Ë þÍ:j÷Pãó—UddlZddlmZddlmZddlmZmZmZm Z m Z ddlZddlmZm Z ddlmZddlmZdd giZd ddd ddddddddœZeee eeffed<dZd=d„Zdedeefd„Zdedededeeeffd„Z d>d eed!eed"ede eeeeffd#„Zd$eeed eed!eedeefd%„Zd&eeed!eedeefd'„Z d?d(ed)e ed*e eegefd+e eegeefdeef d,„Z d-eed.eed/edeeeffd0„Z!d-eed.eedeeeffd1„Z"d-eeed.eeedeeeffd2„Z# d?d3eed.eeed4e$e eefd5ed)e ed*e eegefd+e eegeefdee eefe$eeefffd6„Z%d7eeeefdeeeffd8„Z& d@d3e eeefd.e eeeeeefd5edd9ed*e eegefd+e eegeefd:e ee'ed;ffdeeeffd<„Z(y)AéN)ÚCounter)ÚSequence)ÚAnyÚCallableÚListÚOptionalÚUnion)ÚTensorÚtensor)ÚLiteral)Ú_NLTK_AVAILABLE)Úrouge_scoreÚ_rouge_score_updateÚnltkééééééééé ÚLÚLsum)Úrouge1Úrouge2Úrouge3Úrouge4Úrouge5Úrouge6Úrouge7Úrouge8Úrouge9ÚrougeLÚ rougeLsumÚALLOWED_ROUGE_KEYS)ÚavgÚbestÚreturncóÊ—ddl} |jjd«y#t$r7 |j ddddd¬«Yy#t $r}t d«|‚d}~wwxYwwxYw) zyCheck whether `nltk` `punkt` is downloaded. If not, try to download if a machine is connected to the internet. rNztokenizers/punkt_tabÚ punkt_tabTF)ÚquietÚforceÚ halt_on_errorÚraise_on_errorzz`nltk` resource `punkt` is not available on a disk and cannot be downloaded as a machine is not connected to the internet.)rÚdataÚfindÚLookupErrorÚdownloadÚ ValueErrorÚOSError)rÚerrs úw/home/mcse/projects/srt_converter/srt-converter-venv/lib/python3.12/site-packages/torchmetrics/functional/text/rouge.pyÚ _ensure_nltk_punkt_is_downloadedr9+so€óð Ø ‰ ‰Ð-Õ.øÜòð ØM‰M˜+¨T¸ÈeÐdhˆMÖiøÜò Üð-óðð ûð úðús&†"¢ A"¬AÁ AÁ AÁAÁA"Úxcó—tstd«‚ddl}t«t j dd|«|j |«S)z]Split sentence to get rougeLsum scores matching published rougeL scores for BART and PEGASUS.zQROUGE-Lsum calculation requires that `nltk` is installed. Use `pip install nltk`.rNzÚ)r ÚModuleNotFoundErrorrr9ÚreÚsubÚ sent_tokenize)r:rs r8Ú_split_sentencerA?s;€åÜ!Ð"uÓvÐvÛä$Ô&ä‡FFˆ5"aÔØ×Ñ˜aÓ Ð óÚhits_or_lcsÚpred_lenÚ target_lencóÒ—||z}||z}||cxk(rdk(r$nn!td«td«td«dœSd|z|z||zz}t|«t|«t|«dœS)aCompute overall metrics. This function computes precision, recall and F1 score based on hits/lcs, the length of lists of tokenizer predicted and target sentences. Args: hits_or_lcs: A number of matches or a length of the longest common subsequence. pred_len: A length of a tokenized predicted sentence. target_len: A length of a tokenized target sentence. ç©Ú precisionÚrecallÚfmeasurer)r)rCrDrErIrJrKs r8Ú_compute_metricsrLKst€ð˜hÑ&€IØ ˜:Ñ %€FØFÔ!