1
0
mirror of https://github.com/ilri/csv-metadata-quality.git synced 2024-11-10 16:25:45 +01:00
csv-metadata-quality/csv_metadata_quality/app.py
Alan Orth 0dc66c5c4e
Expand check/fix for multi-value separators
I just came across some metadata that had unnecessary multi-value
separators at the end of a field, causing a blank value to be used.

For example: "Kenya||Tanzania||"
2021-01-03 15:30:03 +02:00

178 lines
6.0 KiB
Python

import argparse
import re
import signal
import sys
import pandas as pd
import csv_metadata_quality.check as check
import csv_metadata_quality.experimental as experimental
import csv_metadata_quality.fix as fix
from csv_metadata_quality.version import VERSION
def parse_args(argv):
parser = argparse.ArgumentParser(description="Metadata quality checker and fixer.")
parser.add_argument(
"--agrovoc-fields",
"-a",
help="Comma-separated list of fields to validate against AGROVOC, for example: dc.subject,cg.coverage.country",
)
parser.add_argument(
"--experimental-checks",
"-e",
help="Enable experimental checks like language detection",
action="store_true",
)
parser.add_argument(
"--input-file",
"-i",
help="Path to input file. Can be UTF-8 CSV or Excel XLSX.",
required=True,
type=argparse.FileType("r", encoding="UTF-8"),
)
parser.add_argument(
"--output-file",
"-o",
help="Path to output file (always CSV).",
required=True,
type=argparse.FileType("w", encoding="UTF-8"),
)
parser.add_argument(
"--unsafe-fixes", "-u", help="Perform unsafe fixes.", action="store_true"
)
parser.add_argument(
"--version", "-V", action="version", version=f"CSV Metadata Quality v{VERSION}"
)
parser.add_argument(
"--exclude-fields",
"-x",
help="Comma-separated list of fields to skip, for example: dc.contributor.author,dc.identifier.citation",
)
args = parser.parse_args()
return args
def signal_handler(signal, frame):
sys.exit(1)
def run(argv):
args = parse_args(argv)
# set the signal handler for SIGINT (^C)
signal.signal(signal.SIGINT, signal_handler)
# Read all fields as strings so dates don't get converted from 1998 to 1998.0
df = pd.read_csv(args.input_file, dtype=str)
for column in df.columns:
# Check if the user requested to skip any fields
if args.exclude_fields:
skip = False
# Split the list of excludes on ',' so we can test exact matches
# rather than fuzzy matches with regexes or "if word in string"
for exclude in args.exclude_fields.split(","):
if column == exclude and skip is False:
skip = True
if skip:
print(f"Skipping {column}")
continue
# Fix: whitespace
df[column] = df[column].apply(fix.whitespace, field_name=column)
# Fix: newlines
if args.unsafe_fixes:
df[column] = df[column].apply(fix.newlines)
# Fix: missing space after comma. Only run on author and citation
# fields for now, as this problem is mostly an issue in names.
if args.unsafe_fixes:
match = re.match(r"^.*?(author|citation).*$", column)
if match is not None:
df[column] = df[column].apply(fix.comma_space, field_name=column)
# Fix: perform Unicode normalization (NFC) to convert decomposed
# characters into their canonical forms.
if args.unsafe_fixes:
df[column] = df[column].apply(fix.normalize_unicode, field_name=column)
# Fix: unnecessary Unicode
df[column] = df[column].apply(fix.unnecessary_unicode)
# Check: invalid and unnecessary multi-value separators
df[column] = df[column].apply(check.separators, field_name=column)
# Check: suspicious characters
df[column] = df[column].apply(check.suspicious_characters, field_name=column)
# Fix: invalid and unnecessary multi-value separators
if args.unsafe_fixes:
df[column] = df[column].apply(fix.separators, field_name=column)
# Run whitespace fix again after fixing invalid separators
df[column] = df[column].apply(fix.whitespace, field_name=column)
# Fix: duplicate metadata values
df[column] = df[column].apply(fix.duplicates, field_name=column)
# Check: invalid AGROVOC subject
if args.agrovoc_fields:
# Identify fields the user wants to validate against AGROVOC
for field in args.agrovoc_fields.split(","):
if column == field:
df[column] = df[column].apply(check.agrovoc, field_name=column)
# Check: invalid language
match = re.match(r"^.*?language.*$", column)
if match is not None:
df[column] = df[column].apply(check.language)
# Check: invalid ISSN
match = re.match(r"^.*?issn.*$", column)
if match is not None:
df[column] = df[column].apply(check.issn)
# Check: invalid ISBN
match = re.match(r"^.*?isbn.*$", column)
if match is not None:
df[column] = df[column].apply(check.isbn)
# Check: invalid date
match = re.match(r"^.*?date.*$", column)
if match is not None:
df[column] = df[column].apply(check.date, field_name=column)
# Check: filename extension
if column == "filename":
df[column] = df[column].apply(check.filename_extension)
##
# Perform some checks on rows so we can consider items as a whole rather
# than simple on a field-by-field basis. This allows us to check whether
# the language used in the title and abstract matches the language indi-
# cated in the language field, for example.
#
# This is slower and apparently frowned upon in the Pandas community be-
# cause it requires iterating over rows rather than using apply over a
# column. For now it will have to do.
##
if args.experimental_checks:
# Transpose the DataFrame so we can consider each row as a column
df_transposed = df.T
for column in df_transposed.columns:
experimental.correct_language(df_transposed[column])
# Write
df.to_csv(args.output_file, index=False)
# Close the input and output files before exiting
args.input_file.close()
args.output_file.close()
sys.exit(0)