File size: 17,680 Bytes
3f53269 2b42f63 3f53269 2b42f63 3f53269 2b42f63 3f53269 2b42f63 3f53269 2b42f63 3f53269 2b42f63 3f53269 2b42f63 3f53269 2b42f63 3f53269 2b42f63 3f53269 2b42f63 3f53269 2b42f63 3f53269 2b42f63 3f53269 2b42f63 3f53269 2b42f63 3f53269 2b42f63 3f53269 2b42f63 3f53269 6a91027 3f53269 6a91027 3f53269 6a91027 3f53269 6a91027 3f53269 6a91027 3f53269 6a91027 3f53269 6a91027 3f53269 6a91027 3f53269 6a91027 3f53269 6a91027 3f53269 6a91027 3f53269 6a91027 3f53269 6a91027 3f53269 6a91027 3f53269 6a91027 3f53269 6a91027 3f53269 6a91027 3f53269 2b42f63 3f53269 2b42f63 3f53269 2b42f63 3f53269 2b42f63 3f53269 2b42f63 3f53269 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 | import re
import os
import sys
import math
import tokenize
import tiktoken
import tempfile
import jsonlines
import subprocess
import scipy.stats
from io import StringIO
from statistics import mean
from tree_sitter import Language, Parser
from nltk.tokenize import wordpunct_tokenize
from nltk.translate.bleu_score import sentence_bleu, SmoothingFunction
def pass_at_k_continuous_vals(n, k, vals):
#Score@k,n is the continuous value version of pass@k,n defined in MGCoder paper.
assert len(vals) == n
assert n >= k, (n, k)
assert all(vals[i-1] >= vals[i] for i in range(1, len(vals))), (n, k, vals)
isum = 0
for i in range(1, n-k+2):
# i ranges from 1 to n-k+1
isum += (vals[i-1]*math.comb(n-i,k-1))
return isum/math.comb(n,k)
def remove_comments(code: str, language: str) -> str:
#Using re module to remove comments for specific languages
if language.lower() == "python":
try:
return remove_py_comments(code)
except:
pattern = r'\'{3}[\s\S]*?\'{3}|\"{3}[\s\S]*?\"{3}|\#[^\n]*'
elif language in ["java", "javascript", "scala", "kotlin", "c++", "c", "ino", "objectivec"]:
pattern = r"\/\*[\s\S]*?\*\/|\/\/[^\n]*"
elif language == 'assembly':
pattern = r';.*|\#.*|\/*[\s\S]*?\*\/'
elif language == 'javascript xml':
pattern = r"\/\*[\s\S]*?\*\/|\/\/[^\n]*|<!--.*?-->"
code=re.sub(pattern, '', code)
return code
def remove_blank_lines(code) -> str:
#Remove blank lines in a string
try:
lines = code.split("\n")
non_blank_lines = [line for line in lines if line.strip() != ""]
return "\n".join(non_blank_lines)
except:
return code
def diff_bleu(source_code, target, generated_answers, pl):
"""Calculating the DiffBleu score.
It is the bleu score between the git diff of the source and generated code and git diff of the source and target."""
with tempfile.NamedTemporaryFile(mode = 'w', delete = False) as source_temp, tempfile.NamedTemporaryFile(mode = 'w', delete = False) as target_temp, tempfile.NamedTemporaryFile(mode = 'w', delete = False) as generated_temp:
source_temp.write(remove_blank_lines(remove_comments(source_code,pl.lower())))
target_temp.write(remove_blank_lines(remove_comments(target,pl.lower())))
generated_temp.write(remove_blank_lines(remove_comments(generated_answers,pl.lower())))
source_path = source_temp.name
target_path = target_temp.name
generated_answers_path = generated_temp.name
command_diff_generated = "git diff -U0 --no-index --ignore-all-space --ignore-blank-lines {} {} | tail -n +5 | grep -v 'No newline at end of file'".format(source_path,generated_answers_path)
command_diff_target = "git diff -U0 --no-index --ignore-all-space --ignore-blank-lines {} {} | tail -n +5 | grep -v 'No newline at end of file'".format(source_path,target_path)
diff_generated = subprocess.run(command_diff_generated, shell=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE).stdout.decode()
diff_target = subprocess.run(command_diff_target, shell=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE).stdout.decode()
diff_generated = wordpunct_tokenize(diff_generated)
diff_target = wordpunct_tokenize(diff_target)
diff_score_bleu = sentence_bleu([diff_target], diff_generated, smoothing_function=SmoothingFunction().method1)
if remove_blank_lines(remove_comments(generated_answers, pl.lower())).strip() == "":
return 0
return diff_score_bleu
def get_welch_t_test_p(m1: float, s1: float, m2: float, s2: float, n1: int, n2: int) -> float:
"""Returns the p-value of a Welch's t-test. The null hypothesis is that the two samples have the same mean.