˜cÕ!Ü# C›[´F¸3³KÌVÐTWË[ÑYÐYà9‰}˜vÑ%¨°VÑ);Ñ<€HÜ Ó*´f¸V³nÔRXÐYaÓRbÑcÐcrBÚpred_tokensÚ target_tokensÚreturn_full_tablecóš—tt|«dz«Dcgc]}dgt|«dzz‘Œ}}tdt|«dz«D]l}tdt|«dz«D]O}||dz ||dz k(r||dz |dz dz|||<Œ,t||dz ||||dz «|||<ŒQŒn|r|S|ddScc}w)a0DP algorithm to compute the length of the longest common subsequence. Args: pred_tokens: A tokenized predicted sentence. target_tokens: A tokenized target sentence. return_full_table: If the full table of logest common subsequence should be returned or just the largest rréÿÿÿÿ)ÚrangeÚlenÚmax)rMrNrOÚ_ÚlcsÚiÚjs r8Ú_lcsrY`sþ€ô27´s¸=Ó7IÈAÑ7MÓ1NÖ O¨AˆAˆ3”#kÓ" QÑ&Ó'Ð O€CÐ OÜ 1”c˜-Ó(¨1Ñ,Ó -ò>ˆÜqœ#˜kÓ*¨QÑ.Ó/ò >ˆAØ˜Q ™UÑ# {°1°q±5Ñ'9Ò9Ø A¡™J q¨1¡uÑ-°Ñ1A‘q’ ä A¨¡E¡ ¨1¡ ¨s°1©v°a¸!±e©}Ó=A‘q’ ñ >ð>ñØˆ Øˆr‰72‰;ÐùòPsšCÚ lcs_tablecó—t|«}t|«}g}|dkDrc|dkDr^||dz ||dz k(r |jd|dz «|dz}|dz}n"|||dz ||dz |kDr|dz}n|dz}|dkDr|dkDrŒ^|S)z÷Backtrack LCS table. Args: lcs_table: A table containing information for the calculation of the longest common subsequence. pred_tokens: A tokenized predicted sentence. target_tokens: A tokenized target sentence. rr)rSÚinsert)rZrMrNrWrXÚbacktracked_lcss r8Ú_backtracked_lcsr^ws¶€ô ˆKÓ€AÜˆMÓ€AØ!#€OØ ˆaŠ%A˜’EØq˜1‘uÑ ¨q°1©uÑ!5Ò5Ø×"Ñ" 1 a¨!¡eÔ,Ø ‰FˆAØ ‰F‰AØ q‰\˜!˜a™%Ñ 9¨Q°©UÑ#3°AÑ#6Ò 6Ø ‰F‰Aà ‰FˆAðˆaŠ%A˜“EðÐrBÚpred_tokens_listcó—dttdttdttfd„}dtttdttfd„}|Dcgc]}|||«‘Œ }}||«Dcgc]}||‘Œ c}Scc}wcc}w)züFind union LCS between a target sentence and iterable of predicted tokens. Args: pred_tokens_list: A tokenized predicted sentence split by ``'\n'``. target_tokens: A tokenized single part of target sentence split by ``'\n'``. rMrNr*có8—t||d¬«}t|||«S)zRReturn one of the longest of longest common subsequence via backtracked lcs table.T)rO)rYr^)rMrNrZs r8Úlcs_indz_union_lcs..lcs_indšs €ä-1°+¸}Ð`dÔ-eˆ Ü ¨;¸ ÓFÐFrBÚ lcs_tablescóB—tt«j|Ž«S)z#Find union LCS given a list of LCS.)ÚsortedÚsetÚunion)rcs r8Ú find_unionz_union_lcs..find_unionŸs€äk”c“e—k‘k :Ð.