# This file is part of UDPipe Future <http://github.com/CoNLL-UD-2018/UDPipe-Future>.
# Copyright 2019 Institute of Formal and Applied Linguistics, Faculty of
# Mathematics and Physics, Charles University in Prague, Czech Republic.
# This Source Code Form is subject to the terms of the Mozilla Public
# License, v. 2.0. If a copy of the MPL was not distributed with this
# file, You can obtain one at http://mozilla.org/MPL/2.0/.
# Note that most of this file content is taken from original
# Bert sources from https://github.com/google-research/bert,
# under the Apache license 2.0, modified to return character-length
# of [UNK] subwords.

"""Extract pre-computed feature vectors from BERT."""

from __future__ import division
from __future__ import print_function

import collections
import copy
import json
import math
import os
import re
import six
import sys
import unicodedata

import numpy as np
import tensorflow as tf

class BertWrapper:
    LANGUAGE_ENGLISH = "english"
    LANGUAGE_CHINESE = "chinese"
    LANGUAGE_MULTILINGUAL = "multilingual"

    SIZE_BASE = "base"
    SIZE_LARGE = "large"

    CASING_CASED = "cased"
    CASING_UNCASED = "uncased"


    def __init__(self, language, size="base", casing="uncased", layer_indices=[-1,-2,-3,-4], with_cls=False, threads=1, batch_size=16, context_sentences=0):
        """Construct BertWrapper instance

        Construct BertWrapper instance. If the Bert model for requested options
        does not exist, ValueError is raised.

            language: Language of the Bert model, see BertWrapper.LANGUAGE_*

            size: Size of the Bert model, see BertWrapper.SIZE_*
              default: `base` (all models except for the English one have been
                released only as `base` size)

            casing: Casing of the Bert mode, see BertWrapper.CASING_*
              default: `uncased`

            layer_indices: list of Bert model layers to average and return
              default: `[-1, -2, -3, -4]`

            with_cls: Also return sentence-level embedding (the embedding of
              the CLS token as described in the Bert paper)

            threads: Number of CPU threads to use
              default: 1

            batch_size: Maximum batch size
              default: 16 (should fit on 8GB GPU with maximum sentence length)

        assert len(layer_indices) > 0
        self._layer_indices = layer_indices
        self._uncased = casing == self.CASING_UNCASED
        self._with_cls = 1 if with_cls else 0
        self._batch_size = batch_size
        self._context_sentences = context_sentences

        model_path = "bert/models/{}-{}-{}".format(language, size, casing)
        if not os.path.exists(model_path):
            raise ValueError("The requested Bert model combination {}-{}-{} does not exist".format(language, size, casing))

        # Initialize the Bert model
        self._tokenizer = FullTokenizer(vocab_file="{}/vocab.txt".format(model_path), do_lower_case=self._uncased)

        bert_config = BertConfig.from_json_file("{}/bert_config.json".format(model_path))

        def model_fn(features, labels, mode, params):
            model = BertModel(config=bert_config, is_training=False, input_ids=features["input_ids"],
                              input_mask=features["input_mask"], token_type_ids=features["input_type_ids"], use_one_hot_embeddings=False)
            assert mode == tf.estimator.ModeKeys.PREDICT

            tvars = tf.trainable_variables()
            (assignment_map, initialized_variable_names) = get_assignment_map_from_checkpoint(tvars, "{}/bert_model.ckpt".format(model_path))
            tf.train.init_from_checkpoint("{}/bert_model.ckpt".format(model_path), assignment_map)
            all_layers = model.get_all_encoder_layers()

            predictions = {"unique_id": features["unique_ids"]}
            for layer_index in layer_indices:
                predictions["layer_output_{}".format(layer_index)] = all_layers[layer_index]

            return tf.estimator.EstimatorSpec(mode=mode, predictions=predictions)

        self._estimator = tf.estimator.Estimator(
            config=tf.estimator.RunConfig(session_config=tf.ConfigProto(inter_op_parallelism_threads=threads, intra_op_parallelism_threads=threads)))

    def bert_embeddings(self, sentences):
        """Returns pretrained BERT embeddings for all word in sentences

        Returns pretrained BERT embeddings for all words in sentences.
        The embeddings are predicted from the pretrained BERT model and the
        given layers are averaged. If the input word is tokenized to more
        subwords by BERT, all subword embeddings are averaged into one vector
        for each word in input. Optionally, CLS embedding is also returned
        (as the first embedding).

        The TensorFlow computational graph for the Bert model is constructed
        and the checkpoint loaded during each call.

            sentences: an array sentences, each being an array of strings.
            A Python list of sentence embeddings, each a numpy array of shape
            [(1 if cls_embedding else 0) + sentence_length, embedding_size]."""

        def normalize_token(token):
            token = convert_to_unicode(token)
            token = self._tokenizer.basic_tokenizer._clean_text(token)
            token = "".join(c for c in token if not c.isspace())
            if self._uncased: token = token.lower()
            token = self._tokenizer.basic_tokenizer._run_strip_accents(token)
            return token

        # Tokenize input data
        InputFeatures = collections.namedtuple("InputFeatures", "unique_id sentence subwords token_ids input_ids input_mask input_type_ids")
        features, token_subwords = [], []
        stat_tokens, stat_subwords, stat_subword_unks = 0, 0, 0
        for index, sentence in enumerate(sentences):
            # Tokenize into subwords
            subwords = self._tokenizer.tokenize(" ".join(sentence))

            stat_tokens += len(sentence)
            stat_subwords += len(subwords)
            stat_subword_unks += len([subword for subword in subwords if subword.startswith("[UNK]")])

            # Align with original tokens
            token_ids, subwords_str, current_token, current_token_normalized = [-1] * len(subwords), "", 0, None
            for i, subword in enumerate(subwords):
                if subword in ["[CLS]", "[SEP]"]: continue

                while current_token_normalized is None:
                    current_token_normalized = normalize_token(sentence[current_token])

                    if not current_token_normalized:
                        current_token += 1
                        current_token_normalized = None

                if subword.startswith("[UNK]"):
                    unk_length = int(subword[6:])
                    subwords[i] = subword[:5]
                    subwords_str += current_token_normalized[len(subwords_str):len(subwords_str) + unk_length]
                    subwords_str += subword[2:] if subword.startswith("##") else subword
                assert current_token_normalized.startswith(subwords_str)

                token_ids[i] = current_token
                token_subwords[-1][current_token] += 1
                if current_token_normalized == subwords_str:
                    subwords_str = ""
                    current_token += 1
                    current_token_normalized = None

            assert current_token_normalized is None
            while current_token < len(sentence):
                assert not normalize_token(sentence[current_token])
                current_token += 1
            assert current_token == len(sentence)

