# original wikiq headers are: title articleid revid date_time anon
# editor editor_id minor text_size text_entropy text_md5 reversion
# additions_size deletions_size
-import pdb
+
import argparse
import sys
import os, os.path
TO_ENCODE = ('title', 'editor')
PERSISTENCE_RADIUS=7
from deltas import SequenceMatcher
+from deltas import SegmentMatcher
+
+from dataclasses import dataclass
+import pandas as pd
+import pyarrow as pa
+import pyarrow.parquet as pq
+
+from typing import List
+
+class PersistMethod:
+ none = 0
+ sequence = 1
+ segment = 2
+ legacy = 3
def calculate_persistence(tokens_added):
return(sum([(len(x.revisions)-1) for x in tokens_added]),
def __init__(self, page, namespace_map, collapse_user=False):
self.id = page.id
self.namespace = page.namespace
+ # following mwxml, we assume namespace 0 in cases where
+ # page.namespace is inconsistent with namespace_map
+ if page.namespace not in namespace_map:
+ self.title = page.title
+ page.namespace = 0
if page.namespace != 0:
self.title = ':'.join([namespace_map[page.namespace], page.title])
else:
def __next__(self):
return next(self.__revisions)
-class WikiqParser():
- def __init__(self, input_file, output_file, collapse_user=False, persist=False, urlencode=False, persist_legacy=False):
+class RegexPair(object):
+ def __init__(self, pattern, label):
+ self.pattern = re.compile(pattern)
+ self.label = label
+ self.has_groups = bool(self.pattern.groupindex)
+ if self.has_groups:
+ self.capture_groups = list(self.pattern.groupindex.keys())
+
+ def _make_key(self, cap_group):
+ return ("{}_{}".format(self.label, cap_group))
+
+ def matchmake(self, content, rev_data):
+ temp_dict = {}
+ # if there are named capture groups in the regex
+ if self.has_groups:
+
+ # if there are matches of some sort in this revision content, fill the lists for each cap_group
+ if self.pattern.search(content) is not None:
+ m = self.pattern.finditer(content)
+ matchobjects = list(m)
+
+ for cap_group in self.capture_groups:
+ key = self._make_key(cap_group)
+ temp_list = []
+ for match in matchobjects:
+ # we only want to add the match for the capture group if the match is not None
+ if match.group(cap_group) != None:
+ temp_list.append(match.group(cap_group))
+
+ # if temp_list of matches is empty just make that column None
+ if len(temp_list)==0:
+ temp_dict[key] = None
+ # else we put in the list we made in the for-loop above
+ else:
+ temp_dict[key] = ', '.join(temp_list)
+
+ # there are no matches at all in this revision content, we default values to None
+ else:
+ for cap_group in self.capture_groups:
+ key = self._make_key(cap_group)
+ temp_dict[key] = None
+
+ # there are no capture groups, we just search for all the matches of the regex
+ else:
+ #given that there are matches to be made
+ if type(content) in(str, bytes):
+ if self.pattern.search(content) is not None:
+ m = self.pattern.findall(content)
+ temp_dict[self.label] = ', '.join(m)
+ else:
+ temp_dict[self.label] = None
+
+ # update rev_data with our new columns
+ rev_data.update(temp_dict)
+ return rev_data
+
+@dataclass
+class RevData():
+ revid: int
+ date_time: datetime
+ articleid: int
+ editorid: int
+ title: str
+ namespace: int
+ deleted: bool
+ text_chars: int
+ revert: bool
+ reverteds: list[bool]
+ sha1: str
+ text_chars: int
+ revert: bool
+ reverteds: list[int]
+ minor: bool
+ editor: str
+ anon: bool
+ collapsed_revs:int
+ token_revs:int
+ tokens_added:int
+ tokens_removed:int
+ tokens_window:int
+
+class WikiqParser():
+ def __init__(self, input_file, output_file, regex_match_revision, regex_match_comment, regex_revision_label, regex_comment_label, collapse_user=False, persist=None, urlencode=False, namespaces = None, revert_radius=15, output_parquet=True, parquet_buffer_size=2000):
