Merge branch 'master' into master

pytorch_transformers/tokenization_utils.py → transformers/tokenization_utils.py

@@ -23,7 +23,12 @@ import six
 import copy
 from io import open
 
-from .file_utils import cached_path
+from .file_utils import cached_path, is_tf_available, is_torch_available
+
+if is_tf_available():
+    import tensorflow as tf
+if is_torch_available():
+    import torch
 
 logger = logging.getLogger(__name__)
 
@@ -231,13 +236,13 @@ class PreTrainedTokenizer(object):
     @classmethod
     def from_pretrained(cls, *inputs, **kwargs):
         r"""
-        Instantiate a :class:`~pytorch_transformers.PreTrainedTokenizer` (or a derived class) from a predefined tokenizer.
+        Instantiate a :class:`~transformers.PreTrainedTokenizer` (or a derived class) from a predefined tokenizer.
 
         Args:
             pretrained_model_name_or_path: either:
 
                 - a string with the `shortcut name` of a predefined tokenizer to load from cache or download, e.g.: ``bert-base-uncased``.
-                - a path to a `directory` containing vocabulary files required by the tokenizer, for instance saved using the :func:`~pytorch_transformers.PreTrainedTokenizer.save_pretrained` method, e.g.: ``./my_model_directory/``.
+                - a path to a `directory` containing vocabulary files required by the tokenizer, for instance saved using the :func:`~transformers.PreTrainedTokenizer.save_pretrained` method, e.g.: ``./my_model_directory/``.
                 - (not applicable to all derived classes) a path or url to a single saved vocabulary file if and only if the tokenizer only requires a single vocabulary file (e.g. Bert, XLNet), e.g.: ``./my_model_directory/vocab.txt``.
 
             cache_dir: (`optional`) string:
@@ -252,7 +257,7 @@ class PreTrainedTokenizer(object):
 
             inputs: (`optional`) positional arguments: will be passed to the Tokenizer ``__init__`` method.
 
-            kwargs: (`optional`) keyword arguments: will be passed to the Tokenizer ``__init__`` method. Can be used to set special tokens like ``bos_token``, ``eos_token``, ``unk_token``, ``sep_token``, ``pad_token``, ``cls_token``, ``mask_token``, ``additional_special_tokens``. See parameters in the doc string of :class:`~pytorch_transformers.PreTrainedTokenizer` for details.
+            kwargs: (`optional`) keyword arguments: will be passed to the Tokenizer ``__init__`` method. Can be used to set special tokens like ``bos_token``, ``eos_token``, ``unk_token``, ``sep_token``, ``pad_token``, ``cls_token``, ``mask_token``, ``additional_special_tokens``. See parameters in the doc string of :class:`~transformers.PreTrainedTokenizer` for details.
 
         Examples::
 
@@ -425,9 +430,9 @@ class PreTrainedTokenizer(object):
                 - tokenizer instantiation positional and keywords inputs (e.g. do_lower_case for Bert).
 
             This won't save modifications other than (added tokens and special token mapping) you may have
-            applied to the tokenizer after the instantion (e.g. modifying tokenizer.do_lower_case after creation).
+            applied to the tokenizer after the instantiation (e.g. modifying tokenizer.do_lower_case after creation).
 
-            This method make sure the full tokenizer can then be re-loaded using the :func:`~pytorch_transformers.PreTrainedTokenizer.from_pretrained` class method.
+            This method make sure the full tokenizer can then be re-loaded using the :func:`~transformers.PreTrainedTokenizer.from_pretrained` class method.
         """
         if not os.path.isdir(save_directory):
             logger.error("Saving directory ({}) should be a directory".format(save_directory))
@@ -464,7 +469,7 @@ class PreTrainedTokenizer(object):
         """ Save the tokenizer vocabulary to a directory. This method does *NOT* save added tokens
             and special token mappings.
 