            # Split into segments with maximum size
            while subwords:
                segment_size = min(len(subwords), self._MAX_SENTENCE_LEN - 2)
                if segment_size < len(subwords):
                    while segment_size > 0 and token_ids[segment_size - 1] == token_ids[segment_size]:
                        segment_size -= 1
                    assert segment_size > 0

                input_subwords = []
                subwords = subwords[segment_size:]

                input_token_ids = np.array([-1] + token_ids[:segment_size] + [-1], dtype=np.int32)
                token_ids = token_ids[segment_size:]

                input_ids = np.zeros(self._MAX_SENTENCE_LEN, dtype=np.int32)
                input_ids[:len(input_subwords)] = self._tokenizer.convert_tokens_to_ids(input_subwords)

                input_mask = np.zeros(self._MAX_SENTENCE_LEN, dtype=np.int8)
                input_mask[:len(input_subwords)] = 1

                input_type_ids = np.zeros(self._MAX_SENTENCE_LEN, dtype=np.int8)


        print("Tokenized {} tokens into {} subwords ({:.3f} per token) with {} UNKs ({:.3f}%)".format(
            stat_tokens, stat_subwords, stat_subwords / stat_tokens,
            stat_subword_unks, 100 * stat_subword_unks / stat_subwords), file=sys.stderr, flush=True)

        if self._context_sentences:
            with_context = []
            for i in range(len(features)):
                current = features[i]
                pre, pre_ids, post, post_ids = [], [], [], []
                for j in range(1, self._context_sentences + 1):
                    if i - j >= 0:
                        pre = features[i - j].subwords[1:-1] + pre
                        pre_ids = list(features[i - j].input_ids[1:len(features[i - j].subwords) - 1]) + pre_ids
                    if i + j < len(features):
                        post = post + features[i + j].subwords[1:-1]
                        post_ids = post_ids + list(features[i + j].input_ids[1:len(features[i + j].subwords) - 1])
                    total_subwords = len(current.subwords) + len(pre) + len(post)
                    if total_subwords >= self._MAX_SENTENCE_LEN or j == self._context_sentences:
                        if total_subwords > self._MAX_SENTENCE_LEN:
                            pre = pre[(total_subwords - self._MAX_SENTENCE_LEN) // 2:]
                            pre_ids = pre_ids[(total_subwords - self._MAX_SENTENCE_LEN) // 2:]
                        total_subwords = len(current.subwords) + len(pre) + len(post)
                        if total_subwords > self._MAX_SENTENCE_LEN:
                            post = post[:-(total_subwords - self._MAX_SENTENCE_LEN)]
                            post_ids = post_ids[:-(total_subwords - self._MAX_SENTENCE_LEN)]
                        total_subwords = len(current.subwords) + len(pre) + len(post)
                        if total_subwords > self._MAX_SENTENCE_LEN:
                            pre = pre[total_subwords - self._MAX_SENTENCE_LEN:]
                            pre_ids = pre_ids[total_subwords - self._MAX_SENTENCE_LEN:]
                        total_subwords = len(current.subwords) + len(pre) + len(post)

                        sw, rest = len(current.subwords[1:-1]), self._MAX_SENTENCE_LEN - total_subwords
                            subwords=current.subwords[0:1] + pre + current.subwords[1:-1] + post + current.subwords[-1:],
                            token_ids=np.array([-1] * len(pre) + list(current.token_ids[:2+sw]) + [-1] * len(post), np.int32),
                            input_ids=np.array([current.input_ids[0]] + pre_ids + list(current.input_ids[1:1+sw]) + post_ids + [current.input_ids[1+sw]] + [0] * rest, np.int32),
                            input_mask=np.array([1] * total_subwords + [0] * rest, np.int8),
                        print("CS {}".format(j), file=sys.stderr, flush=True)
            features = with_context

        def input_generator():
            for feature in features:
                yield {"unique_ids": feature.unique_id, "input_ids": feature.input_ids,
                       "input_mask": feature.input_mask, "input_type_ids": feature.input_type_ids}

        def input_fn(params):
            dataset = tf.data.Dataset.from_generator(
                {"unique_ids": tf.int32, "input_ids": tf.int32, "input_mask": tf.int32, "input_type_ids": tf.int32},
                {"unique_ids": [], "input_ids": [self._MAX_SENTENCE_LEN], "input_mask": [self._MAX_SENTENCE_LEN], "input_type_ids": [self._MAX_SENTENCE_LEN]})
            dataset = dataset.batch(batch_size=self._batch_size, drop_remainder=False)
            return dataset

        embedding_dim = None
        for result in self._estimator.predict(input_fn, yield_single_examples=True):
            if embedding_dim is None:
                embedding_dim = result["layer_output_{}".format(self._layer_indices[0])].shape[1]

            feature_index = int(result["unique_id"])
            sentence = features[feature_index].sentence
            if feature_index == 0 or features[feature_index].sentence != features[feature_index - 1].sentence:
                current_embeddings = np.zeros((len(sentences[sentence]) + self._with_cls, embedding_dim), dtype=np.float32)
                if self._with_cls:
                    for layer_index in self._layer_indices:
                        current_embeddings[0] += result["layer_output_{}".format(layer_index)][0] / len(self._layer_indices)

            for i, token_id in enumerate(features[feature_index].token_ids):
                if token_id >= 0:
                    for layer_index in self._layer_indices:
                        current_embeddings[token_id + self._with_cls] += \
                            result["layer_output_{}".format(layer_index)][i] / (token_subwords[sentence][token_id] * len(self._layer_indices))

            # Yield results if the whole sentence was processed
            if feature_index + 1 == len(features) or features[feature_index].sentence != features[feature_index + 1].sentence:
                yield current_embeddings

### Original Bert sources from https://github.com/google-research/bert,
### under the Apache license 2.0, modified to return character-length
### of [UNK] subwords.