Alternative hypothesis is that the first sample has a smaller mean than the second sample.
The first distribution is (m1, s1) and the second distribution is (m2, s2). The number of samples in each distribution is n1 and n2 respectively.
Returns:
float: p-value
"""
return scipy.stats.ttest_ind_from_stats(
mean1 = m1,
mean2 = m2,
std1 = s1,
std2 = s2,
nobs1 = n1,
nobs2 = n2,
equal_var = False, # Welch's t-test - unequal variances
alternative = "less", # one-sided - generated is faster
).pvalue
def num_tokens_from_string(string: str, model: str) -> int:
"""Returns the number of tokens in a text string."""
encoding = tiktoken.encoding_for_model(model)
num_tokens = len(encoding.encode(string))
return num_tokens
def statistical_significance_test(output_path,n,k_values):
avg_speedup = {}
fun_correct = {}
imp = {}
for k in k_values:
avg_speedup[k] = []
fun_correct[k] = []
imp[k] = []
with jsonlines.open(output_path) as f:
for item in f:
prob_id = item['problem_id']
scores = []
correct = []
sig = []
for i in range(n):
if item[f"generated_answers_{i}_acc"] != 1:
scores.append(1)
correct.append(0)
sig.append(0)
else:
m1 = item[f'generated_answers_{i}_time_mean']
s1 = item[f'generated_answers_{i}_time_std']
nobs1 = 25
m2 = item['input_time_mean']
s2 = item['input_time_std']
nobs2 = 25
welch = get_welch_t_test_p(m1,s1,m2,s2,nobs1,nobs2)
if(welch<.05 and item['input_time_mean'] > item[f'generated_answers_{i}_time_mean']):
scores.append(item['input_time_mean']/item[f'generated_answers_{i}_time_mean'])
sig.append(1)
else:
scores.append(1)
sig.append(0)
correct.append(1)
scores = sorted(scores,reverse = True)
correct = sorted(correct,reverse = True)
sig = sorted(sig,reverse = True)
for k in k_values:
avg_speedup[k].append(pass_at_k_continuous_vals(n, k, scores))
fun_correct[k].append(pass_at_k_continuous_vals(n, k, correct))
imp[k].append(pass_at_k_continuous_vals(n, k, sig))
scores = []
for k in k_values:
scores.append(round(mean(avg_speedup[k]), 3))
scores.append(round(mean(fun_correct[k]), 3))
scores.append(round(mean(imp[k]), 3))
return scores
def get_files_with_syntax_errors(generated_code_path, codeql_db_path, query):
#Find the files with syntax errors on running codeql by finding warning pattern in the db logs created.