-            Please use :func:`~pytorch_transformers.PreTrainedTokenizer.save_pretrained` `()` to save the full Tokenizer state if you want to reload it using the :func:`~pytorch_transformers.PreTrainedTokenizer.from_pretrained` class method.
+            Please use :func:`~transformers.PreTrainedTokenizer.save_pretrained` `()` to save the full Tokenizer state if you want to reload it using the :func:`~transformers.PreTrainedTokenizer.from_pretrained` class method.
         """
         raise NotImplementedError
 
@@ -507,7 +512,8 @@ class PreTrainedTokenizer(object):
         for token in new_tokens:
             assert isinstance(token, str) or (six.PY2 and isinstance(token, unicode))
             if token != self.unk_token and \
-                    self.convert_tokens_to_ids(token) == self.convert_tokens_to_ids(self.unk_token):
+                    self.convert_tokens_to_ids(token) == self.convert_tokens_to_ids(self.unk_token) and \
+                    token not in to_add_tokens:
                 to_add_tokens.append(token)
                 logger.info("Adding %s to the vocabulary", token)
 
@@ -518,6 +524,30 @@ class PreTrainedTokenizer(object):
 
         return len(to_add_tokens)
 
+    def num_added_tokens(self, pair=False):
+        """
+        Returns the number of added tokens when encoding a sequence with special tokens.
+
+        Note:
+            This encodes inputs and checks the number of added tokens, and is therefore not efficient. Do not put this
+            inside your training loop.
+
+        Args:
+            pair: Returns the number of added tokens in the case of a sequence pair if set to True, returns the
+                number of added tokens in the case of a single sequence if set to False.
+
+        Returns:
+            Number of tokens added to sequences
+        """
+
+        if pair:
+            initial_tokens_len = len(self.encode("This is a sequence") + self.encode("This is another"))
+            final_tokens_len = len(self.encode("This is a sequence", "This is another", add_special_tokens=True))
+        else:
+            initial_tokens_len = len(self.encode("This is a sequence"))
+            final_tokens_len = len(self.encode("This is a sequence", add_special_tokens=True))
+
+        return final_tokens_len - initial_tokens_len
 
     def add_special_tokens(self, special_tokens_dict):
         """
@@ -663,38 +693,185 @@ class PreTrainedTokenizer(object):
     def _convert_token_to_id(self, token):
         raise NotImplementedError
 
-    def encode(self, text, text_pair=None, add_special_tokens=False, **kwargs):
+    def encode(self,
+                text,
+                text_pair=None,
+                add_special_tokens=False,
+                max_length=None,
+                stride=0,
+                truncate_first_sequence=True,
+                return_tensors=None,
+                **kwargs):
         """
         Converts a string in a sequence of ids (integer), using the tokenizer and vocabulary.
 
         Same as doing ``self.convert_tokens_to_ids(self.tokenize(text))``.
 
