:j`ddlmZddlmZddlmZmZmZddlZddlm Z m Z dee de defd Z d e dee fd Zd efd ee deee de de de de de dee gee fdee e ffdZde de de de de deedede fdZ dd ee ee fdeee ee fde dedeeede f dZy))Counter)Sequence)CallableOptionalUnionN)Tensortensorngram_input_listn_gramreturnct}td|dzD]?}tt||z dzD] }t||||z}||xxdz cc<"A|S)a Count how many times each word appears in a given text with ngram. Args: ngram_input_list: A list of translated text or reference texts n_gram: gram value ranged 1 to 4 Return: ngram_counter: a collections.Counter object of ngram )rrangelentuple)r r ngram_counterij ngram_keys v/home/mcse/projects/srt_converter/srt-converter-venv/lib/python3.12/site-packages/torchmetrics/functional/text/bleu.py _count_ngramrsw%YM 1fqj !*s+,q0145 *A.qAE;> rpredstarget numerator denominator preds_len target_len tokenizerc |D cgc]}|D cgc]} | r|| ngc} } }} |D cgc]} | r|| ng} } t| | D]\} } |t| z }| Dcgc] }t|}}|Dcgc]}tt| |z }}|||jt |z }t | |}t }| D]}|t ||z}||z}|D]}|t|dz xx||z cc< |D]}|t|dz xx||z cc< ||fScc} wcc} }wcc} wcc}wcc}w)aUpdate and returns variables required to compute the BLEU score. Args: preds: An iterable of machine translated corpus target: An iterable of iterables of reference corpus numerator: Numerator of precision score (true positives) denominator: Denominator of precision score (true positives + false positives) preds_len: count of words in a candidate prediction target_len: count of words in a reference translation target: count of words in a reference translation n_gram: gram value ranged 1 to 4 tokenizer: A function that turns sentence into list of words r)ziprabsindexminrr)rrr r!r"r#r r$tlinetarget_preds_predtargetstgttarget_len_listxtarget_len_diff preds_countertarget_counterngram_counter_clip counter_clipcounters r_bleu_score_updater9=s0lr1rfg_`2aW[d9T?PR3R2a1rG1rSX&Y4$yB'F&YF&YVW-D gSY /673s8777FG!3s4y1}-GGoo&;&;Cr)r torchdivaddoneslogexpsum) r"r#r r!r r:r;r>precision_scoreslog_precision_scoresgeometric_meanbrevity_penaltys r_bleu_score_computerKns*  F 9~c&))  99 IIiF6!B C IIk5::fV#D E (l[^;${2!'&9EIIFV?N4= 4JfS0PUPYPYZ[_ilu_uZvPwO ^ ++rc ,t|tr|gn|}|Dcgc]}t|tr|gn|}}t|t|k7r#tdt|dt||(t||k7rtdt|d|| d|z g|z}t j |}t j |} t d} t d} t|||| | | |t\} } t| | || |||Scc}w)a3Calculate `BLEU score`_ of machine translated text with one or more references. Args: preds: An iterable of machine translated corpus target: An iterable of iterables of reference corpus n_gram: Gram value ranged from 1 to 4 smooth: Whether to apply smoothing - see [2] weights: Weights used for unigrams, bigrams, etc. to calculate BLEU score. If not provided, uniform weights are used. Return: Tensor with BLEU Score Raises: ValueError: If ``preds`` and ``target`` corpus have different lengths. ValueError: If a length of a list of weights is not ``None`` and not equal to ``n_gram``. Example: >>> from torchmetrics.functional.text import bleu_score >>> preds = ['the cat is on the mat'] >>> target = [['there is a cat on the mat', 'a cat is on the mat']] >>> bleu_score(preds, target) tensor(0.7598) References: [1] BLEU: a Method for Automatic Evaluation of Machine Translation by Papineni, Kishore, Salim Roukos, Todd Ward, and Wei-Jing Zhu `BLEU`_ [2] Automatic Evaluation of Machine Translation Quality Using Longest Common Subsequence and Skip-Bigram Statistics by Chin-Yew Lin and Franz Josef Och `Machine Translation Evolution`_ zCorpus has different size z != z5List of weights has different weights than `n_gram`: r?r=) isinstancestrr ValueErrorr@zerosr r9rrK) rrr r;r:r-r0r,r r!r"r#s r bleu_scorerQs%P#5#.eWEFAGH# 3,u#5HGH 6{c'l"5c&k]$s7|nUVVs7|v5PQTU\Q]P^^bcibjkll<.6) F#I++f%Ks IJ.KJP\Iz y*ifV]_e ff'IsD)rFN) collectionsrcollections.abcrtypingrrrr@rr rNintrrrr9floatboolrKrQrrrYs& $,, 8C=#'* 3 8C= (0<.! C=.! Xc] #.!.! .!  .!  .! .!x},-.! 66>.!b%,%,%,%, %,  %, e_ %, %, %,V)- <g hsm# $<g U3 -. /<g <g  <g huo & <g  <gr