error_files = []
error_pattern = r"\[WARN\] \[\d*\] Failed to analyse imports of ([a-zA-Z0-9\\/.:_\-\(\)\']*) : Syntax Error \(line \d*\)"
error_pattern_expr = re.compile(error_pattern)
parent_path = os.path.abspath(generated_code_path + "/")
log_dir = codeql_db_path + "/{}/log/".format(query)
if not os.path.exists(log_dir):
raise FileNotFoundError(log_dir)
for p in os.listdir(log_dir):
if p.startswith('database-create'):
log_path = p
with open(log_dir + log_path, 'r') as f:
logs = f.read()
files_with_syntax_error = error_pattern_expr.findall(logs)
for file_with_error in files_with_syntax_error:
if not os.path.exists(file_with_error):
print("danger: ", file_with_error)
sys.exit(-1)
child_path = os.path.abspath(file_with_error)
if os.path.commonpath([parent_path]) == os.path.commonpath([parent_path, child_path]):
error_files.append(file_with_error.split("/")[-1])
return error_files
def remove_py_comments(source):
#Robustly remove python specific comments
io_obj = StringIO(source)
out = ""
prev_toktype = tokenize.INDENT
last_lineno = -1
last_col = 0
for tok in tokenize.generate_tokens(io_obj.readline):
token_type = tok[0]
token_string = tok[1]
start_line, start_col = tok[2]
end_line, end_col = tok[3]
ltext = tok[4]
if start_line > last_lineno:
last_col = 0
if start_col > last_col:
out += (" " * (start_col - last_col))
if token_type == tokenize.COMMENT:
pass
elif token_type == tokenize.STRING:
if prev_toktype != tokenize.INDENT:
if prev_toktype != tokenize.NEWLINE:
if start_col > 0:
out += token_string
else:
out += token_string
prev_toktype = token_type
last_col = end_col
last_lineno = end_line
return out
def check_syntax(code,language):
#Checks if the code passed parses without any errors using tree-sitter
code += '\n'
if(language.lower() == "java"):
path = 'src/evaluation/tree-sitter/tree-sitter-java'
elif(language.lower() == "python"):
path = 'src/evaluation/tree-sitter/tree-sitter-python'
elif(language.lower() == "scala"):
path = 'src/evaluation/tree-sitter/tree-sitter-scala'
elif(language.lower() == "c"):
path = 'src/evaluation/tree-sitter/tree-sitter-c'
elif(language.lower() == "c++"):
path = 'src/evaluation/tree-sitter/tree-sitter-cpp'
elif(language.lower() == "objectivec"):
path = 'src/evaluation/tree-sitter/tree-sitter-objc'
elif(language.lower() == "javascript"):
path = 'src/evaluation/tree-sitter/tree-sitter-javascript'
elif(language.lower() == "kotlin"):
path = 'src/evaluation/tree-sitter/tree-sitter-kotlin'
else:
return(False)
Language.build_library(
'src/evaluation/tree-sitter/build/my-languages_{}.so'.format(language.lower()),
[
path
]
)
if(language.lower() == "java"):
LANGUAGE = Language('src/evaluation/tree-sitter/build/my-languages_java.so', 'java')
elif(language.lower() == "python"):
LANGUAGE = Language('src/evaluation/tree-sitter/build/my-languages_python.so', 'python')
elif(language.lower() == "scala"):
LANGUAGE = Language('src/evaluation/tree-sitter/build/my-languages_scala.so', 'scala')
elif(language.lower() == "c"):
LANGUAGE = Language('src/evaluation/tree-sitter/build/my-languages_c.so', 'c')
elif(language.lower() == "c++"):
LANGUAGE = Language('src/evaluation/tree-sitter/build/my-languages_c++.so', 'cpp')
elif(language.lower() == "objectivec"):
LANGUAGE = Language('src/evaluation/tree-sitter/build/my-languages_objectivec.so', 'objc')
elif(language.lower() == "javascript"):
LANGUAGE = Language('src/evaluation/tree-sitter/build/my-languages_javascript.so', 'javascript')
elif(language.lower() == "kotlin"):
LANGUAGE = Language('src/evaluation/tree-sitter/build/my-languages_kotlin.so', 'kotlin')
parser = Parser()
parser.set_language(LANGUAGE)
tree = parser.parse(bytes(code, "utf8"))
def find_error(node):
if node.type == 'ERROR':
print(f'Error found from line {node.start_point[0]+1}, column {node.start_point[1]+1} to line {node.end_point[0]+1}, column {node.end_point[1]+1}')
for child in node.children:
find_error(child)
return not(tree.root_node.has_error)
def extract_parsable_code(start, code, code_list,top,bottom, pl):
#Checking thw largest parsable combination in a sentence of text
while top < bottom:
code1 = " ".join(code_list[top:bottom])
code2 = " ".join(code_list[top:bottom-1])
code3 = " ".join(code_list[top+1:bottom])
if(start):
if check_syntax(code1 + "\n" + code,pl):
return code1
elif check_syntax(code2 + "\n" + code,pl):
return code2
elif check_syntax(code3 + "\n" + code,pl):
return code3
else:
top += 1
bottom -= 1
else:
if check_syntax(code+"\n" + code1,pl):
return code1
elif check_syntax(code+"\n" + code2,pl):
return code2
elif check_syntax(code+"\n" + code3,pl):
return code3
else:
top += 1
bottom -= 1
return None
def post_process_generations(generated_answers: str, model: str, prompt:str, pl: str) -> str:
"""Post processing outputs to first extract the code between backquotes after response as defined in the template.