         Args:
-            text: The first sequence to be encoded.
-            text_pair: Optional second sequence to be encoded.
+            text: The first sequence to be encoded. This can be a string, a list of strings (tokenized string using
+                the `tokenize` method) or a list of integers (tokenized string ids using the `convert_tokens_to_ids`
+                method)
+            text_pair: Optional second sequence to be encoded. This can be a string, a list of strings (tokenized
+                string using the `tokenize` method) or a list of integers (tokenized string ids using the
+                `convert_tokens_to_ids` method)
             add_special_tokens: if set to ``True``, the sequences will be encoded with the special tokens relative
                 to their model.
+            max_length: if set to a number, will limit the total sequence returned so that it has a maximum length.
+                If there are overflowing tokens, those will be added to the returned dictionary
+            stride: if set to a number along with max_length, the overflowing tokens returned will contain some tokens
+                from the main sequence returned. The value of this argument defined the number of additional tokens.
+            truncate_first_sequence: if there is a specified max_length, this flag will choose which sequence
+                will be truncated.
+            return_tensors: (optional) can be set to 'tf' or 'pt' to return respectively TensorFlow tf.constant
+                or PyTorch torch.Tensor instead of a list of python integers.
             **kwargs: passed to the `self.tokenize()` method
         """
-        if text_pair is None:
-            if add_special_tokens:
-                return self.add_special_tokens_single_sentence(self.convert_tokens_to_ids(self.tokenize(text, **kwargs)))
-            else:
+        encoded_inputs = self.encode_plus(text,
+                                          text_pair=text_pair,
+                                          max_length=max_length,
+                                          add_special_tokens=add_special_tokens,
+                                          stride=stride,
+                                          truncate_first_sequence=truncate_first_sequence,
+                                          return_tensors=return_tensors,
+                                          **kwargs)
+
+        return encoded_inputs["input_ids"]
+
+    def encode_plus(self,
+                    text,
+                    text_pair=None,
+                    add_special_tokens=False,
+                    max_length=None,
+                    stride=0,
+                    truncate_first_sequence=True,
+                    return_tensors=None,
+                    **kwargs):
+        """
+        Returns a dictionary containing the encoded sequence or sequence pair and additional informations:
+        the mask for sequence classification and the overflowing elements if a ``max_length`` is specified.
+
+        Args:
+            text: The first sequence to be encoded. This can be a string, a list of strings (tokenized string using
+                the `tokenize` method) or a list of integers (tokenized string ids using the `convert_tokens_to_ids`
+                method)
+            text_pair: Optional second sequence to be encoded. This can be a string, a list of strings (tokenized
+                string using the `tokenize` method) or a list of integers (tokenized string ids using the
+                `convert_tokens_to_ids` method)
+            add_special_tokens: if set to ``True``, the sequences will be encoded with the special tokens relative
+                to their model.
+            max_length: if set to a number, will limit the total sequence returned so that it has a maximum length.
+                If there are overflowing tokens, those will be added to the returned dictionary
+            stride: if set to a number along with max_length, the overflowing tokens returned will contain some tokens
+                from the main sequence returned. The value of this argument defined the number of additional tokens.
+            truncate_first_sequence: if there is a specified max_length, this flag will choose which sequence
+                will be truncated.
+            return_tensors: (optional) can be set to 'tf' or 'pt' to return respectively TensorFlow tf.constant
+                or PyTorch torch.Tensor instead of a list of python integers.
+            **kwargs: passed to the `self.tokenize()` method
+        """
+
+        def get_input_ids(text):
+            if isinstance(text, six.string_types):
                 return self.convert_tokens_to_ids(self.tokenize(text, **kwargs))
+            elif isinstance(text, (list, tuple)) and len(text) > 0 and isinstance(text[0], six.string_types):
+                return self.convert_tokens_to_ids(text)
+            elif isinstance(text, (list, tuple)) and len(text) > 0 and isinstance(text[0], int):
+                return text
+            else:
+                raise ValueError("Input is not valid. Should be a string, a list/tuple of strings or a list/tuple of integers.")
+
+        first_ids = get_input_ids(text)
+        second_ids = get_input_ids(text_pair) if text_pair is not None else None
 
-        first_sentence_tokens = [self._convert_token_to_id(token) for token in self.tokenize(text, **kwargs)]
-        second_sentence_tokens = [self._convert_token_to_id(token) for token in self.tokenize(text_pair, **kwargs)]
+        return self.prepare_for_model(first_ids,
+                                      pair_ids=second_ids,
+                                      max_length=max_length,
+                                      add_special_tokens=add_special_tokens,
+                                      stride=stride,
+                                      truncate_first_sequence=truncate_first_sequence,
+                                      return_tensors=return_tensors)
+
+
+    def prepare_for_model(self, ids, pair_ids=None, max_length=None, add_special_tokens=False, stride=0,
+                          truncate_first_sequence=True, return_tensors=None):
+        """
+        Prepares a sequence of input id, or a pair of sequences of inputs ids so that it can be used by the model.
+        It adds special tokens, truncates
+        sequences if overflowing while taking into account the special tokens and manages a window stride for
+        overflowing tokens
+
+        Args:
+            ids: list of tokenized input ids. Can be obtained from a string by chaining the
+                `tokenize` and `convert_tokens_to_ids` methods.
+            pair_ids: Optional second list of input ids. Can be obtained from a string by chaining the
+                `tokenize` and `convert_tokens_to_ids` methods.
+            max_length: maximum length of the returned list. Will truncate by taking into account the special tokens.
+            add_special_tokens: if set to ``True``, the sequences will be encoded with the special tokens relative
+                to their model.
+            stride: window stride for overflowing tokens. Can be useful for edge effect removal when using sequential
+                list of inputs.
+            truncate_first_sequence: if set to `True` and an optional second list of input ids is provided,
+                alongside a specified `max_length`, will truncate the first sequence if the total size is superior
+                than the specified `max_length`. If set to `False`, will truncate the second sequence instead.
+            return_tensors: (optional) can be set to 'tf' or 'pt' to return respectively TensorFlow tf.constant
+                or PyTorch torch.Tensor instead of a list of python integers.
+
+        Return:
+            a dictionary containing the `input_ids` as well as the `overflowing_tokens` if a `max_length` was given.
+        """
+        pair = bool(pair_ids is not None)
+        len_ids = len(ids)
+        len_pair_ids = len(pair_ids) if pair else 0
+
+        encoded_inputs = {}
+        if max_length:
+            n_added_tokens = self.num_added_tokens(pair=pair) if add_special_tokens else 0
+            if pair and n_added_tokens + (len_pair_ids if truncate_first_sequence else len_ids) >= max_length:
+                logger.warning(
+                    "You supplied a pair of sequence in which the sequence that will not be truncated is longer than the maximum specified length."
+                    "This pair of sequences will not be truncated.")
+            else:
+                if n_added_tokens + len_ids + len_pair_ids > max_length:
+                    if truncate_first_sequence or not pair:
+                        encoded_inputs["overflowing_tokens"] = ids[max_length - len_pair_ids - n_added_tokens - stride:]
+                        ids = ids[:max_length - len_pair_ids - n_added_tokens]
+                    elif not truncate_first_sequence and pair:
+                        encoded_inputs["overflowing_tokens"] = pair_ids[max_length - len_ids - n_added_tokens - stride:]
+                        pair_ids = pair_ids[:max_length - len_ids - n_added_tokens]
+                    else:
+                        logger.warning(
+                            "Cannot truncate second sequence as it is not provided. No truncation.")
 
