indo_wiki / cleanse_wiki_data.py
sabilmakbar's picture
Init PR (#1)
7725656
raw
history blame
10.9 kB
'''
Script on Cleansing Wikipedia Data that has been extracted from extract_raw_wiki_data.py
'''
#core functionality modules
import os, gc
import logging
import argparse
import warnings
from functools import partial
#text preprocess modules
import re
import urllib
from xml.etree import ElementTree as ET
#dataset related modules
import numpy as np
import pandas as pd
### MODULES DEFINITION ###
#create custom type-checking of incoming ArgParse
def argparse_bool_check(value: str):
#cast str with value like float into actual float
try:
value = float(value)
#can't be parsed as float, keep as it is
except ValueError:
pass
#cast float-like value (incl int) into str
if isinstance(value, float) and int(value) == value:
value = str(int(value))
#raise ArgumentTypeError if the value isn't in string already
else:
if not isinstance(value, str):
raise argparse.ArgumentTypeError(f"Not the correct value (args: {value})! Expected is cast-able to '1' or '0' or already in string. Please rectify!")
#check for these combinations of values
if value.lower() in ("yes", "true", "t", "y", "1"):
return True
elif value.lower() in ("no", "false", "f", "n", "0"):
return False
else:
raise argparse.ArgumentTypeError(f"Value Error! Not the correct value (args: {value})! Please rectify!")
def set_logger():
# Set up the logger
logging.basicConfig(
level=logging.INFO, # Set the desired logging level (DEBUG, INFO, WARNING, ERROR, CRITICAL)
format='%(asctime)s [%(levelname)s]: %(message)s', # Customize the log message format
datefmt='%Y-%m-%d %H:%M:%S' # Customize the date/time format
)
# Create a file handler to write logs into a file
file_handler = logging.FileHandler('app.log')
# Set the log level for the file handler
file_handler.setLevel(logging.INFO)
# Create a formatter for the file handler (customize the log format for the file)
file_formatter = logging.Formatter('%(asctime)s [%(levelname)s]: %(message)s', datefmt='%Y-%m-%d %H:%M:%S')
file_handler.setFormatter(file_formatter)
logger = logging.getLogger("Wiki Dataset Generation")
logger.addHandler(file_handler)
return logger
#wrapper fn of text-cleansing
def text_cleansing_wrapper(fn, exception_class_names = []):
#ensure caught exception class names passed to decorator is a list (if provided)
if not isinstance(exception_class_names, list):
raise TypeError("Exception Class Name for Wrapper is not a list!")
#ensure all values of caught exception class name list is a string
if not all([isinstance(val, str) for val in exception_class_names]):
raise ValueError("Found an element of Exception Class Name for Wrapper that is not a string!")
#lowercase all exception class name
exception_class_names = [val.lower() for val in exception_class_names]
if len(exception_class_names) == 0:
warnings.warn("The wrapper receives 0 `exception_class_names` to be warned! Will return the function value with its input!")
def text_fn_wrapper(text: str):
try:
return fn(text)
except Exception as e:
_exc_name = type(e).__name__
if _exc_name.lower() not in exception_class_names and len(exception_class_names)>0:
raise Exception(f"Exception Occured of {_exc_name}!") from e
else:
_followup_msg = "Returning the input as it is..."
_text_warn = f"An exception of {_exc_name} occured! {_followup_msg}"
warnings.warn(_text_warn)
return text
return text_fn_wrapper
#create html tags cleanser of a given text
partial_decorator = partial(text_cleansing_wrapper, exception_class_names=["parseerror"])
@partial_decorator
def remove_html_tags(text: str):
#extracted from "https://stackoverflow.com/a/9662410", w/ additional decorator of error handler
return ''.join(ET.fromstring(text).itertext())
#create non-ascii removal of text
@text_cleansing_wrapper
def decode_url_and_remove_non_ascii(text: str):
# return (urllib.parse.unquote(text)).encode('utf8', errors='ignore').decode().strip()
return (urllib.parse.unquote(text)).encode('ascii', errors='ignore').decode().strip()
#create excessive whitespace removal of text
@text_cleansing_wrapper
def remove_excessive_whitespace(text: str):
return re.sub("(\s)(\s+)", "\1", text).strip()
#create non-alphanumeric removal of text
@text_cleansing_wrapper
def remove_non_alphanumeric(text: str):
return re.sub("[^a-z0-9\s]", "", text, flags=re.I).strip()
def cleanse_wiki_text(text: str):
return remove_html_tags(decode_url_and_remove_non_ascii(text))
def normalize_wiki_title(text: str):
return remove_non_alphanumeric(remove_excessive_whitespace(text.lower()))
### MAIN CODE ###
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("--raw-csv-path", help="Relative location of csv file containing raw Wikipedia data")
parser.add_argument("--drop-hard-dupl", help="""Flag whether to drop hard duplicates
(exact values of data of relevant text fields, Titles & Desc)""",
default=True, type=argparse_bool_check)
parser.add_argument("--drop-soft-dupl", help="""Flag whether to drop soft duplicates
(duplicates after cleansed and normalized relevant text fields, Titles & Desc)""",
default=True, type=argparse_bool_check)
parser.add_argument("--save-dir-path", help="""Relative dir path of saved Wikipedia CSV data
to the `cleanse_wiki_data.py` script dir""",
default=os.path.dirname(os.path.abspath(__file__)))
args = parser.parse_args()
expected_colnames = ["id", "url", "title", "text"]
logger = set_logger()
logger.info("Parsing arguments...")