+ """
+ Parameters:
+ persist : what persistence method to use. Takes a PersistMethod value
+ """
self.input_file = input_file
- self.output_file = output_file
+
self.collapse_user = collapse_user
self.persist = persist
- self.persist_legacy = persist_legacy
self.printed_header = False
self.namespaces = []
self.urlencode = urlencode
+ self.revert_radius = revert_radius
+ self.parquet_buffer = []
+ self.parquet_buffer_size = parquet_buffer_size
+
+ if namespaces is not None:
+ self.namespace_filter = set(namespaces)
+ else:
+ self.namespace_filter = None
+
+ self.regex_schemas = []
+ self.regex_revision_pairs = self.make_matchmake_pairs(regex_match_revision, regex_revision_label)
+ self.regex_comment_pairs = self.make_matchmake_pairs(regex_match_comment, regex_comment_label)
+
+
+ if output_parquet is True:
+ self.output_parquet = True
+ self.pq_writer = None
+ self.output_file = output_file
+ else:
+ self.output_file = open(output_file,'w')
+
+
+ def make_matchmake_pairs(self, patterns, labels):
+ if (patterns is not None and labels is not None) and \
+ (len(patterns) == len(labels)):
+ result = []
+ for pattern, label in zip(patterns, labels):
+ result.append(RegexPair(pattern, label))
+ self.regex_schemas.append(pa.field(label, pa.list_(pa.string())))
+
+ return result
+ elif (patterns is None and labels is None):
+ return []
+ else:
+ sys.exit('Each regular expression *must* come with a corresponding label and vice versa.')
+
+ def matchmake(self, rev, rev_data):
+ rev_data = self.matchmake_revision(rev.text, rev_data)
+ rev_data = self.matchmake_comment(rev.comment, rev_data)
+ return rev_data
+
+ def matchmake_revision(self, text, rev_data):
+ return self.matchmake_pairs(text, rev_data, self.regex_revision_pairs)
+
+ def matchmake_comment(self, comment, rev_data):
+ return self.matchmake_pairs(comment, rev_data, self.regex_comment_pairs)
+
+ def matchmake_pairs(self, text, rev_data, pairs):
+ for pair in pairs:
+ rev_data = pair.matchmake(text, rev_data)
+ return rev_data
+
def __get_namespace_from_title(self, title):
default_ns = None
# if we've made it this far with no matches, we return the default namespace
return default_ns
+
def process(self):
# create a regex that creates the output filename
# Iterate through pages
for page in dump:
- rev_detector = mwreverts.Detector()
+ namespace = page.namespace if page.namespace is not None else self.__get_namespace_from_title(page.title)
+
+ # skip namespaces not in the filter
+ if self.namespace_filter is not None:
+ if namespace not in self.namespace_filter:
+ continue
- if self.persist or self.persist_legacy:
+ rev_detector = mwreverts.Detector(radius = self.revert_radius)
+
+ if self.persist != PersistMethod.none:
window = deque(maxlen=PERSISTENCE_RADIUS)
- if not self.persist_legacy:
+ if self.persist == PersistMethod.sequence:
state = mwpersistence.DiffState(SequenceMatcher(tokenizer = wikitext_split),
revert_radius=PERSISTENCE_RADIUS)
+ elif self.persist == PersistMethod.segment:
+ state = mwpersistence.DiffState(SegmentMatcher(tokenizer = wikitext_split),
+ revert_radius=PERSISTENCE_RADIUS)
+
+ # self.persist == PersistMethod.legacy
else:
from mw.lib import persistence
state = persistence.State()
# Iterate through a page's revisions
for rev in page:
-
- rev_data = {'revid' : rev.id,
- 'date_time' : rev.timestamp.strftime('%Y-%m-%d %H:%M:%S'),
- 'articleid' : page.id,
- 'editor_id' : "" if rev.deleted.user == True or rev.user.id is None else rev.user.id,
- 'title' : '"' + page.title + '"',
