@contextlib.contextmanager
def timer(name: str = "-"):
start = time.time()
yield None
delay = time.time() - start
print(f"{name} took {delay:.1f}s")
def benchmark(tmp_path: Path):
segments = [
"crawl-data/CC-MAIN-2019-09/segments/1550249406966.99/wet/CC-MAIN-20190222220601-20190223002601-00441.warc.wet.gz"
]
seg_file = tmp_path / "CC-MAIN-20190222220601-20190223002601-00441.warc.wet.gz"
with timer("from network"):
list(CCSegmentsReader(segments))
with timer("from network, with caching"):
list(CCSegmentsReader(segments, cache_dir=tmp_path))
assert seg_file.exists()
with timer("from disk"):
CCSegmentsReader(segments, cache_dir=tmp_path)
seg_file.unlink()
if __name__ == "__main__":
func_argparse.main(ls, dl)
urls.append(CC_NET_ROOT_FOLDER + dump + "/" + file)
return urls
def reproduce(
language: List[str] = None,
dump: str = "2019-09",
bucket: str = "head",
output_dir: Path = DATA / "reconstruct",
execution: str = "mp",
parallelism: int = -1,
cache_dir: Path = None,
):
"""Reproduce paper results from official CC snapshot and precomputed results.
- dump: CC dump id
- bucket: can be one of ("head", "middle", "tail", "all")
- ouput_dir: output directory
- execution: how to parallelize ("mp", "debug", "slurm", ...)
- cache_dir: where the CC .wet files will be downloaded.
"""
output_dir.mkdir(exist_ok=True, parents=True)
urls = select_urls(dump, language, bucket)
unminify(urls, output_dir / dump, execution, parallelism, cache_dir)
if __name__ == "__main__":
import func_argparse
func_argparse.main(reproduce, minify_file, minify, unminify, unminify_file)
with open(src_file) as src_f, open(trg_file) as trg_f:
src_l = src_f.readline()
trg_l = trg_f.readline()
while src_l and trg_l:
src: SimpleBitext = parser(src_l)
trg: SimpleBitext = parser(trg_l)
if src.line_no <= trg.line_no:
lines_src += 1
src_l = src_f.readline()
if trg.line_no <= src.line_no:
lines_trg += 1
trg_l = trg_f.readline()
if trg.line_no == src.line_no:
found_pairs += 1
if found_pairs == lines_src and found_pairs == lines_trg:
logging.info(
f"Validated {src_file} and {trg_file}. Found {found_pairs} bitexts."
)
else:
logging.error(
f"Validated {src_file} and {trg_file}. "
f"Found {found_pairs} bitexts, from {lines_src} in {src_file} and {lines_trg} in {trg_file}"
)
if __name__ == "__main__":
import func_argparse
func_argparse.main(dl, finalize)
dumps.remove("..")
dumps.remove("current")
# We take the oldest dump since the most recent might be incomplete.
# The page only link to the N latest dumps so the dump won't be too old.
date = min(dumps)
cirrus_url = "/".join((CIRRUS_URL, date))
print("Will use the Wikipedia dump from:", date, cirrus_url)
cirrus_page = BeautifulSoup(urllib.request.urlopen(cirrus_url),
features="html.parser")
urls = {}
for link in cirrus_page.findAll("a"):
match = CIRRUS_DUMP_RE.match(link.get("href"))
if not match:
continue
urls[match.group(1)] = "/".join([cirrus_url, link.get("href")])
assert urls, f"No valid download urls found at {cirrus_url}"
return urls
def wget(url: str, output: Path):
subprocess.run(["wget", url, "-O", tmp(output), "-q"], check=True)
tmp(output).replace(output)
assert (output.stat().st_size >
10_000), f"File {output} downloaded from {url} looks too small"
if __name__ == "__main__":
func_argparse.main(dl, opening)
assert len(files) > 0, "No files given."
output_dir.mkdir(exist_ok=True)
outputs = [output_dir / str(f).split("/")[-1] for f in files]
if cache_dir is None:
cache_dir = output_dir / "wet_cache"
cache_dir.mkdir(exist_ok=True)
if str(cache_dir) == "none":
cache_dir = None
files = [f for f, o in zip(files, outputs) if not o.exists()]
outputs = [o for o in outputs if not o.exists()]
if not files:
return
ex = get_executor(
"unminify",
output_dir / "logs",
execution,
timeout_hour=8,
cpus=1,
task_parallelism=parallelism,
mem_gb=32,
)
ex(fetch_metadata_file, files, outputs, itertools.repeat(cache_dir))
if __name__ == "__main__":
import func_argparse
func_argparse.main(minify_file, minify, fetch_metadata,
fetch_metadata_file)
ex = submitit.AutoExecutor(output_dir / "mining_logs")
ex.update_parameters(
name="mine",
cpus_per_task=PROCESSES,
timeout_min=60 * 24 // PROCESSES,
mem_gb=10,
)
jobs = ex.map_array(_mine, files, outputs, sp, lm, thresholds)
print("Submited job array:", jobs[0])
for j in submitit.helpers.as_completed(jobs):
(i, o) = j.result()
print("Mined sentences from", i, "to", o)
return outputs
def _mine(file: Path, output: Path, sp: Path, lm: Path,
threshold: float) -> Tuple[Path, Path]:
extractor = ExtractSentences(sp,
lm,
field="raw_content",
threshold=threshold)
jsonql.run_pipes(extractor, file=file, output=output, processes=PROCESSES)
return (file, output)
if __name__ == "__main__":
func_argparse.main(sample, mine)
"""Say hello or goodbye to the user."""
import func_argparse
def hello(user: str, times: int = None):
"""Say hello.
Arguments:
user: name of the user
"""
print(f"Hello {user}" * (1 if times is None else times))
def bye(user: str, see_you: float = 1.0):
"""Say goodbye."""
print(f"Goodbye {user}, see you in {see_you:.1f} days")
if __name__ == "__main__":
func_argparse.main()
