Spaces:
Sleeping
Sleeping
File size: 9,334 Bytes
4a1df2e |
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 |
"""
AugmentCommand class
===========================
"""
from argparse import ArgumentDefaultsHelpFormatter, ArgumentError, ArgumentParser
import csv
import os
import time
import tqdm
import textattack
from textattack.augment_args import AUGMENTATION_RECIPE_NAMES
from textattack.commands import TextAttackCommand
class AugmentCommand(TextAttackCommand):
"""The TextAttack attack module:
A command line parser to run data augmentation from user
specifications.
"""
def run(self, args):
"""Reads in a CSV, performs augmentation, and outputs an augmented CSV.
Preserves all columns except for the input (augmneted) column.
"""
args = textattack.AugmenterArgs(**vars(args))
if args.interactive:
print("\nRunning in interactive mode...\n")
augmenter = eval(AUGMENTATION_RECIPE_NAMES[args.recipe])(
pct_words_to_swap=args.pct_words_to_swap,
transformations_per_example=args.transformations_per_example,
high_yield=args.high_yield,
fast_augment=args.fast_augment,
enable_advanced_metrics=args.enable_advanced_metrics,
)
print("--------------------------------------------------------")
while True:
print(
'\nEnter a sentence to augment, "q" to quit, "c" to view/change arguments:\n'
)
text = input()
if text == "q":
break
elif text == "c":
print(
f"\nCurrent Arguments:\n\n\t augmentation recipe: {args.recipe}, "
f"\n\t pct_words_to_swap: {args.pct_words_to_swap}, "
f"\n\t transformations_per_example: {args.transformations_per_example}\n"
)
change = input(
"Enter 'c' again to change arguments, any other keys to opt out\n"
)
if change == "c":
print("\nChanging augmenter arguments...\n")
recipe = input(
"\tAugmentation recipe name ('r' to see available recipes): "
)
if recipe == "r":
recipe_display = " ".join(AUGMENTATION_RECIPE_NAMES.keys())
print(f"\n\t{recipe_display}\n")
args.recipe = input("\tAugmentation recipe name: ")
else:
args.recipe = recipe
args.pct_words_to_swap = float(
input("\tPercentage of words to swap (0.0 ~ 1.0): ")
)
args.transformations_per_example = int(
input("\tTransformations per input example: ")
)
print("\nGenerating new augmenter...\n")
augmenter = eval(AUGMENTATION_RECIPE_NAMES[args.recipe])(
pct_words_to_swap=args.pct_words_to_swap,
transformations_per_example=args.transformations_per_example,
)
print(
"--------------------------------------------------------"
)
continue
elif not text:
continue
print("\nAugmenting...\n")
print("--------------------------------------------------------")
if args.enable_advanced_metrics:
results = augmenter.augment(text)
print("Augmentations:\n")
for augmentation in results[0]:
print(augmentation, "\n")
print()
print(
f"Average Original Perplexity Score: {results[1]['avg_original_perplexity']}"
)
print(
f"Average Augment Perplexity Score: {results[1]['avg_attack_perplexity']}"
)
print(
f"Average Augment USE Score: {results[2]['avg_attack_use_score']}\n"
)
else:
for augmentation in augmenter.augment(text):
print(augmentation, "\n")
print("--------------------------------------------------------")
else:
textattack.shared.utils.set_seed(args.random_seed)
start_time = time.time()
if not (args.input_csv and args.input_column and args.output_csv):
raise ArgumentError(
"The following arguments are required: --csv, --input-column/--i"
)
# Validate input/output paths.
if not os.path.exists(args.input_csv):
raise FileNotFoundError(f"Can't find CSV at location {args.input_csv}")
if os.path.exists(args.output_csv):
if args.overwrite:
textattack.shared.logger.info(
f"Preparing to overwrite {args.output_csv}."
)
else:
raise OSError(
f"Outfile {args.output_csv} exists and --overwrite not set."
)
# Read in CSV file as a list of dictionaries. Use the CSV sniffer to
# try and automatically infer the correct CSV format.
csv_file = open(args.input_csv, "r")
# mark where commas and quotes occur within the text value
def markQuotes(lines):
for row in lines:
row = row.replace('"', '"/')
yield row
dialect = csv.Sniffer().sniff(csv_file.readline(), delimiters=";,")
csv_file.seek(0)
rows = [
row
for row in csv.DictReader(
markQuotes(csv_file),
dialect=dialect,
skipinitialspace=True,
)
]
# replace markings with quotations and commas
for row in rows:
for item in row:
i = 0
while i < len(row[item]):
if row[item][i] == "/":
if row[item][i - 1] == '"':
row[item] = row[item][:i] + row[item][i + 1 :]
else:
row[item] = row[item][:i] + '"' + row[item][i + 1 :]
i += 1
# Validate input column.
row_keys = set(rows[0].keys())
if args.input_column not in row_keys:
raise ValueError(
f"Could not find input column {args.input_column} in CSV. Found keys: {row_keys}"
)
textattack.shared.logger.info(
f"Read {len(rows)} rows from {args.input_csv}. Found columns {row_keys}."
)
augmenter = eval(AUGMENTATION_RECIPE_NAMES[args.recipe])(
pct_words_to_swap=args.pct_words_to_swap,
transformations_per_example=args.transformations_per_example,
high_yield=args.high_yield,
fast_augment=args.fast_augment,
)
output_rows = []
for row in tqdm.tqdm(rows, desc="Augmenting rows"):
text_input = row[args.input_column]
if not args.exclude_original:
output_rows.append(row)
for augmentation in augmenter.augment(text_input):
augmented_row = row.copy()
augmented_row[args.input_column] = augmentation
output_rows.append(augmented_row)
# Print to file.
with open(args.output_csv, "w") as outfile:
csv_writer = csv.writer(
outfile, delimiter=",", quotechar="/", quoting=csv.QUOTE_MINIMAL
)
# Write header.
csv_writer.writerow(output_rows[0].keys())
# Write rows.
for row in output_rows:
csv_writer.writerow(row.values())
textattack.shared.logger.info(
f"Wrote {len(output_rows)} augmentations to {args.output_csv} in {time.time() - start_time}s."
)
# Remove extra markings in output file
with open(args.output_csv, "r") as file:
data = file.readlines()
for i in range(len(data)):
data[i] = data[i].replace("/", "")
with open(args.output_csv, "w") as file:
file.writelines(data)
@staticmethod
def register_subcommand(main_parser: ArgumentParser):
parser = main_parser.add_parser(
"augment",
help="augment text data",
formatter_class=ArgumentDefaultsHelpFormatter,
)
parser = textattack.AugmenterArgs._add_parser_args(parser)
parser.set_defaults(func=AugmentCommand())
|