         if add_special_tokens:
-            return self.add_special_tokens_sentences_pair(first_sentence_tokens, second_sentence_tokens)
+            sequence = self.add_special_tokens_sequence_pair(ids, pair_ids) if pair else self.add_special_tokens_single_sequence(ids)
+            token_type_ids = self.create_token_type_ids_from_sequences(ids, pair_ids) if pair else [0] * len(sequence)
         else:
-            return first_sentence_tokens, second_sentence_tokens
+            sequence = ids + pair_ids if pair else ids
+            token_type_ids = [0] * len(ids) + ([1] * len(pair_ids) if pair else [])
+
+        if return_tensors == 'tf' and is_tf_available():
+            sequence = tf.constant([sequence])
+            token_type_ids = tf.constant([token_type_ids])
+        elif return_tensors == 'pt' and is_torch_available():
+            sequence = torch.tensor([sequence])
+            token_type_ids = torch.tensor([token_type_ids])
+        elif return_tensors is not None:
+            logger.warning("Unable to convert output to tensors format {}, PyTorch or TensorFlow is not available.".format(return_tensors))
+
+        encoded_inputs["input_ids"] = sequence
+        encoded_inputs["token_type_ids"] = token_type_ids
+
+        return encoded_inputs
 
-    def add_special_tokens_single_sentence(self, token_ids):
+    def create_token_type_ids_from_sequences(self, token_ids_0, token_ids_1):
+        logger.warning("This tokenizer does not make use of special tokens.")
+        return [0] * len(token_ids_0) + [1] * len(token_ids_1)
+
+    def add_special_tokens_single_sequence(self, token_ids):
         logger.warning("This tokenizer does not make use of special tokens. The sequence has been returned with no modification.")
         return token_ids
 
-    def add_special_tokens_sentences_pair(self, token_ids_0, token_ids_1):
+    def add_special_tokens_sequence_pair(self, token_ids_0, token_ids_1):
         logger.warning("This tokenizer does not make use of special tokens. The two sequences have been concatenated.")
         return token_ids_0 + token_ids_1
 
@@ -740,7 +917,7 @@ class PreTrainedTokenizer(object):
 
         # To avoid mixing byte-level and unicode for byte-level BPT
         # we need to build string separatly for added tokens and byte-level tokens
-        # cf. https://github.com/huggingface/pytorch-transformers/issues/1133
+        # cf. https://github.com/huggingface/transformers/issues/1133
         sub_texts = []
         current_sub_text = []
         for token in filtered_tokens:
@@ -757,15 +934,6 @@ class PreTrainedTokenizer(object):
             sub_texts.append(self.convert_tokens_to_string(current_sub_text))
         text = ''.join(sub_texts)
 