Ó/Ð/rB)rÚstrÚint)r_rNrbrhrMrcrWs r8Ú _union_lcsrk‘s€ðGœX¤c™]ðG¼8ÄC¹=ðGÌXÔVYÉ]óGð 0œx¬´© Ñ6ð0¼8ÄC¹=ó0ðJZÖZ¸+‘'˜+ }Õ5ÐZ€JÐZÙ&0°Ó&<Ö= ˆM˜!ÓÒ=Ð=ùò[ùÚ=sÁBÁ1BÚtextÚstemmerÚ normalizerÚ tokenizercó˜—t|«r||«n$tjdd|j««}t|«r||«ntjd|«}|r.|Dcgc]#}t|«dkDr|j |«n|‘Œ%}}|Dcgc]$}t|t«sŒt|«dkDsŒ#|‘Œ&c}Scc}wcc}w)aRouge score should be calculated only over lowercased words and digits. Optionally, Porter stemmer can be used to strip word suffixes to improve matching. The text normalization follows the implemantion from `Rouge score_Text Normalizition`_. Args: text: An input sentence. stemmer: Porter stemmer instance to strip word suffixes to improve matching. normalizer: A user's own normalizer function. If this is ``None``, replacing any non-alpha-numeric characters with spaces is default. This function must take a ``str`` and return a ``str``. tokenizer: A user's own tokenizer function. If this is ``None``, splitting by spaces is default This function must take a ``str`` and return ``Sequence[str]`` z [^a-z0-9]+ú z\s+rr) Úcallabler>r?ÚlowerÚsplitrSÚstemÚ isinstanceri)rlrmrnroÚtokensr:s r8Ú_normalize_and_tokenize_textrx§s¬€ô. (¨ Ô3‰:dÔ¼¿¹À ÈsÐTX×T^ÑT^ÓT`Ó9a€Dô!)¨Ô 3‰YtŒ_¼¿¹À&È$Ó9O€Fáà@FÖG¸1¤S¨£V¨a¢Z'—,‘,˜q”/°QÑ6ÐGˆÐGðÖE!¤*¨Q´Õ"4¼¸Q»À!»ŠAÒEÐEùòHùòFsÁ((CÂCÂ,CÂ;CÚpredÚtargetÚn_gramcó–‡‡—dttdtdtfd„}|||«|||«cŠŠt ‰j««t ‰j««}}d||fvr!t d«t d«t d«dœSt ˆˆfd„t‰«D««}t|t|d «t|d ««S) z¶Compute precision, recall and F1 score for the Rouge-N metric. Args: pred: A predicted sentence. target: A target sentence. n_gram: N-gram overlap. rwÚnr*cóŠ‡‡—t«}ˆˆfd„tt‰«‰z dz«D«D]}||xxdz cc<Œ|S)Nc3ó@•K—|]}t‰||‰z«–—Œyw©N)Útuple)Ú.0rWr}rws €€r8ú z9_rouge_n_score.._create_ngrams..×s!øèø€ÒS°1”e˜F 1 q¨1¡uÐ-×.ÑSùsƒr)rrRrS)rwr}ÚngramsÚngrams`` r8Ú_create_ngramsz&_rouge_n_score.._create_ngramsÕsDù€Ü!›)ˆÜS¼¼cÀ&»kÈA¹oÐPQÑ>QÓ8RÔSò ˆEØ5‹M˜QÑŒMð àˆ rBrrGrHc3óB•K—|]}t‰|‰|«–—Œywr€)Úmin)r‚ÚwÚpred_ngramsÚ target_ngramss €€r8rƒz!_rouge_n_score..ás!øèø€ÒO¸Œs;˜q‘> =°Ñ#3×4ÑOùsƒr) rrirjrÚsumÚvaluesrrfrLrT) ryrzr{r†rDrEÚhitsrŠr‹s @@r8Ú_rouge_n_scorerËs¾ù€ðœx¬™}ð´ð¼óñ"0°°fÓ!=¹~ÈfÐV\Ó?]