### tokenization.py
def convert_to_unicode(text):
    """Converts `text` to Unicode (if it's not already), assuming utf-8 input."""
    if six.PY3:
        if isinstance(text, str):
            return text
        elif isinstance(text, bytes):
            return text.decode("utf-8", "ignore")
            raise ValueError("Unsupported string type: %s" % (type(text)))
    elif six.PY2:
        if isinstance(text, str):
            return text.decode("utf-8", "ignore")
        elif isinstance(text, unicode):
            return text
            raise ValueError("Unsupported string type: %s" % (type(text)))
        raise ValueError("Not running on Python2 or Python 3?")

def printable_text(text):
    """Returns text encoded in a way suitable for print or `tf.logging`."""

    # These functions want `str` for both Python2 and Python3, but in one case
    # it's a Unicode string and in the other it's a byte string.
    if six.PY3:
        if isinstance(text, str):
            return text
        elif isinstance(text, bytes):
            return text.decode("utf-8", "ignore")
            raise ValueError("Unsupported string type: %s" % (type(text)))
    elif six.PY2:
        if isinstance(text, str):
            return text
        elif isinstance(text, unicode):
            return text.encode("utf-8")
            raise ValueError("Unsupported string type: %s" % (type(text)))
        raise ValueError("Not running on Python2 or Python 3?")

def load_vocab(vocab_file):
    """Loads a vocabulary file into a dictionary."""
    vocab = collections.OrderedDict()
    index = 0
    with tf.gfile.GFile(vocab_file, "r") as reader:
        while True:
            token = convert_to_unicode(reader.readline())
            if not token:
            token = token.strip()
            vocab[token] = index
            index += 1
    return vocab

def convert_by_vocab(vocab, items):
    """Converts a sequence of [tokens|ids] using the vocab."""
    output = []
    for item in items:
    return output

def convert_tokens_to_ids(vocab, tokens):
    return convert_by_vocab(vocab, tokens)

def convert_ids_to_tokens(inv_vocab, ids):
    return convert_by_vocab(inv_vocab, ids)

def whitespace_tokenize(text):
    """Runs basic whitespace cleaning and splitting on a piece of text."""
    text = text.strip()
    if not text:
        return []
    tokens = text.split()
    return tokens

class FullTokenizer(object):
    """Runs end-to-end tokenziation."""

    def __init__(self, vocab_file, do_lower_case=True):
        self.vocab = load_vocab(vocab_file)
        self.inv_vocab = {v: k for k, v in self.vocab.items()}
        self.basic_tokenizer = BasicTokenizer(do_lower_case=do_lower_case)
        self.wordpiece_tokenizer = WordpieceTokenizer(vocab=self.vocab)

    def tokenize(self, text):
        split_tokens = []
        for token in self.basic_tokenizer.tokenize(text):
            for sub_token in self.wordpiece_tokenizer.tokenize(token):

        return split_tokens

    def convert_tokens_to_ids(self, tokens):
        return convert_by_vocab(self.vocab, tokens)

    def convert_ids_to_tokens(self, ids):
        return convert_by_vocab(self.inv_vocab, ids)

class BasicTokenizer(object):
    """Runs basic tokenization (punctuation splitting, lower casing, etc.)."""

    def __init__(self, do_lower_case=True):
        """Constructs a BasicTokenizer.

            do_lower_case: Whether to lower case the input.
        self.do_lower_case = do_lower_case

    def tokenize(self, text):
        """Tokenizes a piece of text."""
        text = convert_to_unicode(text)
        text = self._clean_text(text)

        # This was added on November 1st, 2018 for the multilingual and Chinese
        # models. This is also applied to the English models now, but it doesn't
        # matter since the English models were not trained on any Chinese data
        # and generally don't have any Chinese data in them (there are Chinese
        # characters in the vocabulary because Wikipedia does have some Chinese
        # words in the English Wikipedia.).
        text = self._tokenize_chinese_chars(text)

        orig_tokens = whitespace_tokenize(text)
        split_tokens = []
        for token in orig_tokens:
            if self.do_lower_case:
                token = token.lower()
                token = self._run_strip_accents(token)

        output_tokens = whitespace_tokenize(" ".join(split_tokens))
        return output_tokens

    def _run_strip_accents(self, text):
        """Strips accents from a piece of text."""
        text = unicodedata.normalize("NFD", text)
        output = []
        for char in text:
            cat = unicodedata.category(char)
            if cat == "Mn":
        return "".join(output)

    def _run_split_on_punc(self, text):
        """Splits punctuation on a piece of text."""
        chars = list(text)
        i = 0
        start_new_word = True
        output = []
        while i < len(chars):
            char = chars[i]
            if _is_punctuation(char):
                start_new_word = True
                if start_new_word:
                start_new_word = False
            i += 1

        return ["".join(x) for x in output]

    def _tokenize_chinese_chars(self, text):
        """Adds whitespace around any CJK character."""
        output = []
        for char in text:
            cp = ord(char)
            if self._is_chinese_char(cp):
                output.append(" ")
                output.append(" ")
        return "".join(output)

    def _is_chinese_char(self, cp):
        """Checks whether CP is the codepoint of a CJK character."""
        # This defines a "chinese character" as anything in the CJK Unicode block:
        #     https://en.wikipedia.org/wiki/CJK_Unified_Ideographs_(Unicode_block)
        # Note that the CJK Unicode block is NOT all Japanese and Korean characters,
        # despite its name. The modern Korean Hangul alphabet is a different block,
        # as is Japanese Hiragana and Katakana. Those alphabets are used to write
        # space-separated words, so they are not treated specially and handled
        # like the all of the other languages.
        if ((cp >= 0x4E00 and cp <= 0x9FFF) or    #
                (cp >= 0x3400 and cp <= 0x4DBF) or    #
                (cp >= 0x20000 and cp <= 0x2A6DF) or    #
                (cp >= 0x2A700 and cp <= 0x2B73F) or    #
                (cp >= 0x2B740 and cp <= 0x2B81F) or    #
                (cp >= 0x2B820 and cp <= 0x2CEAF) or
                (cp >= 0xF900 and cp <= 0xFAFF) or    #
                (cp >= 0x2F800 and cp <= 0x2FA1F)):    #
            return True

        return False

    def _clean_text(self, text):
        """Performs invalid character removal and whitespace cleanup on text."""
        output = []
        for char in text:
            cp = ord(char)
            if cp == 0 or cp == 0xfffd or _is_control(char):
            if _is_whitespace(char):
                output.append(" ")
        return "".join(output)

class WordpieceTokenizer(object):
    """Runs WordPiece tokenziation."""

    def __init__(self, vocab, unk_token="[UNK]", max_input_chars_per_word=200):
        self.vocab = vocab
        self.unk_token = unk_token
        self.max_input_chars_per_word = max_input_chars_per_word

    def tokenize(self, text):
        """Tokenizes a piece of text into its word pieces.

