U 4Af[@s dZddlZddlZddlZddlmZddlmZmZm Z m Z m Z ddl Z ddlmZmZmZmZmZddlmZmZddlmZmZmZmZdd lmZerddlZe e!Z"d d d Z#Gd ddeZ$dZ%eddZ&ddZ'GdddZ(GdddeZ)dS)zTokenization classes for TAPEX.N) lru_cache)DictListOptionalTupleUnion) ExplicitEnumPaddingStrategy TensorTypeadd_end_docstringsis_pandas_available) AddedTokenPreTrainedTokenizer)ENCODE_KWARGS_DOCSTRING BatchEncoding TextInputTruncationStrategy)loggingz vocab.jsonz merges.txt) vocab_file merges_filec@seZdZdZdZdS)TapexTruncationStrategyz} Possible values for the `truncation` argument in [`~TapasTokenizer.__call__`]. Useful for tab-completion in an IDE. Zdrop_rows_to_fitN)__name__ __module__ __qualname____doc__DROP_ROWS_TO_FITrr[/tmp/pip-unpacked-wheel-zw5xktn0/transformers/models/deprecated/tapex/tokenization_tapex.pyr(sra add_special_tokens (`bool`, *optional*, defaults to `True`): Whether or not to encode the sequences with the special tokens relative to their model. padding (`bool`, `str` or [`~file_utils.PaddingStrategy`], *optional*, defaults to `False`): Activates and controls padding. Accepts the following values: - `True` or `'longest'`: Pad to the longest sequence in the batch (or no padding if only a single sequence if provided). - `'max_length'`: Pad to a maximum length specified with the argument `max_length` or to the maximum acceptable input length for the model if that argument is not provided. - `False` or `'do_not_pad'` (default): No padding (i.e., can output a batch with sequences of different lengths). truncation (`bool`, `str`, [`TapexTruncationStrategy`] or [`~tokenization_utils_base.TruncationStrategy`], *optional*, defaults to `False`): Activates and controls truncation. Accepts the following values: - `'drop_rows_to_fit'`: Truncate to a maximum length specified with the argument `max_length` or to the maximum acceptable input length for the model if that argument is not provided. This will truncate row by row, removing rows from the table. - `True` or `'longest_first'`: Truncate to a maximum length specified with the argument `max_length` or to the maximum acceptable input length for the model if that argument is not provided. This will truncate token by token, removing a token from the longest sequence in the pair if a pair of sequences (or a batch of pairs) is provided. - `'only_first'`: Truncate to a maximum length specified with the argument `max_length` or to the maximum acceptable input length for the model if that argument is not provided. This will only truncate the first sequence of a pair if a pair of sequences (or a batch of pairs) is provided. - `'only_second'`: Truncate to a maximum length specified with the argument `max_length` or to the maximum acceptable input length for the model if that argument is not provided. This will only truncate the second sequence of a pair if a pair of sequences (or a batch of pairs) is provided. - `False` or `'do_not_truncate'` (default): No truncation (i.e., can output batch with sequence lengths greater than the model maximum admissible input size). max_length (`int`, *optional*): Controls the maximum length to use by one of the truncation/padding parameters. If left unset or set to `None`, this will use the predefined model maximum length if a maximum length is required by one of the truncation/padding parameters. If the model has no specific maximum input length (like XLNet) truncation/padding to a maximum length