Failing which we try to use tree-sitter to obtain maximum parsable block of code but it is the models failure to not follow the tempate
"""
failed = 0
generated_answers = remove_blank_lines(generated_answers)
if(prompt == "one-shot" or prompt == "chain_of_thought"):
index = 2
else:
index = 1
#Extracting the code after the Response keyword within triple backquotes
try:
generated_answers=generated_answers.split('Response:')[index].strip().split("Instruction:")[0]
generated_answers_post=generated_answers.split("```")[1:]
generated_answers_post = "\n".join("```".join(generated_answers_post).split('\n')[1:])
generated_pass = 0
if(generated_answers_post.find("```") != -1):
generated_answers_post=generated_answers_post.split('```')[0]
passed = 1
if(generated_answers_post == ""):
passed = 0
return [passed,generated_answers_post]
else:
failed = 1
except:
if(prompt=="chain_of_thought"):
try:
generated_answers="```".join(generated_answers.split("Thought")[2].split("```")[1:])
if(generated_answers.find("```")!=-1):
generated_answers = "\n".join(generated_answers.split("```")[0].split("\n")[1:])
return [1,generated_answers]
else:
failed = 1
except:
failed = 1
else:
failed = 1
unsupported_pl = ['javascript xml', 'ino', 'assembly']
if(pl.lower() in unsupported_pl):
generated_answers = remove_blank_lines(generated_answers.strip())
return [0,generated_answers]
# If the template is not followed, we try to use tree-sitter to extract parsable code
if (failed):
generated_answers = generated_answers.replace('\"','"').replace("\'","'").replace("\/\/","//").replace("\/*","/*").replace("*\/","*/")
example_list = generated_answers.splitlines()
for i in range(len(example_list)):
if(check_syntax(example_list[i],pl)):
start_index = i
break
try:
start_index
except:
start_index = 0
if len(example_list) == 0:
return [0,generated_answers]
example_list[start_index] = example_list[start_index].strip()
last_index = len(example_list)
#Finding largest possible codeblock that parses
while(last_index > start_index):
code=""
for j in range(start_index, last_index):
code += example_list[j]+"\n"
if(check_syntax(code,pl)):
last_index = last_index-1
break
else:
last_index -= 1
line_parse_start = ""
line_parse_end = ""
#Checking if the line before the current block can also be parsed after tokeninizing by spaces.
if(start_index != 0):
ind = start_index-1
line = example_list[ind]
try:
tokenized_sentence = line.strip("```").split(" ")
except:
tokenized_sentence = line.strip("```")
line_parse_start = extract_parsable_code(start = True, code = code, code_list = tokenized_sentence, top = 0, bottom = len(tokenized_sentence), pl = pl)
flag = 1
if line_parse_start == None:
line_parse_start = ""
end_split = line_parse_start.strip().split(" ")
if(len(end_split) == 1):
if(end_split[0].isalnum()):
flag = 0
if(check_syntax(line_parse_start+"\n"+code,pl) and flag):
code = line_parse_start + "\n"+ code
#Checking if the line after the current block can also be parsed after tokeninizing by spaces.
if(last_index != (len(example_list)-1)):
ind = last_index+1
line = example_list[ind]
try:
tokenized_sentence = line.strip("```").split(" ")
except:
tokenized_sentence = line.strip("```")
line_parse_end=extract_parsable_code(start = False, code = code, code_list = tokenized_sentence, top = 0, bottom = len(tokenized_sentence), pl = pl)
if line_parse_end == None:
line_parse_end = ""
end_split = line_parse_end.strip().split(" ")
if(len(end_split) == 1):
if(end_split[0].isalnum()):
code = remove_blank_lines(code.strip())
return [0,code]
if(check_syntax(code + "\n" + line_parse_end, pl)):
code+="\n" + line_parse_end
code = remove_blank_lines(code.strip())
return [0,code] |