- 'namespace' : page.namespace if page.namespace is not None else self.__get_namespace_from_title(page.title),
- 'deleted' : "TRUE" if rev.deleted.text else "FALSE" }
+
+ # initialize rev_data
+ rev_data = {
+ 'revid':rev.id,
+ 'date_time' : rev.timestamp.strftime('%Y-%m-%d %H:%M:%S'),
+ 'articleid' : page.id,
+ 'editorid' : "" if rev.deleted.user == True or rev.user.id is None else rev.user.id,
+ 'title' : '"' + page.title + '"',
+ 'namespace' : namespace,
+ 'deleted' : "TRUE" if rev.deleted.text else "FALSE"
+ }
+
+ rev_data = self.matchmake(rev, rev_data)
# if revisions are deleted, /many/ things will be missing
if rev.deleted.text:
# TODO rev.bytes doesn't work.. looks like a bug
rev_data['text_chars'] = len(rev.text)
-
+
# generate revert data
revert = rev_detector.process(text_sha1, rev.id)
if self.collapse_user:
rev_data['collapsed_revs'] = rev.collapsed_revs
- if self.persist or self.persist_legacy:
+ if self.persist != PersistMethod.none:
if rev.deleted.text:
-
for k in ["token_revs", "tokens_added", "tokens_removed", "tokens_window"]:
old_rev_data[k] = None
else:
- if not self.persist_legacy:
+ if self.persist != PersistMethod.legacy:
_, tokens_added, tokens_removed = state.update(rev.text, rev.id)
else:
rev_count += 1
- if self.persist or self.persist_legacy:
+ if self.persist != PersistMethod.none:
# print out metadata for the last RADIUS revisions
for i, item in enumerate(window):
# if the window was full, we've already printed item 0
rev_data["tokens_added"] = num_tokens
rev_data["tokens_removed"] = len(tokens_removed)
rev_data["tokens_window"] = len(window)-(i+1)
-
self.print_rev_data(rev_data)
page_count += 1
print("Done: %s revisions and %s pages." % (rev_count, page_count),
file=sys.stderr)
+ if self.output_parquet is True:
+ self.flush_parquet_buffer()
+ self.pq_writer.close()
+
+ else:
+ output_file.close()
+
+
+ def write_parquet_row(self, rev_data):
+ if 'deleted' in rev_data.keys():
+ rev_data['deleted'] = True if rev_data['deleted'] == "TRUE" else False
+
+ if 'minor' in rev_data.keys():
+ rev_data['minor'] = True if rev_data['minor'] == "TRUE" else False
+
+
+ if 'anon' in rev_data.keys():
+ rev_data['anon'] = True if rev_data['anon'] == "TRUE" else False
+
+
+ self.parquet_buffer.append(rev_data)
+
+ if len(self.parquet_buffer) >= self.parquet_buffer_size:
+ self.flush_parquet_buffer()
+
+ def flush_parquet_buffer(self):
+ outtable = pd.DataFrame.from_records(self.parquet_buffer)
+ outtable = pa.Table.from_pandas(outtable)
+ if self.pq_writer is None:
+ schema = outtable.schema
+ for regex_schema in self.regex_schemas:
+ schema.append(regex_schema)
+
+ self.pq_writer = pq.ParquetWriter(self.output_file, schema, flavor='spark')
+
+ self.pq_writer.write_table(outtable)
+
def print_rev_data(self, rev_data):
+
+ if self.output_parquet is False:
+ printfunc = lambda rev_data: print("\t".join(rev_data), file=self.output_file)
+ else:
+ printfunc = self.write_parquet_row
+
# if it's the first time through, print the header
if self.urlencode:
for field in TO_ENCODE:
rev_data[field] = quote(str(rev_data[field]))
if not self.printed_header:
- print("\t".join([str(k) for k in sorted(rev_data.keys())]), file=self.output_file)
+ printfunc(rev_data)
self.printed_header = True
- print("\t".join([str(v) for k, v in sorted(rev_data.items())]), file=self.output_file)
+ printfunc(rev_data)
def open_input_file(input_filename):
if re.match(r'.*\.7z$', input_filename):
- cmd = ["7za", "x", "-so", input_filename, '*']
+ cmd = ["7za", "x", "-so", input_filename, "*.xml"]
elif re.match(r'.*\.gz$', input_filename):
cmd = ["zcat", input_filename]
elif re.match(r'.*\.bz2$', input_filename):
return input_file