will be deactivated. stride (`int`, *optional*, defaults to 0): If set to a number along with `max_length`, the overflowing tokens returned when `return_overflowing_tokens=True` will contain some tokens from the end of the truncated sequence returned to provide some overlap between truncated and overflowing sequences. The value of this argument defines the number of overlapping tokens. pad_to_multiple_of (`int`, *optional*): If set will pad the sequence to a multiple of the provided value. This is especially useful to enable the use of Tensor Cores on NVIDIA hardware with compute capability `>= 7.5` (Volta). return_tensors (`str` or [`~file_utils.TensorType`], *optional*): If set, will return tensors instead of list of python integers. Acceptable values are: - `'tf'`: Return TensorFlow `tf.constant` objects. - `'pt'`: Return PyTorch `torch.Tensor` objects. - `'np'`: Return Numpy `np.ndarray` objects. cCstttdtddtttdtddtttdtdd}|dd}d }td D],}||krf|||d ||d7}qfd d |D}tt||S) a3 Returns list of utf-8 byte and a mapping to unicode strings. We specifically avoids mapping to whitespace/control characters the bpe code barfs on. The reversible bpe codes work on unicode strings. This means you need a large # of unicode characters in your vocab if you want to avoid UNKs. When you're at something like a 10B token dataset you end up needing around 5K for decent coverage. This is a significant percentage of your normal, say, 32K bpe vocab. To avoid that, we want lookup tables between utf-8 bytes and unicode strings. !~¡¬®ÿNrcSsg|] }t|qSr)chr).0nrrr ysz$bytes_to_unicode..)listrangeordappenddictzip)bscsr)brrrbytes_to_unicodefs L    r4cCs6t}|d}|ddD]}|||f|}q|S)z Return set of symbol pairs in a word. Word is represented as tuple of symbols (symbols being variable-length strings). rr!N)setadd)wordpairsZ prev_charcharrrr get_pairs}s r:c@s<eZdZdZedddZedddZeedd d Z d S) IndexedRowTableLinearizezQ FORMAT: col: col1 | col2 | col 3 row 1 : val1 | val2 | val3 row 2 : ... ) table_contentcCsdd|krd|kst|j||dd}t|dD]"\}}||j||ddd7}q8|S)zs Given a table, TableLinearize aims at converting it into a flatten sequence with special symbols. headerrows r!) row_index)AssertionErrorZPROMPT_MESSAGEprocess_header enumerate process_rowstrip)selfr<Z table_stri row_examplerrr process_tables z&IndexedRowTableLinearize.process_table)headerscCsdd|S)z} Given a list of headers, TableLinearize aims at converting it into a flatten sequence with special symbols. zcol :  | )join)rFrJrrrrBsz'IndexedRowTableLinearize.process_header)rowr@cCsXd}g}|D](}t|tr*|t|q ||q |d|7}dt|d|S)zq Given a row, TableLinearize aims at converting it into a flatten sequence with special symbols. rKzrow z : ) isinstanceintr.strrL)rFrMr@Zrow_strZrow_cell_values cell_valuerrrrDs  z$IndexedRowTableLinearize.process_rowN) rrrrrrIrrBrPrDrrrrr;s r;cseZdZdZeZddgZdcfd d Zddee e ee ee dddZ deee e ee e ee dfdd Z dfee e ee ee dddZdgddZeddZddZddZd d!Zd"d#Zd$d%Zd&d'Zdhee eeed(d)d*Zeeedied,ed,fe eeeefeeeefe ee ee fee ee!fe e e e e e eee"fe e e e e e e e e e#d-d.d/Z$djed,ed,fe eeeefeeeefe ee ee fee ee!fe e e e e e eee"fe e e e e e e e e e#d-d0d1Z%eeedked,ed,fe eeeee ee ee fee efe e e e e eee"fe e e e e e e e e e#d2d3d4Z&ddde j'e!j(dd+ddddd d d d dfed,ed,fe eee eee e e!e e e e e e eee"fe e e e e e e e e e#d5d6d7Z)eeeddde j'e!j(dd+ddddd d d dfed,ed,fe eeeefe eeeefe e e!e e e e e e ee e e e e e e e e#d8d9d:Z*eedld,e ee ee ee ee