        This uses a greedy longest-match-first algorithm to perform tokenization
        using the given vocabulary.

        For example:
            input = "unaffable"
            output = ["un", "##aff", "##able"]

            text: A single token or whitespace separated tokens. This should have
                already been passed through `BasicTokenizer.

            A list of wordpiece tokens.

        text = convert_to_unicode(text)

        output_tokens = []
        for token in whitespace_tokenize(text):
            chars = list(token)
            if len(chars) > self.max_input_chars_per_word:
                output_tokens.append("{}-{}".format(self.unk_token, len(chars)))

            is_bad = False
            start = 0
            sub_tokens = []
            while start < len(chars):
                end = len(chars)
                cur_substr = None
                while start < end:
                    substr = "".join(chars[start:end])
                    if start > 0:
                        substr = "##" + substr
                    if substr in self.vocab:
                        cur_substr = substr
                    end -= 1
                if cur_substr is None:
                    is_bad = True
                start = end

            if is_bad:
                output_tokens.append("{}-{}".format(self.unk_token, len(chars)))
        return output_tokens

def _is_whitespace(char):
    """Checks whether `chars` is a whitespace character."""
    # \t, \n, and \r are technically contorl characters but we treat them
    # as whitespace since they are generally considered as such.
    if char == " " or char == "\t" or char == "\n" or char == "\r":
        return True
    cat = unicodedata.category(char)
    if cat == "Zs":
        return True
    return False

def _is_control(char):
    """Checks whether `chars` is a control character."""
    # These are technically control characters but we count them as whitespace
    # characters.
    if char == "\t" or char == "\n" or char == "\r":
        return False
    cat = unicodedata.category(char)
    if cat.startswith("C"):
        return True
    return False

def _is_punctuation(char):
    """Checks whether `chars` is a punctuation character."""
    cp = ord(char)
    # We treat all non-letter/number ASCII as punctuation.
    # Characters such as "^", "$", and "`" are not in the Unicode
    # Punctuation class but we treat them as punctuation anyways, for
    # consistency.
    if ((cp >= 33 and cp <= 47) or (cp >= 58 and cp <= 64) or
            (cp >= 91 and cp <= 96) or (cp >= 123 and cp <= 126)):
        return True
    cat = unicodedata.category(char)
    if cat.startswith("P"):
        return True
    return False

### modeling.py
class BertConfig(object):
    """Configuration for `BertModel`."""

    def __init__(self,
        """Constructs BertConfig.

            vocab_size: Vocabulary size of `inputs_ids` in `BertModel`.
            hidden_size: Size of the encoder layers and the pooler layer.
            num_hidden_layers: Number of hidden layers in the Transformer encoder.
            num_attention_heads: Number of attention heads for each attention layer in
                the Transformer encoder.
            intermediate_size: The size of the "intermediate" (i.e., feed-forward)
                layer in the Transformer encoder.
            hidden_act: The non-linear activation function (function or string) in the
                encoder and pooler.
            hidden_dropout_prob: The dropout probability for all fully connected
                layers in the embeddings, encoder, and pooler.
            attention_probs_dropout_prob: The dropout ratio for the attention
            max_position_embeddings: The maximum sequence length that this model might
                ever be used with. Typically set this to something large just in case
                (e.g., 512 or 1024 or 2048).
            type_vocab_size: The vocabulary size of the `token_type_ids` passed into
            initializer_range: The stdev of the truncated_normal_initializer for
                initializing all weight matrices.
        self.vocab_size = vocab_size
        self.hidden_size = hidden_size
        self.num_hidden_layers = num_hidden_layers
        self.num_attention_heads = num_attention_heads
        self.hidden_act = hidden_act
        self.intermediate_size = intermediate_size
        self.hidden_dropout_prob = hidden_dropout_prob
        self.attention_probs_dropout_prob = attention_probs_dropout_prob
        self.max_position_embeddings = max_position_embeddings
        self.type_vocab_size = type_vocab_size
        self.initializer_range = initializer_range

    def from_dict(cls, json_object):
        """Constructs a `BertConfig` from a Python dictionary of parameters."""
        config = BertConfig(vocab_size=None)
        for (key, value) in six.iteritems(json_object):
            config.__dict__[key] = value
        return config

    def from_json_file(cls, json_file):
        """Constructs a `BertConfig` from a json file of parameters."""
        with tf.gfile.GFile(json_file, "r") as reader:
            text = reader.read()
        return cls.from_dict(json.loads(text))

    def to_dict(self):
        """Serializes this instance to a Python dictionary."""
        output = copy.deepcopy(self.__dict__)
        return output

    def to_json_string(self):
        """Serializes this instance to a JSON string."""
        return json.dumps(self.to_dict(), indent=2, sort_keys=True) + "\n"

class BertModel(object):
    """BERT model ("Bidirectional Embedding Representations from a Transformer").

    Example usage:

    # Already been converted into WordPiece token ids
    input_ids = tf.constant([[31, 51, 99], [15, 5, 0]])
    input_mask = tf.constant([[1, 1, 1], [1, 1, 0]])
    token_type_ids = tf.constant([[0, 0, 1], [0, 2, 0]])

    config = modeling.BertConfig(vocab_size=32000, hidden_size=512,
        num_hidden_layers=8, num_attention_heads=6, intermediate_size=1024)

    model = modeling.BertModel(config=config, is_training=True,
        input_ids=input_ids, input_mask=input_mask, token_type_ids=token_type_ids)

    label_embeddings = tf.get_variable(...)
    pooled_output = model.get_pooled_output()
    logits = tf.matmul(pooled_output, label_embeddings)

    def __init__(self,
        """Constructor for BertModel.

            config: `BertConfig` instance.
            is_training: bool. rue for training model, false for eval model. Controls
                whether dropout will be applied.
            input_ids: int32 Tensor of shape [batch_size, seq_length].
            input_mask: (optional) int32 Tensor of shape [batch_size, seq_length].
            token_type_ids: (optional) int32 Tensor of shape [batch_size, seq_length].
            use_one_hot_embeddings: (optional) bool. Whether to use one-hot word
                embeddings or tf.embedding_lookup() for the word embeddings. On the TPU,
                it is must faster if this is True, on the CPU or GPU, it is faster if
                this is False.
            scope: (optional) variable scope. Defaults to "bert".