-def open_output_file(input_filename):
- # create a regex that creates the output filename
+def get_output_filename(input_filename, parquet = False):
output_filename = re.sub(r'\.(7z|gz|bz2)?$', '', input_filename)
output_filename = re.sub(r'\.xml', '', output_filename)
- output_filename = output_filename + ".tsv"
- output_file = open(output_filename, "w")
+ if parquet is False:
+ output_filename = output_filename + ".tsv"
+ else:
+ output_filename = output_filename + ".parquet"
+ return output_filename
+def open_output_file(input_filename):
+ # create a regex that creates the output filename
+ output_filename = get_output_filename(input_filename, parquet = False)
+ output_file = open(output_filename, "w")
return output_file
parser = argparse.ArgumentParser(description='Parse MediaWiki XML database dumps into tab delimitted data.')
help="Filename of the compressed or uncompressed XML database dump. If absent, we'll look for content on stdin and output on stdout.")
parser.add_argument('-o', '--output-dir', metavar='DIR', dest='output_dir', type=str, nargs=1,
- help="Directory for output files.")
+ help="Directory for output files. If it ends with .parquet output will be in parquet format.")
parser.add_argument('-s', '--stdout', dest="stdout", action="store_true",
help="Write output to standard out (do not create dump file)")
parser.add_argument('--collapse-user', dest="collapse_user", action="store_true",
help="Operate only on the final revision made by user a user within all sequences of consecutive edits made by a user. This can be useful for addressing issues with text persistence measures.")
-parser.add_argument('-p', '--persistence', dest="persist", action="store_true",
- help="Compute and report measures of content persistent: (1) persistent token revisions, (2) tokens added, and (3) number of revision used in computing the first measure.")
+parser.add_argument('-p', '--persistence', dest="persist", default=None, const='', type=str, choices = ['','segment','sequence','legacy'], nargs='?',
+ help="Compute and report measures of content persistent: (1) persistent token revisions, (2) tokens added, and (3) number of revision used in computing the first measure. This may by slow. The defualt is -p=sequence, which uses the same algorithm as in the past, but with improvements to wikitext parsing. Use -p=legacy for old behavior used in older research projects. Use -p=segment for advanced persistence calculation method that is robust to content moves, but prone to bugs, and slower.")
parser.add_argument('-u', '--url-encode', dest="urlencode", action="store_true",
help="Output url encoded text strings. This works around some data issues like newlines in editor names. In the future it may be used to output other text data.")
-parser.add_argument('--persistence-legacy', dest="persist_legacy", action="store_true",
- help="Legacy behavior for persistence calculation. Output url encoded text strings. This works around some data issues like newlines in editor names. In the future it may be used to output other text data.")
+parser.add_argument('-n', '--namespace-include', dest="namespace_filter", type=int, action='append',
+ help="Id number of namspace to include. Can be specified more than once.")
+
+parser.add_argument('-rr',
+ '--revert-radius',
+ dest="revert_radius",
+ type=int,
+ action='store',
+ default=15,
+ help="Number of edits to check when looking for reverts (default: 15)")
+
+parser.add_argument('-RP', '--revision-pattern', dest="regex_match_revision", default=None, type=str, action='append',
+ help="The regular expression to search for in revision text. The regex must be surrounded by quotes.")