Ð€KÜ˜{×1Ñ1Ó3Ó4´c¸-×:NÑ:NÓ:PÓ6Qˆj€HØˆXzÐ"Ñ"Ü# C›[´F¸3³KÌVÐTWË[ÑYÐYôÔO¼cÀ+Ó>NÔOÓO€DÜ˜D¤# h°Ó"2´C¸ ÀAÓ4FÓGÐGrBcó®—t|«t|«}}d||fvr!td«td«td«dœSt||«}t|||«S)z–Compute precision, recall and F1 score for the Rouge-L metric. Args: pred: A predicted sentence. target: A target sentence. rrGrH)rSrrYrL)ryrzrDrErVs r8Ú_rouge_l_scorer‘åsW€ô˜t›9¤c¨&£kˆj€HØˆXzÐ"Ñ"Ü# C›[´F¸3³KÌVÐTWË[ÑYÐYäD˜&Ó!€CÜ˜C ¨:Ó6Ð6rBcóÄ—ttt|««}ttt|««}d||fvr!td«td«td«dœSdttt dtfd„}||«}||«}d}|D]F}t||«} | D]3} || dkDsŒ|| dkDsŒ|dz }|| xxdzcc<|| xxdzcc<Œ5ŒHt|||«S)a>Compute precision, recall and F1 score for the Rouge-LSum metric. More information can be found in Section 3.2 of the referenced paper [1]. This implementation follow the official implementation from: https://github.com/google-research/google-research/blob/master/rouge/rouge_scorer.py. Args: pred: An iterable of predicted sentence split by '\n'. target: An iterable target sentence split by '\n'. References: [1] ROUGE: A Package for Automatic Evaluation of Summaries by Chin-Yew Lin. https://aclanthology.org/W04-1013/ rrGrHÚ sentencesr*cóJ—t«}|D]}|j|«Œ|Sr€)rÚupdate)r“r„Úsentences r8Ú_get_token_countsz,_rouge_lsum_score.._get_token_counts s)€Ü!›)ˆØ!ò $ˆHØM‰M˜(Õ#ð $àˆ rBr) rŒÚmaprSrrrirrkrL)ryrzrDrEr—Úpred_tokens_countÚtarget_tokens_countrŽÚtgtrVÚtokens r8Ú_rouge_lsum_scorerõs€ô”3”s˜D“>Ó"€HÜ”Sœ˜fÓ%Ó&€JØˆXzÐ"Ñ"Ü# C›[´F¸3³KÌVÐTWË[ÑYÐYð¤X¬h´s©mÑ%<ðÄóñ*¨$Ó/ÐÙ+¨FÓ3Ðð €DØò0ˆÜ˜˜sÓ#ˆØò 0ˆEØ Ñ'¨!Ó+Ð0CÀEÑ0JÈQÓ0NØ˜‘ Ø! %Ó(¨AÑ-Ó(Ø# EÓ*¨aÑ/Ô*ñ 0ð0ô˜D (¨JÓ7Ð7rBÚpredsÚrouge_keys_valuesÚ accumulatecó—|Dcic]}|g“Œ}}t||«D]B\} } |Dcic]}|i“Œ}}|Dcic]}|g“Œ}}g} t| |||«}d|vr$t| «Dcgc]}t||||«‘Œ}}| D]¸}t||||«}d|vr$t|«Dcgc]}t||||«‘Œ}}|D]\}t|t«rt|||«}n#|dk(r t ||«}n|dk(rt«}||<||j|«Œ^| j|j««Œº|dk(r_|D]X}tjtj| Dcgc] }||d‘Œc}««}||j| ||«ŒZŒ€|dk(sŒ‡|Dcic]}|i“Œ}}|j«D]{\}}i}|D]7}|j«D]"\}}||vrg||<||j|«Œ$Œ9|Dcic])}|tj||«j«“Œ+c}||<Œ}|D]}||j||«ŒŒE|Scc}wcc}wcc}wcc}wcc}wcc}wcc}wcc}w)a«Update the rouge score with the current set of predicted and target sentences. Args: preds: An iterable of predicted sentences. target: An iterable of iterable of target sentences. rouge_keys_values: List of N-grams/'L'/'Lsum' arguments. accumulate: Useful in