File size: 25,907 Bytes
9c6594c |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 |
# Copyright The Lightning team.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# Copyright 2017 Maja Popovic
# The code is derived from https://github.com/m-popovic/chrF/blob/6d3c384/chrF%2B%2B.py
# The original author and copyright holder have agreed to relicense the derived code under the Apache License,
# Version 2.0 (the "License")
# Reference to the approval: https://github.com/Lightning-AI/torchmetrics/pull/2701#issuecomment-2316891785
from collections import defaultdict
from collections.abc import Sequence
from itertools import chain
from typing import List, Optional, Union
import torch
from torch import Tensor, tensor
from torchmetrics.functional.text.helper import _validate_inputs
_EPS_SMOOTHING = tensor(1e-16)
# Taken from https://github.com/mjpost/sacrebleu/blob/master/sacrebleu/metrics/chrf.py
_PUNCTUATIONS = set("!\"#$%&'()*+,-./:;<=>?@[\\]^_`{|}~")
def _prepare_n_grams_dicts(
n_char_order: int, n_word_order: int
) -> tuple[
dict[int, Tensor], dict[int, Tensor], dict[int, Tensor], dict[int, Tensor], dict[int, Tensor], dict[int, Tensor]
]:
"""Prepare dictionaries with default zero values for total ref, hypothesis and matching character and word n-grams.
Args:
n_char_order: A character n-gram order.
n_word_order: A word n-gram order.
Return:
Dictionaries with default zero values for total reference, hypothesis and matching character and word
n-grams.
"""
total_preds_char_n_grams: dict[int, Tensor] = {n + 1: tensor(0.0) for n in range(n_char_order)}
total_preds_word_n_grams: dict[int, Tensor] = {n + 1: tensor(0.0) for n in range(n_word_order)}
total_target_char_n_grams: dict[int, Tensor] = {n + 1: tensor(0.0) for n in range(n_char_order)}
total_target_word_n_grams: dict[int, Tensor] = {n + 1: tensor(0.0) for n in range(n_word_order)}
total_matching_char_n_grams: dict[int, Tensor] = {n + 1: tensor(0.0) for n in range(n_char_order)}
total_matching_word_n_grams: dict[int, Tensor] = {n + 1: tensor(0.0) for n in range(n_word_order)}
return (
total_preds_char_n_grams,
total_preds_word_n_grams,
total_target_char_n_grams,
total_target_word_n_grams,
total_matching_char_n_grams,
total_matching_word_n_grams,
)
def _get_characters(sentence: str, whitespace: bool) -> list[str]:
"""Split sentence into individual characters.
Args:
sentence: An input sentence to split.
whitespace: An indication whether to keep whitespaces during character n-gram extraction.
Return:
A list of separated characters.
"""
if whitespace:
return list(sentence)
return list(sentence.strip().replace(" ", ""))
def _separate_word_and_punctuation(word: str) -> list[str]:
"""Separates out punctuations from beginning and end of words for chrF.
Adapted from https://github.com/m-popovic/chrF and
https://github.com/mjpost/sacrebleu/blob/master/sacrebleu/metrics/chrf.py.
Args:
word: An input word to be separated from a punctuation if present.
Return:
A list of a single word or a separated word and punctuation.
"""
if len(word) == 1:
return [word]
if word[-1] in _PUNCTUATIONS:
return [word[:-1], word[-1]]
if word[0] in _PUNCTUATIONS:
return [word[0], word[1:]]
return [word]
def _get_words_and_punctuation(sentence: str) -> list[str]:
"""Separates out punctuations from beginning and end of words for chrF for all words in the sentence.
Args:
sentence: An input sentence to split
Return:
An aggregated list of separated words and punctuations.
"""
return list(chain.from_iterable(_separate_word_and_punctuation(word) for word in sentence.strip().split()))
def _ngram_counts(char_or_word_list: list[str], n_gram_order: int) -> dict[int, dict[tuple[str, ...], Tensor]]:
"""Calculate n-gram counts.
Args:
char_or_word_list: A list of characters of words
n_gram_order: The largest number of n-gram.