fee ee!e+fe e e eee"fee d; dd?d@Z-ddde j'e!j(dd+ddddd d d d dfd,e ee ee e e!e e e e e e eee"fe e e e e e e e e e#d5dAdBZ.dneeeefe ee ee fee ee!fe e e e e e eee"fe e e e e e e e e e#dCdDdEZ/doeee ee ee fee efe e e e e eee"fe e e e e e e e e e#dFdGdHZ0de j'e!j(dd+ddddd d d d dfeee e e!e e e e e e eee"fe e e e e e e e e e#dIdJdKZ1dpee ee ee fee ee!e+fe e e eee"fee dLdMdNZ2dqee ee ee fee efe e e e e eee"fe e e e e e e e e#dOdPdQZ3de j'e!j(dd+ddddd d d d dfee e e!e e e e e e eee"fe e e e e e e e e e#dIdRdSZ4deee!e+fdfdTdUZ5e6eedVdWdXZ7dYdZZ8dre6ee eeeefdVd[d\Z9dse6ed]d^d_Z:e6eee;d`dadbZ<Z=S)tTapexTokenizera Construct a TAPEX tokenizer. Based on byte-level Byte-Pair-Encoding (BPE). This tokenizer can be used to flatten one or more table(s) and concatenate them with one or more related sentences to be used by TAPEX models. The format that the TAPEX tokenizer creates is the following: sentence col: col1 | col2 | col 3 row 1 : val1 | val2 | val3 row 2 : ... The tokenizer supports a single table + single query, a single table and multiple queries (in which case the table will be duplicated for every query), a single query and multiple tables (in which case the query will be duplicated for every table), and multiple tables and queries. In other words, you can provide a batch of tables + questions to the tokenizer for instance to prepare them for the model. Tokenization itself is based on the BPE algorithm. It is identical to the one used by BART, RoBERTa and GPT-2. This tokenizer inherits from [`PreTrainedTokenizer`] which contains most of the main methods. Users should refer to this superclass for more information regarding those methods. Args: vocab_file (`str`): Path to the vocabulary file. merges_file (`str`): Path to the merges file. do_lower_case (`bool`, *optional*, defaults to `True`): Whether or not to lowercase the input when tokenizing. errors (`str`, *optional*, defaults to `"replace"`): Paradigm to follow when decoding bytes to UTF-8. See [bytes.decode](https://docs.python.org/3/library/stdtypes.html#bytes.decode) for more information. bos_token (`str`, *optional*, defaults to `""`): The beginning of sequence token that was used during pretraining. Can be used a sequence classifier token. When building a sequence using special tokens, this is not the token that is used for the beginning of sequence. The token used is the `cls_token`. eos_token (`str`, *optional*, defaults to `""`): The end of sequence token. When building a sequence using special tokens, this is not the token that is used for the end of sequence. The token used is the `sep_token`. sep_token (`str`, *optional*, defaults to `""`): The separator token, which is used when building a sequence from multiple sequences, e.g. two sequences for sequence classification or for a text and a question for question answering. It is also used as the last token of a sequence built with special tokens. cls_token (`str`, *optional*, defaults to `""`): The classifier token which is used when doing sequence classification (classification of the whole sequence instead of per-token classification). It is the first token of the sequence when built with special tokens. unk_token (`str`, *optional*, defaults to `""`): The unknown token. A token that is not in the vocabulary cannot be converted to an ID and is set to be this token instead. pad_token (`str`, *optional*, defaults to `""`): The