            ValueError: The config is invalid or one of the input tensor shapes
                is invalid.
        config = copy.deepcopy(config)
        if not is_training:
            config.hidden_dropout_prob = 0.0
            config.attention_probs_dropout_prob = 0.0

        input_shape = get_shape_list(input_ids, expected_rank=2)
        batch_size = input_shape[0]
        seq_length = input_shape[1]

        if input_mask is None:
            input_mask = tf.ones(shape=[batch_size, seq_length], dtype=tf.int32)

        if token_type_ids is None:
            token_type_ids = tf.zeros(shape=[batch_size, seq_length], dtype=tf.int32)

        with tf.variable_scope(scope, default_name="bert"):
            with tf.variable_scope("embeddings"):
                # Perform embedding lookup on the word ids.
                (self.embedding_output, self.embedding_table) = embedding_lookup(

                # Add positional embeddings and token type embeddings, then layer
                # normalize and perform dropout.
                self.embedding_output = embedding_postprocessor(

            with tf.variable_scope("encoder"):
                # This converts a 2D mask of shape [batch_size, seq_length] to a 3D
                # mask of shape [batch_size, seq_length, seq_length] which is used
                # for the attention scores.
                attention_mask = create_attention_mask_from_input_mask(
                        input_ids, input_mask)

                # Run the stacked transformer.
                # `sequence_output` shape = [batch_size, seq_length, hidden_size].
                self.all_encoder_layers = transformer_model(

            self.sequence_output = self.all_encoder_layers[-1]
            # The "pooler" converts the encoded sequence tensor of shape
            # [batch_size, seq_length, hidden_size] to a tensor of shape
            # [batch_size, hidden_size]. This is necessary for segment-level
            # (or segment-pair-level) classification tasks where we need a fixed
            # dimensional representation of the segment.
            with tf.variable_scope("pooler"):
                # We "pool" the model by simply taking the hidden state corresponding
                # to the first token. We assume that this has been pre-trained
                first_token_tensor = tf.squeeze(self.sequence_output[:, 0:1, :], axis=1)
                self.pooled_output = tf.layers.dense(

    def get_pooled_output(self):
        return self.pooled_output

    def get_sequence_output(self):
        """Gets final hidden layer of encoder.

            float Tensor of shape [batch_size, seq_length, hidden_size] corresponding
            to the final hidden of the transformer encoder.
        return self.sequence_output

    def get_all_encoder_layers(self):
        return self.all_encoder_layers

    def get_embedding_output(self):
        """Gets output of the embedding lookup (i.e., input to the transformer).

            float Tensor of shape [batch_size, seq_length, hidden_size] corresponding
            to the output of the embedding layer, after summing the word
            embeddings with the positional embeddings and the token type embeddings,
            then performing layer normalization. This is the input to the transformer.
        return self.embedding_output

    def get_embedding_table(self):
        return self.embedding_table

def gelu(input_tensor):
    """Gaussian Error Linear Unit.

    This is a smoother version of the RELU.
    Original paper: https://arxiv.org/abs/1606.08415

        input_tensor: float Tensor to perform activation.

        `input_tensor` with the GELU activation applied.
    cdf = 0.5 * (1.0 + tf.erf(input_tensor / tf.sqrt(2.0)))
    return input_tensor * cdf

def get_activation(activation_string):
    """Maps a string to a Python function, e.g., "relu" => `tf.nn.relu`.

        activation_string: String name of the activation function.

        A Python function corresponding to the activation function. If
        `activation_string` is None, empty, or "linear", this will return None.
        If `activation_string` is not a string, it will return `activation_string`.

        ValueError: The `activation_string` does not correspond to a known

    # We assume that anything that"s not a string is already an activation
    # function, so we just return it.
    if not isinstance(activation_string, six.string_types):
        return activation_string

    if not activation_string:
        return None

    act = activation_string.lower()
    if act == "linear":
        return None
    elif act == "relu":
        return tf.nn.relu
    elif act == "gelu":
        return gelu
    elif act == "tanh":
        return tf.tanh
        raise ValueError("Unsupported activation: %s" % act)

def get_assignment_map_from_checkpoint(tvars, init_checkpoint):
    """Compute the union of the current variables and checkpoint variables."""
    assignment_map = {}
    initialized_variable_names = {}

    name_to_variable = collections.OrderedDict()
    for var in tvars:
        name = var.name
        m = re.match("^(.*):\\d+$", name)
        if m is not None:
            name = m.group(1)
        name_to_variable[name] = var

    init_vars = tf.train.list_variables(init_checkpoint)

    assignment_map = collections.OrderedDict()
    for x in init_vars:
        (name, var) = (x[0], x[1])
        if name not in name_to_variable:
        assignment_map[name] = name
        initialized_variable_names[name] = 1
        initialized_variable_names[name + ":0"] = 1

    return (assignment_map, initialized_variable_names)

def dropout(input_tensor, dropout_prob):
    """Perform dropout.

        input_tensor: float Tensor.
        dropout_prob: Python float. The probability of dropping out a value (NOT of
            *keeping* a dimension as in `tf.nn.dropout`).

        A version of `input_tensor` with dropout applied.
    if dropout_prob is None or dropout_prob == 0.0:
        return input_tensor

    output = tf.nn.dropout(input_tensor, 1.0 - dropout_prob)
    return output

def layer_norm(input_tensor, name=None):
    """Run layer normalization on the last dimension of the tensor."""
    return tf.contrib.layers.layer_norm(
            inputs=input_tensor, begin_norm_axis=-1, begin_params_axis=-1, scope=name)

def layer_norm_and_dropout(input_tensor, dropout_prob, name=None):
    """Runs layer normalization followed by dropout."""
    output_tensor = layer_norm(input_tensor, name)
    output_tensor = dropout(output_tensor, dropout_prob)
    return output_tensor

def create_initializer(initializer_range=0.02):
    """Creates a `truncated_normal_initializer` with the given range."""
    return tf.truncated_normal_initializer(stddev=initializer_range)

def embedding_lookup(input_ids,
    """Looks up words embeddings for id tensor.

        input_ids: int32 Tensor of shape [batch_size, seq_length] containing word
        vocab_size: int. Size of the embedding vocabulary.
        embedding_size: int. Width of the word embeddings.
        initializer_range: float. Embedding initialization range.
        word_embedding_name: string. Name of the embedding table.
        use_one_hot_embeddings: bool. If True, use one-hot method for word
            embeddings. If False, use `tf.nn.embedding_lookup()`. One hot is better
            for TPUs.