case of multi-reference rouge score. ``avg`` takes the avg of all references with respect to predictions ``best`` takes the best fmeasure score obtained between prediction and multiple corresponding references. Allowed values are ``avg`` and ``best``. stemmer: Porter stemmer instance to strip word suffixes to improve matching. normalizer: A user's own normalizer function. If this is ``None``, replacing any non-alpha-numeric characters with spaces is default. This function must take a `str` and return a `str`. tokenizer: A user's own tokenizer function. If this is ``None``, splitting by spaces is default This function must take a `str` and return `Sequence[str]` Example: >>> preds = ["My name is John"] >>> target = [["Is your name John"]] >>> from pprint import pprint >>> score = _rouge_score_update(preds, target, rouge_keys_values=[1, 2, 'L'], accumulate='best') >>> pprint(score) {1: [{'fmeasure': tensor(0.7500), 'precision': tensor(0.7500), 'recall': tensor(0.7500)}], 2: [{'fmeasure': tensor(0.), 'precision': tensor(0.), 'recall': tensor(0.)}], 'L': [{'fmeasure': tensor(0.5000), 'precision': tensor(0.5000), 'recall': tensor(0.5000)}]} rrr)rKr()ÚziprxrArvrjrr‘rÚappendÚcopyÚtorchÚargmaxrÚitemsÚmean)ržrzrŸr rmrnroÚ rouge_keyÚresultsÚpred_rawÚ target_rawÚresult_innerÚ result_avgÚlist_resultsryÚ pred_sentenceÚ pred_lsumÚtarget_raw_innerr›Útgt_sentenceÚtarget_lsumÚscoreÚkÚsÚindexÚnew_result_avgÚmetricsÚ_dict_metric_score_batchÚmetricÚ_typeÚvalues r8rr s>€ðV_pÖ>pÐQZ¸yÈ"¹}Ð>p€GÐ>pä # E¨6Ó 2ó5EÑˆ*ØarÖAsÐT]À)ÈRÁ-ÐAsˆÐAsØevÖEwÐXaÀiÐQSÁmÐEwˆ ÐEwØˆÜ+¨H°g¸zÈ9ÓUˆØÐ&Ñ&ô&5°XÓ%>öà!ô-¨]¸GÀZÐQZÕ[ðˆIðð !+ò 5ÐÜ.Ð/?ÀÈ*ÐV_Ó`ˆCàÐ*Ñ*ô)8Ð8HÓ(Iöà$ô1°¸wÈ ÐT]Õ^ððð /ò 4 Ü˜i¬Ô-Ü*¨4°°iÓ@‘EØ #Ò%Ü*¨4°Ó5‘EØ &Ò(Ü-¨i¸ÓEEØ*/˜YÑ'Ø˜9Ñ%×,Ñ,¨UÕ3ð 4ð ×Ñ × 1Ñ 1Ó 3Õ4ð% 5ð(˜ÒØ&ò :ÜŸ™¤U§\¡\È\Ö2ZÈ°1°Q±4¸ Ó3CÒ2ZÓ%[Ó\Ø˜‘ ×!Ñ! ,¨uÑ"5°aÑ"8Õ9ò :ð˜5Ô à/@öHØ"+ ˜2‘ ðHˆNðHð'1×&6Ñ&6Ó&8ò Ñ" ˜7ØDFÐ(Ø%òFFØ(.¯©«òF™˜˜uØ Ð(@Ñ@Ø>@Ð4°UÑ;Ø0°Ñ7×>Ñ>¸uÕEñFðFð^vö-ØTYEœ5Ÿ<™<Ð(@ÀÑ(GÓH×MÑMÓOÑOò-˜yÒ)ð ð/ò E Ø˜ Ñ"×)Ñ)¨.