token used for padding, for example when batching sequences of different lengths. mask_token (`str`, *optional*, defaults to `""`): The token used for masking values. This is the token used when training this model with masked language modeling. This is the token which the model will try to predict. add_prefix_space (`bool`, *optional*, defaults to `False`): Whether or not to add an initial space to the input. This allows to treat the leading word just as any other word. (BART tokenizer detect beginning of words by the preceding space). max_cell_length (`int`, *optional*, defaults to 15): Maximum number of characters per cell when linearizing a table. If this number is exceeded, truncation takes place. input_idsZattention_maskTreplaceFc st|trt|dddn|}t|tr4t|dddn|}t|trPt|dddn|}t|trlt|dddn|}t| trt| dddn| } t| trt| dddn| } t| trt| dddn| } t|dd}t||_W5QRXdd|jD|_||_ t |_ dd|j D|_ t|dd}| d d d }W5QRXd d |D}tt|tt||_i|_| |_||_td|_tjf||||||| ||| | | | d || |_t|_dS)NF)lstriprstripTutf-8encodingcSsi|]\}}||qSrrr(kvrrr sz+TapexTokenizer.__init__..cSsi|]\}}||qSrrrarrrrds r!cSsg|]}t|qSr)tuplesplit)r(mergerrrr*sz+TapexTokenizer.__init__..zJ's|'t|'re|'ve|'m|'ll|'d| ?\p{L}+| ?\p{N}+| ?[^\s\p{L}\p{N}]+|\s+(?!\S)|\s+) rr do_lower_caseerrors bos_token eos_token unk_token sep_token cls_token pad_token mask_tokenadd_prefix_spacemax_cell_length)rOrQropenjsonloadencoderitemsdecoderrkr4 byte_encoder byte_decoderreadrhr/r0r,len bpe_rankscachersrjrecompilepatsuper__init__rtr;table_linearize)rFrrrjrkrlrmrorprnrqrrrsrtkwargsZ vocab_handleZ merges_handleZ bpe_merges __class__rrrsP  zTapexTokenizer.__init__N) token_ids_0 token_ids_1returncCsD|dkr|jg||jgS|jg}|jg}||||||S)a Build model inputs from a sequence or a pair of sequence for sequence classification tasks by concatenating and adding special tokens. A TAPEX sequence has the following format: - single sequence: ` X ` - pair of sequences: ` A B ` Args: token_ids_0 (`List[int]`): List of IDs to which the special tokens will be added. token_ids_1 (`List[int]`, *optional*): Optional second list of IDs for sequence pairs. Returns: `List[int]`: List of [input IDs](../glossary#input-ids) with the appropriate special tokens. N) cls_token_id sep_token_id)rFrrclsseprrr build_inputs_with_special_tokens>s z/TapexTokenizer.build_inputs_with_special_tokens)rralready_has_special_tokensrcsh|rtj||ddS|dkr8dgdgt|dgSdgdgt|ddgdgt|dgS)a Args: Retrieve sequence ids from a token list that has no special tokens added. This method is called when adding special tokens using the tokenizer `prepare_for_model` method. token_ids_0 (`List[int]`): List of IDs. token_ids_1 (`List[int]`, *optional*): Optional second list of IDs for sequence pairs. already_has_special_tokens (`bool`, *optional*, defaults to `False`): Whether or not the token list is already formatted with special tokens for the model. Returns: `List[int]`: A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token. T)rrrNr!r)rget_special_tokens_maskr~)rFrrrrrrrUsz&TapexTokenizer.get_special_tokens_maskcCsP|jg}|jg}|dkr.t|||dgSt||||||dgS)a Args: Create a mask from the two sequences passed to be used in a sequence-pair classification task. TAPEX does not: make use of token type ids, therefore a list of zeros is returned. token_ids_0 (`List[int]`): List of IDs. token_ids_1 (`List[int]`, *optional*): Optional second list of IDs for sequence pairs. Returns: `List[int]`: List of