/
rule_representation_data.py
208 lines (182 loc) · 7.15 KB
/
rule_representation_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
import os
import numpy as np
import torch
import pickle
from torch.utils.data import Dataset
from transformers import AutoModel
from transformers import GPT2TokenizerFast, AutoTokenizer
from transformers import pipeline
from utils import *
from data_utils import RuleDatasetUtils
class RuleReprDataset(Dataset):
def __init__(self, input_data_dir, emb_model_type, tokenizer):
# get all relevant files (in raw form)
files = []
oracles = {}
hole_datas = {}
parse_datas = {}
all_duplicate_files = []
data_type = input_data_dir.split('/')[-1]
for dp, dn, filenames in os.walk(input_data_dir):
for f in filenames:
if f == 'hole_data':
hole_data = pickle.load(open(os.path.join(dp, f), 'rb'))
hole_data = self.update_dict(hole_data, data_type)
hole_datas = {**hole_datas, **hole_data}
if f == 'parsed_data':
parse_data = pickle.load(open(os.path.join(dp, f), 'rb'))
parse_data = self.update_dict(parse_data, data_type)
parse_datas = {**parse_datas, **parse_data}
if f == 'duplicates':
duplicate_files = open(os.path.join(dp, f), 'r').readlines()
all_duplicate_files.extend([x.strip() for x in duplicate_files])
if os.path.splitext(f)[1] == '.java':
files.append(os.path.join(dp, f))
print(len(all_duplicate_files))
self.holes = []
for file in files:
if file in hole_datas and \
file not in all_duplicate_files and \
not file.startswith('rule_classifier_data/train/rsbotownversion/trunk/scripts/'):
for (l,c) in hole_datas[file]:
hole_identity = file + '_' + str(l) + '_' + str(c)
self.holes.append(hole_identity)
print(len(self.holes))
self.num_rules = len(combined_to_index)
self.tokenizer = tokenizer
self.parse_datas = parse_datas
self.model_max_length = self.tokenizer.model_max_length
self.rule_repr_cache = {}
self.emb_model_type = emb_model_type
self.set_embedding_model()
self.repr_size = 768
self.start = 0
self.end = 500000
def update_dict(self, dic, data_type):
mod_dic = {}
for k,v in dic.items():
mod_k = '/'. join(['rule_classifier_data', data_type] + k.split('/')[2:])
mod_dic[mod_k] = v
return mod_dic
def __len__(self):
return len(self.holes)
def __getitem__(self, idx):
if idx >=self.start and idx <= self.end:
return self.generate_data(self.holes[idx])
else:
return None, None, None
def get_start_index(self, repo, start_offset=0, interval=0):
count=0
for i in range(len(self.holes)):
hole = self.holes[i]
repo_name = hole.split('/')[2]
if repo_name == repo:
count+=1
repo_end_idx = i
self.start = repo_end_idx - count + 1
self.start = self.start + start_offset
if interval!=0 :
self.end = self.start + interval
else:
self.end = repo_end_idx
return self.start, self.end
def is_clear_cache(self):
if len(self.rule_repr_cache) < 30:
self.clear_cache = False
else:
self.clear_cache = True
self.rule_repr_cache = {}
def get_representation(self, inputs, mask):
outputs = self.emb_model(inputs, attention_mask=mask)
try:
representation = outputs.pooler_output
except:
representation = outputs.last_hidden_state[:, 0]
#print(representation.shape)
return representation
def get_context_embedding(self, context, attn_mask):
context_embedding = self.get_representation(context, attn_mask)
return context_embedding
def get_rule_context(self, file, hole_pos):
self.is_clear_cache()
rule_dataset_util = RuleDatasetUtils(file, self.parse_datas, hole_pos, self.tokenizer)
rule_prompts, rule_indexes = rule_dataset_util.get_all_rules_context()
rule_contexts = self.tokenizer(rule_prompts, truncation=True, padding='max_length')
rule_inputs = torch.tensor(rule_contexts['input_ids'])
rule_masks = torch.tensor(rule_contexts['attention_mask'])
rule_indexes = torch.tensor(rule_indexes)
# remove rules that are already cached
rule_prompts = self.tokenizer.batch_decode(rule_inputs)
filtered_rule_context = []
filtered_rule_mask = []
filtered_rule_prompts = []
filtered_rule_indexes = []
for i in range(len(rule_prompts)):
rule_prompt = rule_prompts[i]
if rule_prompt not in self.rule_repr_cache:
filtered_rule_indexes.append(rule_indexes[i])
filtered_rule_context.append(rule_inputs[i])
filtered_rule_mask.append(rule_masks[i])
filtered_rule_prompts.append(rule_prompt)
if filtered_rule_context:
filtered_rule_context = torch.stack(filtered_rule_context)
filtered_rule_mask = torch.stack(filtered_rule_mask)
# get rule representations
filtered_representations = self.get_context_embedding(filtered_rule_context, filtered_rule_mask)
# cache the representations
for i in range(len(filtered_representations)):
f_repr = filtered_representations[i]
rule_prompt = filtered_rule_prompts[i]
self.rule_repr_cache[rule_prompt] = f_repr
# obtain full representations
keys = []
j = 0
for ind in range(self.num_rules):
if ind in rule_indexes:
prompt = rule_prompts[j]
j+=1
if prompt in self.rule_repr_cache:
keys.append(self.rule_repr_cache[prompt])
else:
keys.append(torch.zeros(self.repr_size))
else:
keys.append(torch.zeros(self.repr_size))
keys = torch.stack(keys)
return keys
def generate_data(self, hole):
hole_parts = hole.split('/')[-1].split('_')
repo_name = hole.split('/')[2]
if len(hole_parts) > 3:
new_hole_parts = hole_parts[:-2]
filename = '_'.join(new_hole_parts)
filename = [filename]
else:
filename = [hole_parts[0]]
file = '/'.join(hole.split('/')[:-1] + filename)
hole_pos = (int(hole_parts[-2]), int(hole_parts[-1]))
rule_contexts = self.get_rule_context(file, hole_pos)
return rule_contexts, hole, repo_name
def set_tokenizer(self):
if self.emb_model_type == 'codebert':
self.tokenizer = AutoTokenizer.from_pretrained("microsoft/codebert-base")
if self.emb_model_type == 'graphcodebert':
self.tokenizer = AutoTokenizer.from_pretrained("microsoft/graphcodebert-base")
if self.emb_model_type == 'gpt-2':
self.tokenizer = GPT2TokenizerFast.from_pretrained("gpt2")
self.tokenizer.pad_token = self.tokenizer.eos_token
def set_embedding_model(self):
# CodeBERT
if self.emb_model_type == 'codebert':
self.emb_model = AutoModel.from_pretrained("microsoft/codebert-base")
# GraphCodeBERT
if self.emb_model_type == 'graphcodebert':
self.emb_model = AutoModel.from_pretrained("microsoft/graphcodebert-base")
def set_tokenizer(emb_model_type):
if emb_model_type == 'codebert':
tokenizer = AutoTokenizer.from_pretrained("microsoft/codebert-base")
if emb_model_type == 'graphcodebert':
tokenizer = AutoTokenizer.from_pretrained("microsoft/graphcodebert-base")
if emb_model_type == 'gpt-2':
tokenizer = GPT2TokenizerFast.from_pretrained("gpt2")
tokenizer.pad_token = tokenizer.eos_token
return tokenizer