        float Tensor of shape [batch_size, seq_length, embedding_size].
    # This function assumes that the input is of shape [batch_size, seq_length,
    # num_inputs].
    # If the input is a 2D tensor of shape [batch_size, seq_length], we
    # reshape to [batch_size, seq_length, 1].
    if input_ids.shape.ndims == 2:
        input_ids = tf.expand_dims(input_ids, axis=[-1])

    embedding_table = tf.get_variable(
            shape=[vocab_size, embedding_size],

    if use_one_hot_embeddings:
        flat_input_ids = tf.reshape(input_ids, [-1])
        one_hot_input_ids = tf.one_hot(flat_input_ids, depth=vocab_size)
        output = tf.matmul(one_hot_input_ids, embedding_table)
        output = tf.nn.embedding_lookup(embedding_table, input_ids)

    input_shape = get_shape_list(input_ids)

    output = tf.reshape(output,
                                            input_shape[0:-1] + [input_shape[-1] * embedding_size])
    return (output, embedding_table)

def embedding_postprocessor(input_tensor,
    """Performs various post-processing on a word embedding tensor.

        input_tensor: float Tensor of shape [batch_size, seq_length,
        use_token_type: bool. Whether to add embeddings for `token_type_ids`.
        token_type_ids: (optional) int32 Tensor of shape [batch_size, seq_length].
            Must be specified if `use_token_type` is True.
        token_type_vocab_size: int. The vocabulary size of `token_type_ids`.
        token_type_embedding_name: string. The name of the embedding table variable
            for token type ids.
        use_position_embeddings: bool. Whether to add position embeddings for the
            position of each token in the sequence.
        position_embedding_name: string. The name of the embedding table variable
            for positional embeddings.
        initializer_range: float. Range of the weight initialization.
        max_position_embeddings: int. Maximum sequence length that might ever be
            used with this model. This can be longer than the sequence length of
            input_tensor, but cannot be shorter.
        dropout_prob: float. Dropout probability applied to the final output tensor.

        float tensor with same shape as `input_tensor`.

        ValueError: One of the tensor shapes or input values is invalid.
    input_shape = get_shape_list(input_tensor, expected_rank=3)
    batch_size = input_shape[0]
    seq_length = input_shape[1]
    width = input_shape[2]

    output = input_tensor

    if use_token_type:
        if token_type_ids is None:
            raise ValueError("`token_type_ids` must be specified if"
                                             "`use_token_type` is True.")
        token_type_table = tf.get_variable(
                shape=[token_type_vocab_size, width],
        # This vocab will be small so we always do one-hot here, since it is always
        # faster for a small vocabulary.
        flat_token_type_ids = tf.reshape(token_type_ids, [-1])
        one_hot_ids = tf.one_hot(flat_token_type_ids, depth=token_type_vocab_size)
        token_type_embeddings = tf.matmul(one_hot_ids, token_type_table)
        token_type_embeddings = tf.reshape(token_type_embeddings,
                                                                             [batch_size, seq_length, width])
        output += token_type_embeddings

    if use_position_embeddings:
        assert_op = tf.assert_less_equal(seq_length, max_position_embeddings)
        with tf.control_dependencies([assert_op]):
            full_position_embeddings = tf.get_variable(
                    shape=[max_position_embeddings, width],
            # Since the position embedding table is a learned variable, we create it
            # using a (long) sequence length `max_position_embeddings`. The actual
            # sequence length might be shorter than this, for faster training of
            # tasks that do not have long sequences.
            # So `full_position_embeddings` is effectively an embedding table
            # for position [0, 1, 2, ..., max_position_embeddings-1], and the current
            # sequence has positions [0, 1, 2, ... seq_length-1], so we can just
            # perform a slice.
            position_embeddings = tf.slice(full_position_embeddings, [0, 0],
                                                                         [seq_length, -1])
            num_dims = len(output.shape.as_list())

            # Only the last two dimensions are relevant (`seq_length` and `width`), so
            # we broadcast among the first dimensions, which is typically just
            # the batch size.
            position_broadcast_shape = []
            for _ in range(num_dims - 2):
            position_broadcast_shape.extend([seq_length, width])
            position_embeddings = tf.reshape(position_embeddings,
            output += position_embeddings

    output = layer_norm_and_dropout(output, dropout_prob)
    return output

def create_attention_mask_from_input_mask(from_tensor, to_mask):
    """Create 3D attention mask from a 2D tensor mask.

        from_tensor: 2D or 3D Tensor of shape [batch_size, from_seq_length, ...].
        to_mask: int32 Tensor of shape [batch_size, to_seq_length].

        float Tensor of shape [batch_size, from_seq_length, to_seq_length].
    from_shape = get_shape_list(from_tensor, expected_rank=[2, 3])
    batch_size = from_shape[0]
    from_seq_length = from_shape[1]

    to_shape = get_shape_list(to_mask, expected_rank=2)
    to_seq_length = to_shape[1]

    to_mask = tf.cast(
            tf.reshape(to_mask, [batch_size, 1, to_seq_length]), tf.float32)

    # We don't assume that `from_tensor` is a mask (although it could be). We
    # don't actually care if we attend *from* padding tokens (only *to* padding)
    # tokens so we create a tensor of all ones.
    # `broadcast_ones` = [batch_size, from_seq_length, 1]
    broadcast_ones = tf.ones(
            shape=[batch_size, from_seq_length, 1], dtype=tf.float32)

    # Here we broadcast along two dimensions to create the mask.
    mask = broadcast_ones * to_mask

    return mask

def attention_layer(from_tensor,
    """Performs multi-headed attention from `from_tensor` to `to_tensor`.

    This is an implementation of multi-headed attention based on "Attention
    is all you Need". If `from_tensor` and `to_tensor` are the same, then
    this is self-attention. Each timestep in `from_tensor` attends to the
    corresponding sequence in `to_tensor`, and returns a fixed-with vector.

    This function first projects `from_tensor` into a "query" tensor and
    `to_tensor` into "key" and "value" tensors. These are (effectively) a list
    of tensors of length `num_attention_heads`, where each tensor is of shape
    [batch_size, seq_length, size_per_head].