Return:
A dictionary of dictionaries with a counts of given n-grams.
"""
ngrams: dict[int, dict[tuple[str, ...], Tensor]] = defaultdict(lambda: defaultdict(lambda: tensor(0.0)))
for n in range(1, n_gram_order + 1):
for ngram in (tuple(char_or_word_list[i : i + n]) for i in range(len(char_or_word_list) - n + 1)):
ngrams[n][ngram] += tensor(1)
return ngrams
def _get_n_grams_counts_and_total_ngrams(
sentence: str, n_char_order: int, n_word_order: int, lowercase: bool, whitespace: bool
) -> tuple[
dict[int, dict[tuple[str, ...], Tensor]],
dict[int, dict[tuple[str, ...], Tensor]],
dict[int, Tensor],
dict[int, Tensor],
]:
"""Get n-grams and total n-grams.
Args:
sentence: An input sentence
n_char_order: A character n-gram order.
n_word_order: A word n-gram order.
lowercase: An indication whether to enable case-insensitivity.
whitespace: An indication whether to keep whitespaces during character n-gram extraction.
Return:
char_n_grams_counts: A dictionary of dictionaries with sentence character n-grams.
word_n_grams_counts: A dictionary of dictionaries with sentence word n-grams.
total_char_n_grams: A dictionary containing a total number of sentence character n-grams.
total_word_n_grams: A dictionary containing a total number of sentence word n-grams.
"""
def _char_and_word_ngrams_counts(
sentence: str, n_char_order: int, n_word_order: int, lowercase: bool
) -> tuple[dict[int, dict[tuple[str, ...], Tensor]], dict[int, dict[tuple[str, ...], Tensor]]]:
"""Get a dictionary of dictionaries with a counts of given n-grams."""
if lowercase:
sentence = sentence.lower()
char_n_grams_counts = _ngram_counts(_get_characters(sentence, whitespace), n_char_order)
word_n_grams_counts = _ngram_counts(_get_words_and_punctuation(sentence), n_word_order)
return char_n_grams_counts, word_n_grams_counts
def _get_total_ngrams(n_grams_counts: dict[int, dict[tuple[str, ...], Tensor]]) -> dict[int, Tensor]:
"""Get total sum of n-grams over n-grams w.r.t n."""
total_n_grams: dict[int, Tensor] = defaultdict(lambda: tensor(0.0))
for n in n_grams_counts:
total_n_grams[n] = sum(n_grams_counts[n].values()).detach().clone() # type: ignore
return total_n_grams
char_n_grams_counts, word_n_grams_counts = _char_and_word_ngrams_counts(
sentence, n_char_order, n_word_order, lowercase
)
total_char_n_grams = _get_total_ngrams(char_n_grams_counts)
total_word_n_grams = _get_total_ngrams(word_n_grams_counts)
return char_n_grams_counts, word_n_grams_counts, total_char_n_grams, total_word_n_grams
def _get_ngram_matches(
hyp_n_grams_counts: dict[int, dict[tuple[str, ...], Tensor]],
ref_n_grams_counts: dict[int, dict[tuple[str, ...], Tensor]],
) -> dict[int, Tensor]:
"""Get a number of n-gram matches between reference and hypothesis n-grams.
Args:
hyp_n_grams_counts: n-grams counts for hypothesis
ref_n_grams_counts: n-grams counts for reference
Return:
matching_n_grams
"""
matching_n_grams: dict[int, Tensor] = defaultdict(lambda: tensor(0.0))
for n in hyp_n_grams_counts:
min_n_grams = [
torch.min(ref_n_grams_counts[n][n_gram], hyp_n_grams_counts[n][n_gram]) for n_gram in hyp_n_grams_counts[n]
]
matching_n_grams[n] = sum(min_n_grams).detach().clone() # type: ignore
return matching_n_grams
def _sum_over_dicts(total_n_grams: dict[int, Tensor], n_grams: dict[int, Tensor]) -> dict[int, Tensor]:
"""Aggregate total n-grams to keep corpus-level statistics.
