from __future__ import absolute_import, division, print_function import collections import math import numpy as np import torch from pytorch_transformers.tokenization_bert import (BasicTokenizer, whitespace_tokenize) from torch.utils.data import DataLoader, SequentialSampler, TensorDataset class SquadExample(object): """ A single training/test example for the Squad dataset. For examples without an answer, the start and end position are -1. """ def __init__(self, qas_id, question_text, doc_tokens, orig_answer_text=None, start_position=None, end_position=None): self.qas_id = qas_id self.question_text = question_text self.doc_tokens = doc_tokens self.orig_answer_text = orig_answer_text self.start_position = start_position self.end_position = end_position def __str__(self): return self.__repr__() def __repr__(self): s = "" s += "qas_id: %s" % (self.qas_id) s += ", question_text: %s" % ( self.question_text) s += ", doc_tokens: [%s]" % (" ".join(self.doc_tokens)) if self.start_position: s += ", start_position: %d" % (self.start_position) if self.end_position: s += ", end_position: %d" % (self.end_position) return s class InputFeatures(object): """A single set of features of data.""" def __init__(self, unique_id, example_index, doc_span_index, tokens, token_to_orig_map, token_is_max_context, input_ids, input_mask, segment_ids, paragraph_len, start_position=None, end_position=None,): self.unique_id = unique_id self.example_index = example_index self.doc_span_index = doc_span_index self.tokens = tokens self.token_to_orig_map = token_to_orig_map self.token_is_max_context = token_is_max_context self.input_ids = input_ids self.input_mask = input_mask self.segment_ids = segment_ids self.paragraph_len = paragraph_len self.start_position = start_position self.end_position = end_position def input_to_squad_example(passage, question): """Convert input passage and question into a SquadExample.""" def is_whitespace(c): if c == " " or c == "\t" or c == "\r" or c == "\n" or ord(c) == 0x202F: return True return False paragraph_text = passage doc_tokens = [] char_to_word_offset = [] prev_is_whitespace = True for c in paragraph_text: if is_whitespace(c): prev_is_whitespace = True else: if prev_is_whitespace: doc_tokens.append(c) else: doc_tokens[-1] += c prev_is_whitespace = False char_to_word_offset.append(len(doc_tokens) - 1) qas_id = 0 question_text = question start_position = None end_position = None orig_answer_text = None example = SquadExample( qas_id=qas_id, question_text=question_text, doc_tokens=doc_tokens, orig_answer_text=orig_answer_text, start_position=start_position, end_position=end_position) return example def _check_is_max_context(doc_spans, cur_span_index, position): """Check if this is the 'max context' doc span for the token.""" # Because of the sliding window approach taken to scoring documents, a single # token can appear in multiple documents. E.g. # Doc: the man went to the store and bought a gallon of milk # Span A: the man went to the # Span B: to the store and bought # Span C: and bought a gallon of # ... # # Now the word 'bought' will have two scores from spans B and C. We only # want to consider the score with "maximum context", which we define as # the *minimum* of its left and right context (the *sum* of left and # right context will always be the same, of course). # # In the example the maximum context for 'bought' would be span C since # it has 1 left context and 3 right context, while span B has 4 left context # and 0 right context. best_score = None best_span_index = None for (span_index, doc_span) in enumerate(doc_spans): end = doc_span.start + doc_span.length - 1 if position < doc_span.start: continue if position > end: continue num_left_context = position - doc_span.start num_right_context = end - position score = min(num_left_context, num_right_context) + 0.01 * doc_span.length if best_score is None or score > best_score: best_score = score best_span_index = span_index return cur_span_index == best_span_index def squad_examples_to_features(example, tokenizer, max_seq_length, doc_stride, max_query_length,cls_token_at_end=False, cls_token='[CLS]', sep_token='[SEP]', pad_token=0, sequence_a_segment_id=0, sequence_b_segment_id=1, cls_token_segment_id=0, pad_token_segment_id=0, mask_padding_with_zero=True): """Loads a data file into a list of `InputBatch`s.""" unique_id = 1000000000 # cnt_pos, cnt_neg = 0, 0 # max_N, max_M = 1024, 1024 # f = np.zeros((max_N, max_M), dtype=np.float32) example_index = 0 features = [] # if example_index % 100 == 0: # logger.info('Converting %s/%s pos %s neg %s', example_index, len(examples), cnt_pos, cnt_neg) query_tokens = tokenizer.tokenize(example.question_text) if len(query_tokens) > max_query_length: query_tokens = query_tokens[0:max_query_length] tok_to_orig_index = [] orig_to_tok_index = [] all_doc_tokens = [] for (i, token) in enumerate(example.doc_tokens): orig_to_tok_index.append(len(all_doc_tokens)) sub_tokens = tokenizer.tokenize(token) for sub_token in sub_tokens: tok_to_orig_index.append(i) all_doc_tokens.append(sub_token) # The -3 accounts for [CLS], [SEP] and [SEP] max_tokens_for_doc = max_seq_length - len(query_tokens) - 3 # We can have documents that are longer than the maximum sequence length. # To deal with this we do a sliding window approach, where we take chunks # of the up to our max length with a stride of `doc_stride`. _DocSpan = collections.namedtuple( # pylint: disable=invalid-name "DocSpan", ["start", "length"]) doc_spans = [] start_offset = 0 while start_offset < len(all_doc_tokens): length = len(all_doc_tokens) - start_offset if length > max_tokens_for_doc: length = max_tokens_for_doc doc_spans.append(_DocSpan(start=start_offset, length=length)) if start_offset + length == len(all_doc_tokens): break start_offset += min(length, doc_stride) for (doc_span_index, doc_span) in enumerate(doc_spans): tokens = [] token_to_orig_map = {} token_is_max_context = {} segment_ids = [] # CLS token at the beginning if not cls_token_at_end: tokens.append(cls_token) segment_ids.append(cls_token_segment_id) # Query for token in query_tokens: tokens.append(token) segment_ids.append(sequence_a_segment_id) # SEP token tokens.append(sep_token) segment_ids.append(sequence_a_segment_id) # Paragraph for i in range(doc_span.length): split_token_index = doc_span.start + i token_to_orig_map[len(tokens)] = tok_to_orig_index[split_token_index] is_max_context = _check_is_max_context(doc_spans, doc_span_index, split_token_index) token_is_max_context[len(tokens)] = is_max_context tokens.append(all_doc_tokens[split_token_index]) segment_ids.append(sequence_b_segment_id) paragraph_len = doc_span.length # SEP token tokens.append(sep_token) segment_ids.append(sequence_b_segment_id) # CLS token at the end if cls_token_at_end: tokens.append(cls_token) segment_ids.append(cls_token_segment_id) input_ids = tokenizer.convert_tokens_to_ids(tokens) # The mask has 1 for real tokens and 0 for padding tokens. Only real # tokens are attended to. input_mask = [1 if mask_padding_with_zero else 0] * len(input_ids) # Zero-pad up to the sequence length. while len(input_ids) < max_seq_length: input_ids.append(pad_token) input_mask.append(0 if mask_padding_with_zero else 1) segment_ids.append(pad_token_segment_id) assert len(input_ids) == max_seq_length assert len(input_mask) == max_seq_length assert len(segment_ids) == max_seq_length start_position = None end_position = None features.append( InputFeatures( unique_id=unique_id, example_index=example_index, doc_span_index=doc_span_index, tokens=tokens, token_to_orig_map=token_to_orig_map, token_is_max_context=token_is_max_context, input_ids=input_ids, input_mask=input_mask, segment_ids=segment_ids, paragraph_len=paragraph_len, start_position=start_position, end_position=end_position)) unique_id += 1 return features def to_list(tensor): return tensor.detach().cpu().tolist() def _get_best_indexes(logits, n_best_size): """Get the n-best logits from a list.""" index_and_score = sorted(enumerate(logits), key=lambda x: x[1], reverse=True) best_indexes = [] for i in range(len(index_and_score)): if i >= n_best_size: break best_indexes.append(index_and_score[i][0]) return best_indexes RawResult = collections.namedtuple("RawResult",["unique_id", "start_logits", "end_logits"]) def get_final_text(pred_text, orig_text, do_lower_case, verbose_logging=False): """Project the tokenized prediction back to the original text.""" # When we created the data, we kept track of the alignment between original # (whitespace tokenized) tokens and our WordPiece tokenized tokens. So # now `orig_text` contains the span of our original text corresponding to the # span that we predicted. # # However, `orig_text` may contain extra characters that we don't want in # our prediction. # # For example, let's say: # pred_text = steve smith # orig_text = Steve Smith's # # We don't want to return `orig_text` because it contains the extra "'s". # # We don't want to return `pred_text` because it's already been normalized # (the SQuAD eval script also does punctuation stripping/lower casing but # our tokenizer does additional normalization like stripping accent # characters). # # What we really want to return is "Steve Smith". # # Therefore, we have to apply a semi-complicated alignment heuristic between # `pred_text` and `orig_text` to get a character-to-character alignment. This # can fail in certain cases in which case we just return `orig_text`. def _strip_spaces(text): ns_chars = [] ns_to_s_map = collections.OrderedDict() for (i, c) in enumerate(text): if c == " ": continue ns_to_s_map[len(ns_chars)] = i ns_chars.append(c) ns_text = "".join(ns_chars) return (ns_text, ns_to_s_map) # We first tokenize `orig_text`, strip whitespace from the result # and `pred_text`, and check if they are the same length. If they are # NOT the same length, the heuristic has failed. If they are the same # length, we assume the characters are one-to-one aligned. tokenizer = BasicTokenizer(do_lower_case=do_lower_case) tok_text = " ".join(tokenizer.tokenize(orig_text)) start_position = tok_text.find(pred_text) if start_position == -1: return orig_text end_position = start_position + len(pred_text) - 1 (orig_ns_text, orig_ns_to_s_map) = _strip_spaces(orig_text) (tok_ns_text, tok_ns_to_s_map) = _strip_spaces(tok_text) if len(orig_ns_text) != len(tok_ns_text): return orig_text # We then project the characters in `pred_text` back to `orig_text` using # the character-to-character alignment. tok_s_to_ns_map = {} for (i, tok_index) in tok_ns_to_s_map.items(): tok_s_to_ns_map[tok_index] = i orig_start_position = None if start_position in tok_s_to_ns_map: ns_start_position = tok_s_to_ns_map[start_position] if ns_start_position in orig_ns_to_s_map: orig_start_position = orig_ns_to_s_map[ns_start_position] if orig_start_position is None: return orig_text orig_end_position = None if end_position in tok_s_to_ns_map: ns_end_position = tok_s_to_ns_map[end_position] if ns_end_position in orig_ns_to_s_map: orig_end_position = orig_ns_to_s_map[ns_end_position] if orig_end_position is None: return orig_text output_text = orig_text[orig_start_position:(orig_end_position + 1)] return output_text def _compute_softmax(scores): """Compute softmax probability over raw logits.""" if not scores: return [] max_score = None for score in scores: if max_score is None or score > max_score: max_score = score exp_scores = [] total_sum = 0.0 for score in scores: x = math.exp(score - max_score) exp_scores.append(x) total_sum += x probs = [] for score in exp_scores: probs.append(score / total_sum) return probs def get_answer(example, features, all_results, n_best_size, max_answer_length, do_lower_case): example_index_to_features = collections.defaultdict(list) for feature in features: example_index_to_features[feature.example_index].append(feature) unique_id_to_result = {} for result in all_results: unique_id_to_result[result.unique_id] = result _PrelimPrediction = collections.namedtuple( "PrelimPrediction",["feature_index", "start_index", "end_index", "start_logit", "end_logit"]) example_index = 0 features = example_index_to_features[example_index] prelim_predictions = [] for (feature_index, feature) in enumerate(features): result = unique_id_to_result[feature.unique_id] start_indexes = _get_best_indexes(result.start_logits, n_best_size) end_indexes = _get_best_indexes(result.end_logits, n_best_size) for start_index in start_indexes: for end_index in end_indexes: # We could hypothetically create invalid predictions, e.g., predict # that the start of the span is in the question. We throw out all # invalid predictions. if start_index >= len(feature.tokens): continue if end_index >= len(feature.tokens): continue if start_index not in feature.token_to_orig_map: continue if end_index not in feature.token_to_orig_map: continue if not feature.token_is_max_context.get(start_index, False): continue if end_index < start_index: continue length = end_index - start_index + 1 if length > max_answer_length: continue prelim_predictions.append( _PrelimPrediction( feature_index=feature_index, start_index=start_index, end_index=end_index, start_logit=result.start_logits[start_index], end_logit=result.end_logits[end_index])) prelim_predictions = sorted(prelim_predictions,key=lambda x: (x.start_logit + x.end_logit),reverse=True) _NbestPrediction = collections.namedtuple("NbestPrediction", ["text", "start_logit", "end_logit","start_index","end_index"]) seen_predictions = {} nbest = [] for pred in prelim_predictions: if len(nbest) >= n_best_size: break feature = features[pred.feature_index] orig_doc_start = -1 orig_doc_end = -1 if pred.start_index > 0: # this is a non-null prediction tok_tokens = feature.tokens[pred.start_index:(pred.end_index + 1)] orig_doc_start = feature.token_to_orig_map[pred.start_index] orig_doc_end = feature.token_to_orig_map[pred.end_index] orig_tokens = example.doc_tokens[orig_doc_start:(orig_doc_end + 1)] tok_text = " ".join(tok_tokens) # De-tokenize WordPieces that have been split off. tok_text = tok_text.replace(" ##", "") tok_text = tok_text.replace("##", "") # Clean whitespace tok_text = tok_text.strip() tok_text = " ".join(tok_text.split()) orig_text = " ".join(orig_tokens) final_text = get_final_text(tok_text, orig_text,do_lower_case) if final_text in seen_predictions: continue seen_predictions[final_text] = True else: final_text = "" seen_predictions[final_text] = True nbest.append( _NbestPrediction( text=final_text, start_logit=pred.start_logit, end_logit=pred.end_logit, start_index=orig_doc_start, end_index=orig_doc_end)) if not nbest: nbest.append(_NbestPrediction(text="empty", start_logit=0.0, end_logit=0.0,start_index=-1, end_index=-1)) assert len(nbest) >= 1 total_scores = [] for entry in nbest: total_scores.append(entry.start_logit + entry.end_logit) probs = _compute_softmax(total_scores) answer = {"answer" : nbest[0].text, "start" : nbest[0].start_index, "end" : nbest[0].end_index, "confidence" : probs[0], "document" : example.doc_tokens } return answer