zeros. Nr)rrr~)rFrrrrrrr$create_token_type_ids_from_sequencesns z3TapexTokenizer.create_token_type_ids_from_sequencescKs>|d|j}|s|r6t|dkr6|ds6d|}||fS)Nrsrr?)poprsr~isspace)rFtextZis_split_into_wordsrrsrrrprepare_for_tokenizations z'TapexTokenizer.prepare_for_tokenizationcCs t|jSN)r~rxrFrrr vocab_sizeszTapexTokenizer.vocab_sizecCst|jf|jSr)r/rxZadded_tokens_encoderrrrr get_vocabszTapexTokenizer.get_vocabc sd|jkrj|St|}t|}|s,|St|fddd}|jkrNqL|\}}g}d}|t|kr"z|||} Wn,tk r|||dYq"YnX|||| | }|||kr |t|dkr ||d|kr | |||d7}q^| |||d7}q^t|}|}t|dkrBqLq,t|}q,d |}|j|<|S)Ncsj|tdS)Ninf)rgetfloat)pairrrrz$TapexTokenizer.bpe..keyrr!r?) rrgr:minrr~index ValueErrorextendr.rL) rFtokenr7r8ZbigramfirstsecondZnew_wordrGjrrrbpesB    2     zTapexTokenizer.bpecsZg}tj|D]B}dfdd|dD}|dd|dDq|S)zTokenize a string.rNc3s|]}j|VqdSr)r{)r(r3rrr sz+TapexTokenizer._tokenize..r^css|] }|VqdSrr)r(Z bpe_tokenrrrrsr?)rfindallrrLencoderrrh)rFr bpe_tokensrrrr _tokenizes"zTapexTokenizer._tokenizecCs|j||j|jS)z0Converts a token (str) in an id using the vocab.)rxrrn)rFrrrr_convert_token_to_idsz#TapexTokenizer._convert_token_to_idcCs |j|S)z=Converts an index (integer) in a token (str) using the vocab.)rzr)rFrrrr_convert_id_to_tokensz#TapexTokenizer._convert_id_to_tokencs0d|}tfdd|Djdjd}|S)z:Converts a sequence of tokens (string) in a single string.rNcsg|]}j|qSr)r|)r(crrrr*sz;TapexTokenizer.convert_tokens_to_string..r^)rk)rL bytearraydecoderk)rFtokensrrrrconvert_tokens_to_strings "z'TapexTokenizer.convert_tokens_to_string)save_directoryfilename_prefixrc Cs(tj|s"td|ddStj||r6|dndtd}tj||rX|dndtd}t|ddd $}|t j |j d d d d dW5QRXd}t|ddd j}|dt |j dddD]B\}} || krtd|d| }|d|d|d7}qW5QRX||fS)NzVocabulary path (z) should be a directory-rNrrwr^r_rTF)indent sort_keys ensure_asciirerz#version: 0.2 cSs|dS)Nr!r)kvrrrrrz0TapexTokenizer.save_vocabulary..rzSaving vocabulary to zZ: BPE merge indices are not consecutive. Please check that the tokenizer is not corrupted!r?r!)ospathisdirloggererrorrLVOCAB_FILES_NAMESruwritervdumpsrxsortedrrywarning) rFrrrZ merge_filefrwriterrZ token_indexrrrsave_vocabularys2 (  zTapexTokenizer.save_vocabularyrz pd.DataFrame)tablequeryansweradd_special_tokenspadding truncation max_lengthstridepad_to_multiple_ofreturn_tensorsreturn_token_type_idsreturn_attention_maskreturn_overflowing_tokensreturn_special_tokens_maskreturn_offsets_mapping return_lengthverbosercKs|dk r<|jf||||||||| | | | | ||||d|S|dk rt|jf||||||| | | | | ||||d|StddS)aY Main method to tokenize and prepare for the model one or several table-sequence pair(s). Args: table (`pd.DataFrame`, `List[pd.DataFrame]`): Table(s) containing tabular data. query (`str` or `List[str]`, *optional*): Sentence or batch of sentences related to one or more table(s) to be encoded. Note that the number of sentences must match the number of tables. answer (`str` or `List[str]`, *optional*): Optionally, the corresponding answer to the questions as supervision. N)rrrrrrrrrrrrrrrrr)rrrrrrrrrrrrrrrz4You need to provide either a `table` or an `answer`.)source_call_functarget_call_funcr)rFrrrrrrrrrrrrrrrrrrrrr__call__sV#zTapexTokenizer.