Args:
total_n_grams: A dictionary containing a total corpus-level number of n-grams.
n_grams: A dictionary containing a sentence-level number of n-grams.
Return:
A dictionary containing a total corpus-level number of n-grams.
"""
for n in n_grams:
total_n_grams[n] += n_grams[n]
return total_n_grams
def _calculate_fscore(
matching_char_n_grams: dict[int, Tensor],
matching_word_n_grams: dict[int, Tensor],
hyp_char_n_grams: dict[int, Tensor],
hyp_word_n_grams: dict[int, Tensor],
ref_char_n_grams: dict[int, Tensor],
ref_word_n_grams: dict[int, Tensor],
n_order: float,
beta: float,
) -> Tensor:
"""Calculate sentence-level chrF/chrF++ score.
For given hypothesis and reference statistics (either sentence-level or corpus-level)
the chrF/chrF++ score is returned.
Args:
matching_char_n_grams:
A total number of matching character n-grams between the best matching reference and hypothesis.
matching_word_n_grams:
A total number of matching word n-grams between the best matching reference and hypothesis.
hyp_char_n_grams: A total number of hypothesis character n-grams.
hyp_word_n_grams: A total number of hypothesis word n-grams.
ref_char_n_grams: A total number of reference character n-grams.
ref_word_n_grams: A total number of reference word n-grams.
n_order: A sum of character and word n-gram order.
beta: A parameter determining an importance of recall w.r.t. precision. If `beta=1`, their importance is equal.
Return:
A chrF/chrF++ score. This function is universal both for sentence-level and corpus-level calculation.
"""
def _get_n_gram_fscore(
matching_n_grams: dict[int, Tensor], ref_n_grams: dict[int, Tensor], hyp_n_grams: dict[int, Tensor], beta: float
) -> dict[int, Tensor]:
"""Get n-gram level f-score."""
precision: dict[int, Tensor] = {
n: matching_n_grams[n] / hyp_n_grams[n] if hyp_n_grams[n] > 0 else tensor(0.0) for n in matching_n_grams
}
recall: dict[int, Tensor] = {
n: matching_n_grams[n] / ref_n_grams[n] if ref_n_grams[n] > 0 else tensor(0.0) for n in matching_n_grams
}
denominator: dict[int, Tensor] = {
n: torch.max(beta**2 * precision[n] + recall[n], _EPS_SMOOTHING) for n in matching_n_grams
}
f_score: dict[int, Tensor] = {
n: (1 + beta**2) * precision[n] * recall[n] / denominator[n] for n in matching_n_grams
}
return f_score
char_n_gram_f_score = _get_n_gram_fscore(matching_char_n_grams, ref_char_n_grams, hyp_char_n_grams, beta)
word_n_gram_f_score = _get_n_gram_fscore(matching_word_n_grams, ref_word_n_grams, hyp_word_n_grams, beta)
return (sum(char_n_gram_f_score.values()) + sum(word_n_gram_f_score.values())) / tensor(n_order)
def _calculate_sentence_level_chrf_score(
targets: list[str],
pred_char_n_grams_counts: dict[int, dict[tuple[str, ...], Tensor]],
pred_word_n_grams_counts: dict[int, dict[tuple[str, ...], Tensor]],
pred_char_n_grams: dict[int, Tensor],
pred_word_n_grams: dict[int, Tensor],
n_char_order: int,
n_word_order: int,
n_order: float,
beta: float,
lowercase: bool,
whitespace: bool,
) -> tuple[Tensor, dict[int, Tensor], dict[int, Tensor], dict[int, Tensor], dict[int, Tensor]]:
"""Calculate the best sentence-level chrF/chrF++ score.
For a given pre-processed hypothesis, all references are evaluated and score and statistics
for the best matching reference is returned.
Args:
targets: An iterable of references.
pred_char_n_grams_counts: A dictionary of dictionaries with hypothesis character n-grams.
pred_word_n_grams_counts: A dictionary of dictionaries with hypothesis word n-grams.
pred_char_n_grams: A total number of hypothesis character n-grams.
pred_word_n_grams: A total number of hypothesis word n-grams.
n_char_order: A character n-gram order.
n_word_order: A word n-gram order.
n_order: A sum of character and word n-gram order.
beta: A parameter determining an importance of recall w.r.t. precision. If `beta=1`, their importance is equal.
lowercase: An indication whether to enable case-insensitivity.
whitespace: An indication whether to keep whitespaces during character n-gram extraction.