+
+parser.add_argument('-RPl', '--revision-pattern-label', dest="regex_revision_label", default=None, type=str, action='append',
+ help="The label for the outputted column based on matching the regex in revision text.")
+
+parser.add_argument('-CP', '--comment-pattern', dest="regex_match_comment", default=None, type=str, action='append',
+ help="The regular expression to search for in comments of revisions.")
+
+parser.add_argument('-CPl', '--comment-pattern-label', dest="regex_comment_label", default=None, type=str, action='append',
+ help="The label for the outputted column based on matching the regex in comments.")
args = parser.parse_args()
+
+
+# set persistence method
+
+if args.persist is None:
+ persist = PersistMethod.none
+elif args.persist == "segment":
+ persist = PersistMethod.segment
+elif args.persist == "legacy":
+ persist = PersistMethod.legacy
+else:
+ persist = PersistMethod.sequence
+
+if args.namespace_filter is not None:
+ namespaces = args.namespace_filter
+else:
+ namespaces = None
+
if len(args.dumpfiles) > 0:
+ output_parquet = False
for filename in args.dumpfiles:
input_file = open_input_file(filename)
else:
output_dir = "."
+ if output_dir.endswith(".parquet"):
+ output_parquet = True
+
print("Processing file: %s" % filename, file=sys.stderr)
if args.stdout:
output_file = sys.stdout
else:
filename = os.path.join(output_dir, os.path.basename(filename))
- output_file = open_output_file(filename)
+ output_file = get_output_filename(filename, parquet = output_parquet)
- wikiq = WikiqParser(input_file, output_file,
+ wikiq = WikiqParser(input_file,
+ output_file,
collapse_user=args.collapse_user,
- persist=args.persist,
- persist_legacy=args.persist_legacy,
- urlencode=args.urlencode)
-
-
+ persist=persist,
+ urlencode=args.urlencode,
+ namespaces=namespaces,
+ revert_radius=args.revert_radius,
+ regex_match_revision = args.regex_match_revision,
+ regex_revision_label = args.regex_revision_label,
+ regex_match_comment = args.regex_match_comment,
+ regex_comment_label = args.regex_comment_label,
+ output_parquet=output_parquet)
+
+ print(wikiq.output_parquet)
wikiq.process()
# close things
input_file.close()
- output_file.close()
+
else:
- wikiq = WikiqParser(sys.stdin, sys.stdout,
+ wikiq = WikiqParser(sys.stdin,
+ sys.stdout,
collapse_user=args.collapse_user,
- persist=args.persist,
- persist_legacy=args.persist_legacy,
- urlencode=args.urlencode)
- wikiq.process()
+ persist=persist,
+ #persist_legacy=args.persist_legacy,
+ urlencode=args.urlencode,
+ namespaces=namespaces,
+ revert_radius=args.revert_radius,
+ regex_match_revision = args.regex_match_revision,
+ regex_revision_label = args.regex_revision_label,
+ regex_match_comment = args.regex_match_comment,
+ regex_comment_label = args.regex_comment_label)
+
+ wikiq.process()
# stop_words = "a,able,about,across,after,all,almost,also,am,among,an,and,any,are,as,at,be,because,been,but,by,can,cannot,could,dear,did,do,does,either,else,ever,every,for,from,get,got,had,has,have,he,her,hers,him,his,how,however,i,if,in,into,is,it,its,just,least,let,like,likely,may,me,might,most,must,my,neither,no,nor,not,of,off,often,on,only,or,other,our,own,rather,said,say,says,she,should,since,so,some,than,that,the,their,them,then,there,these,they,this,tis,to,too,twas,us,wants,was,we,were,what,when,where,which,while,who,whom,why,will,with,would,yet,you,your"
# stop_words = stop_words.split(",")
projects / mediawiki_dump_tools.git / blobdiff