__call__cKs d}d}t|tjrd}n"t|ttfr This method is deprecated, `__call__` should be used instead. rrrrr)rrrrpadding_strategytruncation_strategyrrrrrrrrrr)"_get_padding_truncation_strategies_batch_encode_plus)rFrrrrrrrrrrrrrrrrrrrrrrrs: z TapexTokenizer.batch_encode_plus)rrrrrrrrrrrrrrrrrrcKs|r tdt|tjr4t|ttfr4|gt|}t|ttfrZt|trZ|gt|}|j||||||||| | | | ||| |d}t |S)Nzreturn_offset_mapping is not available when using Python tokenizers. To use this feature, change your tokenizer to one deriving from transformers.PreTrainedTokenizerFast.)rrrrrrrrrrrrrrrr) NotImplementedErrorrOrrr+rgr~rQ_batch_prepare_for_modelr)rFrrrrrrrrrrrrrrrrrr batch_outputsrrrrs6z!TapexTokenizer._batch_encode_plus)rrrrrrrrrrrrrrrrrcCsi}|dkrdgt|}t|||D]\}}}|j|||||d}|jrR|}||}|j|||tj j |j ||dd| | ||dd|d}| D]&\}}||krg||<|| |qq&|j ||j || | d}t|| d}|S)z This method adds special tokens, truncates sequences if overflowing while taking into account the special tokens and manages a moving window (with user defined stride) for overflowing tokens. NrrFidsrrrrrrrrrrrrprepend_batch_axisrrrrrZ tensor_type)r~r0prepare_table_queryrjlowertokenizeprepare_for_modelconvert_tokens_to_idsr DO_NOT_PADvalueryr.padr)rFrrrrrrrrrrrrrrrrrZ_tableZ_queryZ_answerrroutputsrrrrrrsX  z'TapexTokenizer._batch_prepare_for_model) rrrrrrrrrc Ks*|j|f|||||||d| } | dS)a6 Prepare a table, a string and possible answer for the model. This method does not return token type IDs, attention masks, etc. which are necessary for the model to work correctly. Use this method if you want to build your processing on your own, otherwise refer to `__call__`. )rrrrrrrrT)r) rFrrrrrrrrrZencoded_inputsrrrrOs  zTapexTokenizer.encode)rrrrrrrrrrrrrrrrcKsT|jf|||||d|\}}}}|jf||||||||| | | | | ||d|S)Nr)rrrrrrrrrrrrrrr)r _encode_plus)rFrrrrrrrrrrrrrrrrrrrrrros8 zTapexTokenizer.encode_pluscKsj|r td|j|||||d}|jr.|}||}|j||||j|j||| | d| | | |||dS)N return_offset_mapping is not available when using Python tokenizers. To use this feature, change your tokenizer to one deriving from transformers.PreTrainedTokenizerFast. More information on available tokenizers at https://github.com/huggingface/transformers/pull/2674rTrrrrrrrrrrrrrrr)rrrjrrrrr)rFrrrrrrrrrrrrrrrrrrrrrrrrs> zTapexTokenizer._encode_plus)rrrrrrrrrrrrrrrrcKsrt|ttf}|r@|jf|||||||| | | | | ||d|S|jf|||||||| | | | | ||d|SdS)z The method tokenizes and prepares the answer label for the model. Args: answer (`str` or `List[str]`): Corresponding answer supervision to the queries for training the model. )rrrrrrrrrrrrrrN)rOr+rgtarget_batch_encode_plustarget_encode_plus)rFrrrrrrrrrrrrrrrrrrrrrsLzTapexTokenizer.target_call_func)rrrrrrrrrrrrrrrcKsR|jf|||||d|\}}}}|jf||||||||| | | | | |d|S)z Prepare answer strings for the model. Args: answer `List[str]`: Corresponding answer supervision to the queries for training the model. r)rrrrrrrrrrrrrr)r_target_batch_encode_plus)rFrrrrrrrrrrrrrrrrrrrrrs6 z'TapexTokenizer.target_batch_encode_plus)rrrrrrrrrrrrrrrrcKsi}|D]}|jr|}||}|j|||tjj|j||dd| | | |dd|d}|D]&\}}||krxg||<|| |q`q|j ||j||| d}t ||d}t |S)NFrrr) rjrrrrr rrryr.rr)rFrrrrrrrrrrrrrrrrrrrrrrrrrrPsF  z(TapexTokenizer._target_batch_encode_plus)rrrrrrrc Ks&|jf||||||d|}|dS)a Prepare the answer string for the model. This method does not return token type IDs, attention masks, etc. which are necessary for the model to work correctly. Use this method if you want to build your processing on your own, otherwise refer to `__call__`. Args: answer `str`: Corresponding answer supervision to the queries for training the model )rrrrrrrT)r) rFrrrrrrrZencoded_outputsrrr target_encodes zTapexTokenizer.target_encode)rrrrrrrrrrrrrrcKsP|jf||||| d|\}}}}|jf||||||||| | | | | d |S)z Prepare a answer string for the model. Args: answer `str`: Corresponding answer supervision to the queries for training the model. r) rrrrrrrrrrrrr)r_target_encode_plus)rFrrrrrrrrrrrrrrrrrrrrs4 z!TapexTokenizer.target_encode_pluscKsZ| r td|}|jr|}||}|j||||j|j||||d| | | | ||dS)NrTr)rrjrrrrr)rFrrrrrrrrrrrrrrrrrrrrrrs2 z"TapexTokenizer._target_encode_plusc Cs|js\t|jdd|Dd}|||||tjkrN|j||||d|j |}nd}|dkr~t dd|d|dkrt d |r|rd nd}|r|||n|} | S) z This method can be used to linearize a table and add a corresponding query. Optionally, it also handles truncation of the table (cells). An answer can be provided for more precise truncation. cSsg|]\}}t|jqSr)r+values)r(rGrMrrrr*%sz6TapexTokenizer.prepare_table_query..)r=r>)rrNzUYou provide an empty table, or all cells contain much tokens (e.g., >= 1024 tokens). z@Please carefully check the corresponding table with the query : .z7You provide nothing to query with respect to the table.r?) emptyr+columnsZiterrowstruncate_table_cellsrrtruncate_table_rowsrrIrr) rFrrrrrr<Z linear_table separatorZ joint_inputrrrrs$   z"TapexTokenizer.prepare_table_query)r<questionrc Cs|i}|dD]8}t|D]*\}}||}|dk r|||<|||<qq |dk rxt|D] \}} | |krV|| ||<qVdS)Nr>)rC truncate_cellkeys) rFr<rrZ cell_mappingrMrGcellrZcaserrrr @s    z#TapexTokenizer.truncate_table_cellscCsft|tst|tr|S|dkr^||}t||jkrX|d|j}||}|SdSn|SdS)NrN)rOrPrrErr~rtr)rFrRZ try_tokensZ retain_tokensZretain_cell_valuerrrrPs   zTapexTokenizer.truncate_cellc Cs||||\}}|||||d}t|dD]D\}} |j| |d} t|| } | |krfqx|| 8}|d7}q2|d|d=dS)a Args: table_content: {"header": xxx, "rows": xxx, "id" (Optionally): xxx} question: natural language sentence answer: if for training, is the supervision; otherwise will be empty rr>r!N)estimate_delete_ratiodelete_unrelated_rowsrCrrDr~r) rFr<rrr delete_ratioremain_token_lenZmaximum_keep_rowsindrH value_stringvalue_token_lenrrrr _s z"TapexTokenizer.truncate_table_rows)r<rc Csd|ksd|krtd|j|dd}|j|d}|j|dd}t|t|}||}d} t|dD]\} } | |j| dd 7} qlt|| } | |krd |fSd || |fSdS) Nr=r>z?The table content should contain both 'header' and 'rows' keys.T)rFrNdr?gg?)rrrrBr~rCrD) rFr<rrZquestion_tokensZ header_stringZ header_tokensZused_token_lenrr_rHrrrrr|sz$TapexTokenizer.estimate_delete_ratio)r<rrrc sxg}g|dkst|dkr$t}ndd|D}|dk rH||t|dd}t|d}t|dD]f\} } dd| D} t| |@dkrt| |@dkr|| qt| d | d | | d | d gqtfd d |D}tt|t t|d|} t j || d } t t |D]} | | kr&|d| =q&d|krtt| dkrttdt| |ddS)zC The argument answer is used only during training. NrcSsh|] }|qSr)r)r(Zans_exrrr sz7TapexTokenizer.delete_unrelated_rows..z?!.,r?r>cSsh|]}t|qSr)rQr)r(rrrrrsrr!csg|]}|kr|qSrr)r(_row_idxZrelated_indicesrrr*sz8TapexTokenizer.delete_unrelated_rows..)rbidzDelete {:.2f} rows in table {})r~r5updaterhrErCr.rrrPrandomchoicesreversedr,rrformat)rFr<rrrZtruncated_unrelated_indicesZ answer_setZ question_setZ row_max_lenrrMZ lower_rowZ drop_itemsZdrop_row_indicesrrrrs0   &  z$TapexTokenizer.delete_unrelated_rows) TrUrVrWrWrVrXrYrZFr[)N)NF)N)F)N)NNNTFNNrNNNNFFFFT)NNTFNNrNNNNFFFFT)NNTFNNNNNNFFFFT)NNTFNNN)NNTFNNNNNNFFFT)TFNNrNNNNFFFFT) TFNNNNNNFFFFT)TFNNN) TFNNNNNNFFFT)NN)N)>rrrrrZvocab_files_namesZmodel_input_namesrrrPrrboolrrrpropertyrrrrrrrrQrrr r-TAPEX_ENCODE_PLUS_ADDITIONAL_KWARGS_DOCSTRINGrrr rr rrrrrZDO_NOT_TRUNCATErrrrrrrrrrrrrrr rr rrr __classcell__rrrrrSsGC       *   P   W   <   :F   3 =   D   8 ?  "   6 :  , rS)*rrvrr! functoolsrtypingrrrrrregexrZ file_utilsr r r r r Ztokenization_utilsrrZtokenization_utils_baserrrrutilsrZpandasrZ get_loggerrrrrr'r4r:r;rSrrrrs*    6  '