Return:
Return chrF/chrF++ score and statistics for the best matching hypothesis and reference.
f_score: A sentence-level chrF/chrF++ score.
matching_char_n_grams:
A total number of matching character n-grams between the best matching reference and hypothesis.
matching_word_n_grams:
A total number of matching word n-grams between the best matching reference and hypothesis.
target_char_n_grams: A total number of reference character n-grams.
target_word_n_grams: A total number of reference word n-grams.
"""
best_f_score = tensor(0.0)
best_matching_char_n_grams: dict[int, Tensor] = defaultdict(lambda: tensor(0.0))
best_matching_word_n_grams: dict[int, Tensor] = defaultdict(lambda: tensor(0.0))
best_target_char_n_grams: dict[int, Tensor] = defaultdict(lambda: tensor(0.0))
best_target_word_n_grams: dict[int, Tensor] = defaultdict(lambda: tensor(0.0))
for target in targets:
(
target_char_n_grams_counts,
target_word_n_grams_counts,
target_char_n_grams,
target_word_n_grams,
) = _get_n_grams_counts_and_total_ngrams(target, n_char_order, n_word_order, lowercase, whitespace)
matching_char_n_grams = _get_ngram_matches(target_char_n_grams_counts, pred_char_n_grams_counts)
matching_word_n_grams = _get_ngram_matches(target_word_n_grams_counts, pred_word_n_grams_counts)
f_score = _calculate_fscore(
matching_char_n_grams,
matching_word_n_grams,
pred_char_n_grams,
pred_word_n_grams,
target_char_n_grams,
target_word_n_grams,
n_order,
beta,
)
if f_score > best_f_score:
best_f_score = f_score
best_matching_char_n_grams = matching_char_n_grams
best_matching_word_n_grams = matching_word_n_grams
best_target_char_n_grams = target_char_n_grams
best_target_word_n_grams = target_word_n_grams
return (
best_f_score,
best_matching_char_n_grams,
best_matching_word_n_grams,
best_target_char_n_grams,
best_target_word_n_grams,
)
def _chrf_score_update(
preds: Union[str, Sequence[str]],
target: Union[Sequence[str], Sequence[Sequence[str]]],
total_preds_char_n_grams: dict[int, Tensor],
total_preds_word_n_grams: dict[int, Tensor],
total_target_char_n_grams: dict[int, Tensor],
total_target_word_n_grams: dict[int, Tensor],
total_matching_char_n_grams: dict[int, Tensor],
total_matching_word_n_grams: dict[int, Tensor],
n_char_order: int,
n_word_order: int,
n_order: float,
beta: float,
lowercase: bool,
whitespace: bool,
sentence_chrf_score: Optional[List[Tensor]] = None,
) -> tuple[
dict[int, Tensor],
dict[int, Tensor],
dict[int, Tensor],
dict[int, Tensor],
dict[int, Tensor],
dict[int, Tensor],
Optional[List[Tensor]],
]:
"""Update function for chrf score.
Args:
preds: An iterable of hypothesis corpus.
target: An iterable of iterables of reference corpus.
total_preds_char_n_grams: A dictionary containing a total number of hypothesis character n-grams.
total_preds_word_n_grams: A dictionary containing a total number of hypothesis word n-grams.
total_target_char_n_grams: A dictionary containing a total number of reference character n-grams.
total_target_word_n_grams: A dictionary containing a total number of reference word n-grams.
total_matching_char_n_grams:
A dictionary containing a total number of matching character n-grams between references and hypotheses.
total_matching_word_n_grams:
A dictionary containing a total number of total matching word n-grams between references and hypotheses.
n_char_order: A character n-gram order.
n_word_order: A word n-gram order.
n_order: Sum of character and word n-gram order.
beta: A parameter determining an importance of recall w.r.t. precision. If `beta=1`, their importance is equal.
lowercase: An indication whether to enable case-insensitivity.
whitespace: An indication whether to keep whitespaces during character n-gram extraction.
sentence_chrf_score: A list of sentence-level chrF/chrF++ scores.