¸Ñ*CÕDò Eði5Eðn€Nùòs?qùòBtùÚEwùòùòùò$3[ùòHùò-s-… I%¨ I*¸ I/Á%I4Â I9Å%I>Æ( JÈ.JÚsentence_resultscó”—i}|ik(r|S|j«D]+\}}tj|«j«||<Œ-|S)zÈCompute the combined ROUGE metric for all the input set of predicted and target sentences. Args: sentence_results: Rouge-N/Rouge-L/Rouge-LSum metrics calculated for single sentence. )r§r¥rr¨)r¿rªr©Úscoress r8Ú_rouge_score_computerÂ‡sV€ð"$€Gà˜2ÒØˆà-×3Ñ3Ó5ò9Ñˆ 6Ü"Ÿ\™\¨&Ó1×6Ñ6Ó8ˆ Òð9ð€NrBÚuse_stemmerÚ rouge_keys.c ó\—|rtstd«‚ddl}|r$jjj«nd}t |t«s|f}|D]6} | tvsŒtd| ›dttj««›«‚|D cgc]} t| ‘Œ } } t |t«r6td„|D««r$t |t«r|gn|Dcgc]}|g‘Œc}}t |t«r|g}t |t«r|gg}t||| ||||¬«}| D cic]} dD]}d | ›d |›g“Œ Œ}} }|j«D]>\} }|D]4}|j«D]\}}|d | ›d |›j!|«Œ!Œ6Œ@t#|«Scc} wcc}wcc}} w)az Calculate `Calculate Rouge Score`_ , used for automatic summarization. Args: preds: An iterable of predicted sentences or a single predicted sentence. target: An iterable of iterables of target sentences or an iterable of target sentences or a single target sentence. accumulate: Useful in case of multi-reference rouge score. - ``avg`` takes the avg of all references with respect to predictions - ``best`` takes the best fmeasure score obtained between prediction and multiple corresponding references. use_stemmer: Use Porter stemmer to strip word suffixes to improve matching. normalizer: A user's own normalizer function. If this is ``None``, replacing any non-alpha-numeric characters with spaces is default. This function must take a ``str`` and return a ``str``. tokenizer: A user's own tokenizer function. If this is ``None``, splitting by spaces is default This function must take a ``str`` and return ``Sequence[str]`` rouge_keys: A list of rouge types to calculate. Keys that are allowed are ``rougeL``, ``rougeLsum``, and ``rouge1`` through ``rouge9``. Return: Python dictionary of rouge scores for each input rouge key. Example: >>> from torchmetrics.functional.text.rouge import rouge_score >>> preds = "My name is John" >>> target = "Is your name John" >>> from pprint import pprint >>> pprint(rouge_score(preds, target)) {'rouge1_fmeasure': tensor(0.7500), 'rouge1_precision': tensor(0.7500), 'rouge1_recall': tensor(0.7500), 'rouge2_fmeasure': tensor(0.), 'rouge2_precision': tensor(0.), 