len(subpats) > 1 and subpats[1]:
pat = label
for i, subpat in enumerate(subpats):
pat += ' {}'.format(subpat)
pat = '(' + pat + ')'
else:
return None
elif label == 'head':
pat = label
pat += ' {}'.format(subpats[0])
pat = '(' + pat + ')'
elif label == 'list':
pat = 'list '
if subpats[0]:
pat += '(h {})'.format(subpats[0])
if subpats[0] and subpats[1]:
pat += ' '
if subpats[1]:
pat += '(t {})'.format(subpats[1])
pat = '(' + pat + ')'
if not pat:
for s in subpats:
if s:
pat = s
break
return pat
def get_patterns(tree):
if isinstance(tree, str):
tree = prolog_parse(tree)
if tree is None:
return
tree = ParentedTree.convert(tree)
# get patterns separately for each clause
for clause in tree:
# collect variable nodes in this clause
variables = collections.defaultdict(list)
for node in clause.subtrees():
if isinstance(node, Tree) and node.label() == 'variable':
name = node[0].val
variables[name].append(node)
# yield patterns for singleton variables
for var, nodes in variables.items():
if len(nodes) == 1:
pat = pattern(clause, nodes)
if pat:
yield pat, nodes
# yield patterns for variable-variable pairs (within a clause)
for var, nodes in variables.items():
for selected in combinations(nodes, 2):
pat = pattern(clause, selected)
if pat:
yield pat, selected
# yield patterns for variable-literal / literal-literal pairs
# yield patterns for singleton literals
# (only within a topmost compound / binop / unop)
def patterns_with_literals(node):
if not isinstance(node, Tree):
return
if node.label() in {'compound', 'binop', 'unop'}:
vars = [n for n in node.subtrees() if n.label() == 'variable']
lits = [n for n in node.subtrees() if n.label() == 'literal']
for selected in chain(combinations(lits, 1),
combinations(lits, 2),
product(lits, vars)):
pat = pattern(clause, selected)
if pat:
yield pat, selected
else:
for child in node:
yield from patterns_with_literals(child)
yield from patterns_with_literals(clause)
# Extract edits and other data from existing traces for each problem.
# Run with: python3 -m monkey.patterns <problem ID> <predicate name> <solutions.pickle>
if __name__ == '__main__':
pid = int(sys.argv[1])
name = sys.argv[2]
submissions = pickle.load(open('pickle/programs-{}.pickle'.format(pid), 'rb'))
# find test/train users
users = sorted({user for code, info in submissions.items() for user in info['users']})
random.Random(0).shuffle(users)
split = int(len(users)*0.7)
learn_users = set(users[:split])
test_users = set(users[split:])
# save test users to file
with open('data/users-test-{}.txt'.format(pid), 'wt') as f:
for user in test_users:
print(user, file=f)
# find test/train programs
data = {
'train': [],
'test': []
}
for code, info in submissions.items():
if len(code) > 1000 or prolog_parse(code) is None:
continue
if name not in code:
continue
data['train'] += [(code, info['n_tests'] == info['n_passed'])] * len(info['users'] & learn_users)
data['test'] += [(code, info['n_tests'] == info['n_passed'])] * len(info['users'] & test_users)
# print info about test users and test/train programs
print('Test users:')
print(test_users)
print()
for which in ['train', 'test']:
print('Programs ({}):'.format(which))
print('correct: {} ({} unique)'.format(
len([code for code, correct in data[which] if correct]),
len({code for code, correct in data[which] if correct})))
print('incorrect: {} ({} unique)'.format(
len([code for code, correct in data[which] if not correct]),
len({code for code, correct in data[which] if not correct})))
print()
# extract attributes from training data
patterns = collections.Counter()
for code, correct in data['train']:
for pat, nodes in get_patterns(code):
patterns[pat] += 1
attrs = []
with open('data/attributes-{:03}.tab'.format(pid), 'w') as pattern_file:
for i, (pat, count) in enumerate(patterns.most_common()):
if count < 5:
break
attrs.append(pat)
print('a{}\t{}'.format(i, pat), file=pattern_file)
# check and write attributes for training/test data
for t in ['train', 'test']:
with open('data/programs-{:03}-{}.tab'.format(pid, t), 'w') as f:
# print header
print('\t'.join(['code', 'correct'] + ['a'+str(i) for i in range(len(attrs))]), file=f)
print('\t'.join(['d'] * (len(attrs)+2)), file=f)
print('meta\tclass', file=f)
# print rows (program, correct, attr1, attr2, …)
for code, correct in data[t]:
record = '{}\t{}'.format(repr(code), 'T' if correct else 'F')
code_pats = [pat for pat, nodes in get_patterns(code)]
for pat in attrs:
record += '\t{}'.format('T' if pat in code_pats else 'F')
print(record, file=f)
|