Return:
total_target_char_n_grams: number of reference character n-grams.
total_target_word_n_grams: number of reference word n-grams.
total_preds_char_n_grams: number of hypothesis character n-grams.
total_preds_word_n_grams: number of hypothesis word n-grams.
total_matching_char_n_grams: number of matching character n-grams between references and hypotheses.
total_matching_word_n_grams: number of total matching word n-grams between references and hypotheses.
sentence_chrf_score: A list of sentence-level chrF/chrF++ scores.
Raises:
ValueError:
If length of ``preds`` and ``target`` differs.
"""
target_corpus, preds = _validate_inputs(target, preds)
for pred, targets in zip(preds, target_corpus):
(
pred_char_n_grams_counts,
pred_word_n_grams_counts,
pred_char_n_grams,
pred_word_n_grams,
) = _get_n_grams_counts_and_total_ngrams(pred, n_char_order, n_word_order, lowercase, whitespace)
total_preds_char_n_grams = _sum_over_dicts(total_preds_char_n_grams, pred_char_n_grams)
total_preds_word_n_grams = _sum_over_dicts(total_preds_word_n_grams, pred_word_n_grams)
(
sentence_level_f_score,
matching_char_n_grams,
matching_word_n_grams,
target_char_n_grams,
target_word_n_grams,
) = _calculate_sentence_level_chrf_score(
targets, # type: ignore
pred_char_n_grams_counts,
pred_word_n_grams_counts,
pred_char_n_grams,
pred_word_n_grams,
n_char_order,
n_word_order,
n_order,
beta,
lowercase,
whitespace,
)
if sentence_chrf_score is not None:
sentence_chrf_score.append(sentence_level_f_score.unsqueeze(0))
total_target_char_n_grams = _sum_over_dicts(total_target_char_n_grams, target_char_n_grams)
total_target_word_n_grams = _sum_over_dicts(total_target_word_n_grams, target_word_n_grams)
total_matching_char_n_grams = _sum_over_dicts(total_matching_char_n_grams, matching_char_n_grams)
total_matching_word_n_grams = _sum_over_dicts(total_matching_word_n_grams, matching_word_n_grams)
return (
total_preds_char_n_grams,
total_preds_word_n_grams,
total_target_char_n_grams,
total_target_word_n_grams,
total_matching_char_n_grams,
total_matching_word_n_grams,
sentence_chrf_score,
)
def _chrf_score_compute(
total_preds_char_n_grams: dict[int, Tensor],
total_preds_word_n_grams: dict[int, Tensor],
total_target_char_n_grams: dict[int, Tensor],
total_target_word_n_grams: dict[int, Tensor],
total_matching_char_n_grams: dict[int, Tensor],
total_matching_word_n_grams: dict[int, Tensor],
n_order: float,
beta: float,
) -> Tensor:
"""Compute chrF/chrF++ score based on pre-computed target, prediction and matching character and word n-grams.
Args:
total_preds_char_n_grams: number of hypothesis character n-grams.
total_preds_word_n_grams: number of hypothesis word n-grams.
total_target_char_n_grams: number of reference character n-grams.
total_target_word_n_grams: number of reference word n-grams.
total_matching_char_n_grams: number of matching character n-grams between references and hypotheses.
total_matching_word_n_grams: number of total matching word n-grams between references and hypotheses.
n_order: A sum of character and word n-gram order.
beta:
A parameter determining an importance of recall w.r.t. precision. If `beta=1`, their importance is equal.
Return:
A corpus-level chrF/chrF++ score.