-        if self._sep_token is not None and self._sep_token in text:
-            text = text.replace(self._cls_token, self._sep_token)
-            split_text = list(filter(lambda sentence: len(sentence) > 0, text.split(self._sep_token)))
-            if clean_up_tokenization_spaces:
-                clean_text = [self.clean_up_tokenization(text) for text in split_text]
-                return clean_text
-            else:
-                return split_text
-        else:
         if clean_up_tokenization_spaces:
             clean_text = self.clean_up_tokenization(text)
             return clean_text

pytorch_transformers/tokenization_xlm.py → transformers/tokenization_xlm.py

@@ -754,14 +754,14 @@ class XLMTokenizer(PreTrainedTokenizer):
         out_string = ''.join(tokens).replace('</w>', ' ').strip()
         return out_string
 
-    def add_special_tokens_single_sentence(self, token_ids):
+    def add_special_tokens_single_sequence(self, token_ids):
         """
         Adds special tokens to a sequence for sequence classification tasks.
         An XLM sequence has the following format: [CLS] X [SEP]
         """
         return [self.cls_token_id] + token_ids + [self.sep_token_id]
 
-    def add_special_tokens_sentences_pair(self, token_ids_0, token_ids_1):
+    def add_special_tokens_sequence_pair(self, token_ids_0, token_ids_1):
         """
         Adds special tokens to a sequence pair for sequence classification tasks.
         An XLM sequence pair has the following format: [CLS] A [SEP] B [SEP]
@@ -770,6 +770,18 @@ class XLMTokenizer(PreTrainedTokenizer):
         cls = [self.cls_token_id]
         return cls + token_ids_0 + sep + token_ids_1 + sep
 
+    def create_token_type_ids_from_sequences(self, token_ids_0, token_ids_1):
+        """
+        Creates a mask from the two sequences passed to be used in a sequence-pair classification task.
+        An XLM sequence pair mask has the following format:
+        0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1
+        | first sequence    | second sequence
+        """
+        sep = [self.sep_token_id]
+        cls = [self.cls_token_id]
+
+        return len(cls + token_ids_0 + sep) * [0] + len(token_ids_1 + sep) * [1]
+
     def save_vocabulary(self, save_directory):
         """Save the tokenizer vocabulary and merge files to a directory."""
         if not os.path.isdir(save_directory):

pytorch_transformers/tokenization_xlnet.py → transformers/tokenization_xlnet.py

@@ -181,7 +181,7 @@ class XLNetTokenizer(PreTrainedTokenizer):
         out_string = ''.join(tokens).replace(SPIECE_UNDERLINE, ' ').strip()
         return out_string
 
-    def add_special_tokens_single_sentence(self, token_ids):
+    def add_special_tokens_single_sequence(self, token_ids):
         """
         Adds special tokens to a sequence for sequence classification tasks.
         An XLNet sequence has the following format: X [SEP][CLS]
@@ -190,15 +190,29 @@ class XLNetTokenizer(PreTrainedTokenizer):
         cls = [self.cls_token_id]
         return token_ids + sep + cls
 
-    def add_special_tokens_sentences_pair(self, token_ids_0, token_ids_1):
+    def add_special_tokens_sequence_pair(self, token_ids_0, token_ids_1):
         """
         Adds special tokens to a sequence pair for sequence classification tasks.
         An XLNet sequence pair has the following format: A [SEP] B [SEP][CLS]
         """
+
         sep = [self.sep_token_id]
         cls = [self.cls_token_id]
         return token_ids_0 + sep + token_ids_1 + sep + cls
 
+    def create_token_type_ids_from_sequences(self, token_ids_0, token_ids_1):
+        """
+        Creates a mask from the two sequences passed to be used in a sequence-pair classification task.
+        A BERT sequence pair mask has the following format:
+        0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 2
+        | first sequence    | second sequence     | CLS segment ID
+        """
+        sep = [self.sep_token_id]
+        cls = [self.cls_token_id]
+        cls_segment_id = [2]
+
+        return len(token_ids_0 + sep) * [0] + len(token_ids_1 + sep) * [1] + cls_segment_id
+
     def save_vocabulary(self, save_directory):
         """ Save the sentencepiece vocabulary (copy original file) and special tokens file
             to a directory.