    Then, the query and key tensors are dot-producted and scaled. These are
    softmaxed to obtain attention probabilities. The value tensors are then
    interpolated by these probabilities, then concatenated back to a single
    tensor and returned.

    In practice, the multi-headed attention are done with transposes and
    reshapes rather than actual separate tensors.

        from_tensor: float Tensor of shape [batch_size, from_seq_length,
        to_tensor: float Tensor of shape [batch_size, to_seq_length, to_width].
        attention_mask: (optional) int32 Tensor of shape [batch_size,
            from_seq_length, to_seq_length]. The values should be 1 or 0. The
            attention scores will effectively be set to -infinity for any positions in
            the mask that are 0, and will be unchanged for positions that are 1.
        num_attention_heads: int. Number of attention heads.
        size_per_head: int. Size of each attention head.
        query_act: (optional) Activation function for the query transform.
        key_act: (optional) Activation function for the key transform.
        value_act: (optional) Activation function for the value transform.
        attention_probs_dropout_prob: (optional) float. Dropout probability of the
            attention probabilities.
        initializer_range: float. Range of the weight initializer.
        do_return_2d_tensor: bool. If True, the output will be of shape [batch_size
            * from_seq_length, num_attention_heads * size_per_head]. If False, the
            output will be of shape [batch_size, from_seq_length, num_attention_heads
            * size_per_head].
        batch_size: (Optional) int. If the input is 2D, this might be the batch size
            of the 3D version of the `from_tensor` and `to_tensor`.
        from_seq_length: (Optional) If the input is 2D, this might be the seq length
            of the 3D version of the `from_tensor`.
        to_seq_length: (Optional) If the input is 2D, this might be the seq length
            of the 3D version of the `to_tensor`.

        float Tensor of shape [batch_size, from_seq_length,
            num_attention_heads * size_per_head]. (If `do_return_2d_tensor` is
            true, this will be of shape [batch_size * from_seq_length,
            num_attention_heads * size_per_head]).

        ValueError: Any of the arguments or tensor shapes are invalid.

    def transpose_for_scores(input_tensor, batch_size, num_attention_heads,
                                                     seq_length, width):
        output_tensor = tf.reshape(
                input_tensor, [batch_size, seq_length, num_attention_heads, width])

        output_tensor = tf.transpose(output_tensor, [0, 2, 1, 3])
        return output_tensor

    from_shape = get_shape_list(from_tensor, expected_rank=[2, 3])
    to_shape = get_shape_list(to_tensor, expected_rank=[2, 3])

    if len(from_shape) != len(to_shape):
        raise ValueError(
                "The rank of `from_tensor` must match the rank of `to_tensor`.")

    if len(from_shape) == 3:
        batch_size = from_shape[0]
        from_seq_length = from_shape[1]
        to_seq_length = to_shape[1]
    elif len(from_shape) == 2:
        if (batch_size is None or from_seq_length is None or to_seq_length is None):
            raise ValueError(
                    "When passing in rank 2 tensors to attention_layer, the values "
                    "for `batch_size`, `from_seq_length`, and `to_seq_length` "
                    "must all be specified.")

    # Scalar dimensions referenced here:
    #     B = batch size (number of sequences)
    #     F = `from_tensor` sequence length
    #     T = `to_tensor` sequence length
    #     N = `num_attention_heads`
    #     H = `size_per_head`

    from_tensor_2d = reshape_to_matrix(from_tensor)
    to_tensor_2d = reshape_to_matrix(to_tensor)

    # `query_layer` = [B*F, N*H]
    query_layer = tf.layers.dense(
            num_attention_heads * size_per_head,

    # `key_layer` = [B*T, N*H]
    key_layer = tf.layers.dense(
            num_attention_heads * size_per_head,

    # `value_layer` = [B*T, N*H]
    value_layer = tf.layers.dense(
            num_attention_heads * size_per_head,

    # `query_layer` = [B, N, F, H]
    query_layer = transpose_for_scores(query_layer, batch_size,
                                                                         num_attention_heads, from_seq_length,

    # `key_layer` = [B, N, T, H]
    key_layer = transpose_for_scores(key_layer, batch_size, num_attention_heads,
                                                                     to_seq_length, size_per_head)

    # Take the dot product between "query" and "key" to get the raw
    # attention scores.
    # `attention_scores` = [B, N, F, T]
    attention_scores = tf.matmul(query_layer, key_layer, transpose_b=True)
    attention_scores = tf.multiply(attention_scores,
                                                                 1.0 / math.sqrt(float(size_per_head)))

    if attention_mask is not None:
        # `attention_mask` = [B, 1, F, T]
        attention_mask = tf.expand_dims(attention_mask, axis=[1])

        # Since attention_mask is 1.0 for positions we want to attend and 0.0 for
        # masked positions, this operation will create a tensor which is 0.0 for
        # positions we want to attend and -10000.0 for masked positions.
        adder = (1.0 - tf.cast(attention_mask, tf.float32)) * -10000.0

        # Since we are adding it to the raw scores before the softmax, this is
        # effectively the same as removing these entirely.
        attention_scores += adder

    # Normalize the attention scores to probabilities.
    # `attention_probs` = [B, N, F, T]
    attention_probs = tf.nn.softmax(attention_scores)

    # This is actually dropping out entire tokens to attend to, which might
    # seem a bit unusual, but is taken from the original Transformer paper.
    attention_probs = dropout(attention_probs, attention_probs_dropout_prob)

    # `value_layer` = [B, T, N, H]
    value_layer = tf.reshape(
            [batch_size, to_seq_length, num_attention_heads, size_per_head])

    # `value_layer` = [B, N, T, H]
    value_layer = tf.transpose(value_layer, [0, 2, 1, 3])

    # `context_layer` = [B, N, F, H]
    context_layer = tf.matmul(attention_probs, value_layer)

    # `context_layer` = [B, F, N, H]
    context_layer = tf.transpose(context_layer, [0, 2, 1, 3])

    if do_return_2d_tensor:
        # `context_layer` = [B*F, N*V]
        context_layer = tf.reshape(
                [batch_size * from_seq_length, num_attention_heads * size_per_head])
        # `context_layer` = [B, F, N*V]
        context_layer = tf.reshape(
                [batch_size, from_seq_length, num_attention_heads * size_per_head])

    return context_layer

def transformer_model(input_tensor,
    """Multi-headed, multi-layer Transformer from "Attention is All You Need".

    This is almost an exact implementation of the original Transformer encoder.