"""
return _calculate_fscore(
total_matching_char_n_grams,
total_matching_word_n_grams,
total_preds_char_n_grams,
total_preds_word_n_grams,
total_target_char_n_grams,
total_target_word_n_grams,
n_order,
beta,
)
def chrf_score(
preds: Union[str, Sequence[str]],
target: Sequence[Union[str, Sequence[str]]],
n_char_order: int = 6,
n_word_order: int = 2,
beta: float = 2.0,
lowercase: bool = False,
whitespace: bool = False,
return_sentence_level_score: bool = False,
) -> Union[Tensor, tuple[Tensor, Tensor]]:
"""Calculate `chrF score`_ of machine translated text with one or more references.
This implementation supports both chrF score computation introduced in [1] and chrF++ score introduced in
`chrF++ score`_. This implementation follows the implementations from https://github.com/m-popovic/chrF and
https://github.com/mjpost/sacrebleu/blob/master/sacrebleu/metrics/chrf.py.
Args:
preds: An iterable of hypothesis corpus.
target: An iterable of iterables of reference corpus.
n_char_order:
A character n-gram order. If `n_char_order=6`, the metrics refers to the official chrF/chrF++.
n_word_order:
A word n-gram order. If `n_word_order=2`, the metric refers to the official chrF++. If `n_word_order=0`, the
metric is equivalent to the original chrF.
beta:
A parameter determining an importance of recall w.r.t. precision. If `beta=1`, their importance is equal.
lowercase: An indication whether to enable case-insensitivity.
whitespace: An indication whether to keep whitespaces during character n-gram extraction.
return_sentence_level_score: An indication whether a sentence-level chrF/chrF++ score to be returned.
Return:
A corpus-level chrF/chrF++ score.
(Optionally) A list of sentence-level chrF/chrF++ scores if `return_sentence_level_score=True`.
Raises:
ValueError:
If ``n_char_order`` is not an integer greater than or equal to 1.
ValueError:
If ``n_word_order`` is not an integer greater than or equal to 0.
ValueError:
If ``beta`` is smaller than 0.
Example:
>>> from torchmetrics.functional.text import chrf_score
>>> preds = ['the cat is on the mat']
>>> target = [['there is a cat on the mat', 'a cat is on the mat']]
>>> chrf_score(preds, target)
tensor(0.8640)
References:
[1] chrF: character n-gram F-score for automatic MT evaluation by Maja PopoviΔ `chrF score`_
[2] chrF++: words helping character n-grams by Maja PopoviΔ `chrF++ score`_
"""
if not isinstance(n_char_order, int) or n_char_order < 1:
raise ValueError("Expected argument `n_char_order` to be an integer greater than or equal to 1.")
if not isinstance(n_word_order, int) or n_word_order < 0:
raise ValueError("Expected argument `n_word_order` to be an integer greater than or equal to 0.")
if beta < 0:
raise ValueError("Expected argument `beta` to be greater than 0.")
n_order = float(n_char_order + n_word_order)
(
total_preds_char_n_grams,
total_preds_word_n_grams,
total_target_char_n_grams,
total_target_word_n_grams,
total_matching_char_n_grams,
total_matching_word_n_grams,
) = _prepare_n_grams_dicts(n_char_order, n_word_order)
sentence_chrf_score: Optional[List[Tensor]] = [] if return_sentence_level_score else None
(
total_preds_char_n_grams,
total_preds_word_n_grams,
total_target_char_n_grams,
total_target_word_n_grams,
total_matching_char_n_grams,
total_matching_word_n_grams,
sentence_chrf_score,
) = _chrf_score_update(
preds,
target,
total_preds_char_n_grams,
total_preds_word_n_grams,
total_target_char_n_grams,
total_target_word_n_grams,
total_matching_char_n_grams,
total_matching_word_n_grams,
n_char_order,
n_word_order,
n_order,
beta,
lowercase,
whitespace,
sentence_chrf_score,
)
chrf_f_score = _chrf_score_compute(
total_preds_char_n_grams,
total_preds_word_n_grams,
total_target_char_n_grams,
total_target_word_n_grams,
total_matching_char_n_grams,
total_matching_word_n_grams,
n_order,
beta,
)
if sentence_chrf_score:
return chrf_f_score, torch.cat(sentence_chrf_score)
return chrf_f_score
|