'rouge2_recall': tensor(0.), 'rougeL_fmeasure': tensor(0.5000), 'rougeL_precision': tensor(0.5000), 'rougeL_recall': tensor(0.5000), 'rougeLsum_fmeasure': tensor(0.5000), 'rougeLsum_precision': tensor(0.5000), 'rougeLsum_recall': tensor(0.5000)} Raises: ModuleNotFoundError: If the python package ``nltk`` is not installed. ValueError: If any of the ``rouge_keys`` does not belong to the allowed set of keys. References: [1] ROUGE: A Package for Automatic Evaluation of Summaries by Chin-Yew Lin. https://aclanthology.org/W04-1013/ zBStemmer requires that `nltk` is installed. Use `pip install nltk`.rNzGot unknown rouge key z. Expected to be one of c3ó<K—|]}t|t«–—Œywr€)rvri)r‚r›s r8rƒzrouge_score..æsèø€Ò'OÀ¬ °3¼×(<Ñ'Oùs‚)rmrnror )rKrIrJÚrougerU)r r=rruÚporterÚ PorterStemmerrvrr'r5ÚlistÚkeysÚallrirr§r£rÂ)ržrzr rÃrnrorÄrrmÚkeyrŸr›r¿r©ÚtpÚoutputrºr¼r¾s r8rr™së€ñ~ÝÜ%Ð&jÓkÐkÛá2=ˆdi‰i×Ñ×,Ñ,Ô.À4€Gäj¤%Ô(Ø ]ˆ ØòvˆØÔ(Ò(ÜÐ5°c°UÐ:RÔSWÔXj×XoÑXoÓXqÓSrÐRsÐtÓuÐuðvð=GÖG°SÔ+¨CÓ0ÐGÐÐGä&œ$Ô¤CÑ'OÈÔ'OÔ$OÜ'¨¬sÔ3&‘È6Ö9RÀC¸3º%Ò9Rˆä%œÔØˆä&œ#ÔØ(ˆäGZØ ØØØØØØôHÐð7H÷'Ø)2ÐRuò'ØLNˆ% ˆ{˜!˜B˜4Ð "Ñ$ð'Ø ð'€Fñ'ð/×4Ñ4Ó6ò?Ñˆ 7Øò ?ˆFØ#Ÿ\™\›^ò ?‘ EØ˜˜y˜k¨¨2¨$Ð/Ñ0×7Ñ7¸Õ>ñ ?ñ ?ð?ô Ó'Ð'ùò;Hùò:Sùó$'sÂFÃ F#Ä(F()r*N)F)NNN)r)FNN)rrr%r&))r>ÚcollectionsrÚcollections.abcrÚtypingrrrrr r¥r rÚtyping_extensionsrÚtorchmetrics.utilities.importsr Ú__doctest_requires__r'ÚdictrirjÚ__annotations__ÚALLOWED_ACCUMULATE_VALUESr9rArLÚboolrYr^rkrxrr‘rrÊrrÂrr©rBr8úrÛsUðô ÝÝ$ß7Õ7ãß Ý%å:à>ÀÀÐIÐðØØØØØØØØØØñ2ÐD˜˜e C¨ H™oÐ-Ñ.óð,Ðóð( !sð !˜x¨™}ó !ðd #ðd°ðdÀ#ðdÈ$ÈsÐTZÈ{ÑJ[ódð,Y^ñØ˜#‘ðØ/7¸©}ðØQUðà ˆ3˜ #™Ñ'Ð'Ñ(óð.Ø˜ ™ Ñ&ðØ5=¸c±]ðØS[Ð\_ÑS`ðà ˆc]óð4> ¨(°3©-Ñ!8ð>ÈÐRUÉð>Ð[cÐdgÑ[hó>ð0"Ø15Ø:>ñ !FØ ð!Fà c‰]ð!Fð˜ 3 %¨ *Ñ-Ñ.ð!Fð˜ # ¨°© Ð!5Ñ6Ñ7ð !Fð ˆc]ó!FðHH˜ #™ðH°¸± ðHÀsðHÈtÐTWÐY_ÐT_ÑO`óHð4 7˜ #™ð 7°¸± ð 7À$ÀsÈFÀ{ÑBSó 7ð (8˜H X¨c¡]Ñ3ð(8¸XÀhÈsÁmÑ=Tð(8ÐY]Ð^aÐciÐ^iÑYjó(8ð`"Ø15Ø:>ñdØC‰=ðdàX˜c‘]Ñ#ðdð˜E # s (™OÑ,ðdðð dð c‰]ðdð˜ 3 %¨ *Ñ-Ñ.ð dð˜ # ¨°© Ð!5Ñ6Ñ7ðdð ˆ%S‰/˜4 S¨& [Ñ 1Ñ2Ð 2Ñ3ódðN¨4°°T¸&±\Ð0AÑ+BðÀtÈCÐQWÈKÑGXóð**0ØØ15Ø:>Ø.Yñh(Øh˜s‘mÐ#Ñ$ðh(à#x ‘} h¨x¸©}Ñ&=Ð=Ñ>ðh(ð˜ Ñ&ðh(ðð h(ð ˜ 3 %¨ *Ñ-Ñ.ðh(ð˜ # ¨°© Ð!5Ñ6Ñ7ð h(ðc˜5 c ™?Ð*Ñ+ðh(ð ˆ#ˆvˆ+Ñôh(rB