raw_data_path = args.raw_csv_path
drop_hard_dupl = args.drop_hard_dupl
drop_soft_dupl = args.drop_soft_dupl
save_dir = args.save_dir_path
df = pd.read_csv(raw_data_path)
if len(set(df.columns).difference(set(expected_colnames))) != 0 or len(set(expected_colnames).difference(set(df.columns))) != 0:
raise ValueError(f"The data schema expected, consist of columns: {', '.join(df.columns.to_list())} doesn't match with expected column values of {', '.join(expected_colnames)}!")
if (not drop_hard_dupl) and (not drop_soft_dupl):
raise AssertionError("The script won't run with both `drop-hard-dupl` and `drop-soft-dupl` args turned off!")
elif (not drop_hard_dupl):
warnings.warn("The args of `drop_hard_dupl` isn't turned off! Possibly the data will contain one template value of Wikipedia (usually no contribution text!)")
#will save id identifier colname first (popping first list val)
id_colname = expected_colnames.pop(0)
# if any of the data has duplicate values from columns checked (url, title, or text),
# it means the data integrity is questionable
# i.e. copied from other article or filled with template text
# hence, we will delete those duplicated datasets
suffix_file = "_dedup_cleansed"
#hard duplicate drop (drop all duplicate values that has exact same text on expected unique colnames)
if drop_hard_dupl:
for colname in expected_colnames:
logger.info(f"Checking data integrity on column {colname} on removing hard-duplicate(s)...")
dupl_text_df = df[df.duplicated(subset=colname,keep=False)]
shape_of_dupl_data = dupl_text_df.shape[0]
if shape_of_dupl_data > 0:
logger.info(f"Found {shape_of_dupl_data} data duplicated! Will be dropped")
df.drop_duplicates(subset=colname, keep=False, inplace=True)
#check id/idx of the cleansed data, whether it has duplicate
# (the duplication of id/idx should came from the very first extraction, not from the cleansing)
if df[df.duplicated(subset=id_colname,keep=False)].shape[0] > 0:
logger.info("Duplicated ID found! Re-assigning ID to the new ones based on `df.reset_index` method!")
df[id_colname] = df.reset_index().index
#soft duplicate drop (drop all except one duplicate values that has exact same text on expected unique colnames)
#keep the data that has longest value of its raw form
if drop_soft_dupl:
idx_to_keep = set(df.index.to_list())
#clean from text & title only, url isn't needed for this process
expected_colnames.remove("url")
for colname in expected_colnames:
logger.info(f"Checking data integrity on column {colname} on removing soft-duplicate(s)...")
_df = df.copy(deep=True)
#define text processing fn for soft-duplicate cleansing
#text processing for all colums (text & title)
_df = _df[[colname]]
_df[colname] = _df[colname].astype("str")
logger.info(f"Cleansing the data based on {colname}")
_df[colname+"_raw_len"] = _df[colname].apply(len)
_df[colname+"_cleansed"] = _df[colname].apply(cleanse_wiki_text)
if colname == "title":
# title text has been cleansed by `cleanse_wiki_text` fn, but needs to normalized on
# whitespaces, non alphanum syms (incl. punctuations), and case (all lowercased)
_df[colname+"_cleansed"] = _df[colname+"_cleansed"].apply(normalize_wiki_title)
#choose the data to keep by "ranking" it according to len of its raw text (greatest to keep)
logger.info(f"Ranking and grouping the data based on {colname}")
_df["rk"] = _df.groupby(colname+"_cleansed")[colname+"_raw_len"].rank(method="min", ascending=False)
shape_of_dupl_data = _df[_df["rk"]>1].shape[0]
if shape_of_dupl_data > 0:
logger.info(f"Found {shape_of_dupl_data} data duplicated! Will be dropped")
_idx_to_keep = _df[_df["rk"]==1].index.to_list()
if len(_idx_to_keep)+shape_of_dupl_data != df.shape[0]:
raise AssertionError("Mismatch of data number!")
idx_to_keep = idx_to_keep.intersection(set(_idx_to_keep))
del _df
gc.collect()
logger.info(f"The final data kept is {len(idx_to_keep)} from {df.shape[0]}")
df = df.loc[list(idx_to_keep),:]
logger.info("Saving dataset cleansed form...")
#input path splitted by ("/") for the last entry should return filename
#whereas the filename splitted by (".") except the last value should return the filename w/o ".csv" extension
_save_fn = ".".join(raw_data_path.split("/")[-1].split(".")[:-1]) + suffix_file + ".csv"
df.to_csv(f"{save_dir}/{_save_fn}", index=False)