    See the original paper:

    Also see:

        input_tensor: float Tensor of shape [batch_size, seq_length, hidden_size].
        attention_mask: (optional) int32 Tensor of shape [batch_size, seq_length,
            seq_length], with 1 for positions that can be attended to and 0 in
            positions that should not be.
        hidden_size: int. Hidden size of the Transformer.
        num_hidden_layers: int. Number of layers (blocks) in the Transformer.
        num_attention_heads: int. Number of attention heads in the Transformer.
        intermediate_size: int. The size of the "intermediate" (a.k.a., feed
            forward) layer.
        intermediate_act_fn: function. The non-linear activation function to apply
            to the output of the intermediate/feed-forward layer.
        hidden_dropout_prob: float. Dropout probability for the hidden layers.
        attention_probs_dropout_prob: float. Dropout probability of the attention
        initializer_range: float. Range of the initializer (stddev of truncated
        do_return_all_layers: Whether to also return all layers or just the final

        float Tensor of shape [batch_size, seq_length, hidden_size], the final
        hidden layer of the Transformer.

        ValueError: A Tensor shape or parameter is invalid.
    if hidden_size % num_attention_heads != 0:
        raise ValueError(
                "The hidden size (%d) is not a multiple of the number of attention "
                "heads (%d)" % (hidden_size, num_attention_heads))

    attention_head_size = int(hidden_size / num_attention_heads)
    input_shape = get_shape_list(input_tensor, expected_rank=3)
    batch_size = input_shape[0]
    seq_length = input_shape[1]
    input_width = input_shape[2]

    # The Transformer performs sum residuals on all layers so the input needs
    # to be the same as the hidden size.
    if input_width != hidden_size:
        raise ValueError("The width of the input tensor (%d) != hidden size (%d)" %
                                         (input_width, hidden_size))

    # We keep the representation as a 2D tensor to avoid re-shaping it back and
    # forth from a 3D tensor to a 2D tensor. Re-shapes are normally free on
    # the GPU/CPU but may not be free on the TPU, so we want to minimize them to
    # help the optimizer.
    prev_output = reshape_to_matrix(input_tensor)

    all_layer_outputs = []
    for layer_idx in range(num_hidden_layers):
        with tf.variable_scope("layer_%d" % layer_idx):
            layer_input = prev_output

            with tf.variable_scope("attention"):
                attention_heads = []
                with tf.variable_scope("self"):
                    attention_head = attention_layer(

                attention_output = None
                if len(attention_heads) == 1:
                    attention_output = attention_heads[0]
                    # In the case where we have other sequences, we just concatenate
                    # them to the self-attention head before the projection.
                    attention_output = tf.concat(attention_heads, axis=-1)

                # Run a linear projection of `hidden_size` then add a residual
                # with `layer_input`.
                with tf.variable_scope("output"):
                    attention_output = tf.layers.dense(
                    attention_output = dropout(attention_output, hidden_dropout_prob)
                    attention_output = layer_norm(attention_output + layer_input)

            # The activation is only applied to the "intermediate" hidden layer.
            with tf.variable_scope("intermediate"):
                intermediate_output = tf.layers.dense(

            # Down-project back to `hidden_size` then add the residual.
            with tf.variable_scope("output"):
                layer_output = tf.layers.dense(
                layer_output = dropout(layer_output, hidden_dropout_prob)
                layer_output = layer_norm(layer_output + attention_output)
                prev_output = layer_output

    if do_return_all_layers:
        final_outputs = []
        for layer_output in all_layer_outputs:
            final_output = reshape_from_matrix(layer_output, input_shape)
        return final_outputs
        final_output = reshape_from_matrix(prev_output, input_shape)
        return final_output

def get_shape_list(tensor, expected_rank=None, name=None):
    """Returns a list of the shape of tensor, preferring static dimensions.

        tensor: A tf.Tensor object to find the shape of.
        expected_rank: (optional) int. The expected rank of `tensor`. If this is
            specified and the `tensor` has a different rank, and exception will be
        name: Optional name of the tensor for the error message.

        A list of dimensions of the shape of tensor. All static dimensions will
        be returned as python integers, and dynamic dimensions will be returned
        as tf.Tensor scalars.
    if name is None:
        name = tensor.name

    if expected_rank is not None:
        assert_rank(tensor, expected_rank, name)

    shape = tensor.shape.as_list()

    non_static_indexes = []
    for (index, dim) in enumerate(shape):
        if dim is None:

    if not non_static_indexes:
        return shape

    dyn_shape = tf.shape(tensor)
    for index in non_static_indexes:
        shape[index] = dyn_shape[index]
    return shape

def reshape_to_matrix(input_tensor):
    """Reshapes a >= rank 2 tensor to a rank 2 tensor (i.e., a matrix)."""
    ndims = input_tensor.shape.ndims
    if ndims < 2:
        raise ValueError("Input tensor must have at least rank 2. Shape = %s" %
    if ndims == 2:
        return input_tensor

    width = input_tensor.shape[-1]
    output_tensor = tf.reshape(input_tensor, [-1, width])
    return output_tensor

def reshape_from_matrix(output_tensor, orig_shape_list):
    """Reshapes a rank 2 tensor back to its original rank >= 2 tensor."""
    if len(orig_shape_list) == 2:
        return output_tensor

    output_shape = get_shape_list(output_tensor)

    orig_dims = orig_shape_list[0:-1]
    width = output_shape[-1]

    return tf.reshape(output_tensor, orig_dims + [width])

def assert_rank(tensor, expected_rank, name=None):
    """Raises an exception if the tensor rank is not of the expected rank.

        tensor: A tf.Tensor to check the rank of.
        expected_rank: Python integer or list of integers, expected rank.
        name: Optional name of the tensor for the error message.

        ValueError: If the expected shape doesn't match the actual shape.
    if name is None:
        name = tensor.name

    expected_rank_dict = {}
    if isinstance(expected_rank, six.integer_types):
        expected_rank_dict[expected_rank] = True
        for x in expected_rank:
            expected_rank_dict[x] = True

    actual_rank = tensor.shape.ndims
    if actual_rank not in expected_rank_dict:
        scope_name = tf.get_variable_scope().name
        raise ValueError(
                "For the tensor `%s` in scope `%s`, the actual rank "
                "`%d` (shape = %s) is not equal to the expected rank `%s`" %
                (name, scope_name, actual_rank, str(tensor.shape), str(expected_rank)))