import argparse import collections import json import math import os import sys import numpy as np import onnxruntime as onnxrt import six from tokenizers import BertWordPieceTokenizer from tokenizers import pre_tokenizers RawResult = collections.namedtuple("RawResult", ["unique_id", "start_logits", "end_logits"]) Feature = collections.namedtuple("Feature", [ "unique_id", "tokens", "example_index", "token_to_orig_map", "token_is_max_context" ]) class SquadExample(object): 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.append("qas_id: %s" % (self.qas_id)) s.append("question_text: %s" % (self.question_text)) s.append("doc_tokens: [%s]" % (" ".join(self.doc_tokens))) if self.start_position: s.append("start_position: %d" % (self.start_position)) if self.start_position: s.append("end_position: %d" % (self.end_position)) return ", ".join(s) def check_is_max_context(doc_spans, cur_span_index, position): 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 convert_examples_to_features(examples, tokenizer, max_seq_length, doc_stride, max_query_length): res_input_ids = [] res_input_mask = [] res_segment_ids = [] extra = [] unique_id = 0 for (example_index, example) in enumerate(examples): # 对原始问题文本进行数据处理 query_tokens = tokenizer.encode(example.question_text) if len(query_tokens) > max_query_length: query_tokens = query_tokens[0:max_query_length] # 对上下文文本进行数据处理 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.encode(token, add_special_tokens=False) for sub_token in sub_tokens.tokens: tok_to_orig_index.append(i) all_doc_tokens.append(sub_token) max_tokens_for_doc = max_seq_length - len(query_tokens) - 3 # 当上下文文本的长度大于规定的最大长度，则使用滑动窗口的方法。 _DocSpan = collections.namedtuple("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): # 如果start_offset + length == len(all_doc_tokens)，则不进行滑动窗口操作 break start_offset += min(length, doc_stride) # 拼接原始问题和上下文文本 for (doc_span_index, doc_span) in enumerate(doc_spans): tokens = [] token_to_orig_map = {} token_is_max_context = {} segment_ids = [] tokens.append("[CLS]") segment_ids.append(0) for token in query_tokens.tokens: tokens.append(token) segment_ids.append(0) tokens.append("[SEP]") segment_ids.append(0) 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(1) tokens.append("[SEP]") segment_ids.append(1) input_ids = [] for token in tokens: input_ids.append(tokenizer.token_to_id(token)) # 掩码为1表示真实标记，0表示填充标记。 input_mask = [1] * len(input_ids) # 当序列长度小于max_seq_length时，零填充序列到max_seq_length长度 while len(input_ids) < max_seq_length: input_ids.append(0) input_mask.append(0) segment_ids.append(0) res_input_ids.append(np.array(input_ids, dtype=np.int64)) res_input_mask.append(np.array(input_mask, dtype=np.int64)) res_segment_ids.append(np.array(segment_ids, dtype=np.int64)) feature = Feature(unique_id=unique_id, tokens=tokens, example_index=example_index, token_to_orig_map=token_to_orig_map, token_is_max_context=token_is_max_context) extra.append(feature) unique_id += 1 return np.array(res_input_ids), np.array(res_input_mask), np.array( res_segment_ids), extra # 将SQuAD json文件读入到一个SquadEexample列表中 def read_squad_examples(input_file): with open(input_file, "r") as f: input_data = json.load(f)["data"] def is_whitespace(c): if c == " " or c == "\t" or c == "\r" or c == "\n" or ord(c) == 0x202F: return True return False examples = [] for idx, entry in enumerate(input_data): # 获取上下文文本内容，并存储在doc_tokens列表中 for paragraph in entry["paragraphs"]: paragraph_text = paragraph["context"] 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) # 获取原始问题文本和对应的id for qa in paragraph["qas"]: qas_id = qa["id"] question_text = qa["question"] start_position = None end_position = None orig_answer_text = None # 将上下文文本和原始问题文本等保存在SquadExample列表中 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) examples.append(example) return examples def write_predictions(all_examples, all_features, all_results, n_best_size, max_answer_length, do_lower_case, output_prediction_file, output_nbest_file): example_index_to_features = collections.defaultdict(list) for feature in all_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( # pylint: disable=invalid-name "PrelimPrediction", [ "feature_index", "start_index", "end_index", "start_logit", "end_logit" ]) all_predictions = collections.OrderedDict() all_nbest_json = collections.OrderedDict() for (example_index, example) in enumerate(all_examples): features = example_index_to_features[example_index] prelim_predictions = [] for (feature_index, feature) in enumerate(features): # 取前n_best_size个预测概率值 if not feature.unique_id in unique_id_to_result: print("feature not in unique_Id", feature.unique_id) continue 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) # 筛选与过滤，过滤掉不符合的开始索引和结束索引 for start_index in start_indexes: for end_index in end_indexes: 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])) # 排序，开始索引加结束索引的概率值和最大的排在前面 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"]) seen_predictions = {} nbest = [] for pred in prelim_predictions: # 取前n_best_size个概率值最大的结果 if len(nbest) >= n_best_size: break feature = features[pred.feature_index] 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) # 去标记化已分离的单词以及去除首尾空格 tok_text = tok_text.replace(" ##", "") tok_text = tok_text.replace("##", "") tok_text = tok_text.strip() tok_text = " ".join(tok_text.split()) orig_text = " ".join(orig_tokens) nbest.append( _NbestPrediction(text=orig_text, start_logit=pred.start_logit, end_logit=pred.end_logit)) if not nbest: nbest.append( _NbestPrediction(text="empty", start_logit=0.0, end_logit=0.0)) assert len(nbest) >= 1 total_scores = [] for entry in nbest: total_scores.append(entry.start_logit + entry.end_logit) probs = compute_softmax(total_scores) nbest_json = [] for (i, entry) in enumerate(nbest): output = collections.OrderedDict() output["text"] = entry.text # 获取对应的文本 output["probability"] = probs[i] # 预测概率值 output["start_logit"] = float(entry.start_logit) # 开始位置的概率值 output["end_logit"] = float(entry.end_logit) # 结束位置的概率值 nbest_json.append(output) all_predictions[example.qas_id] = nbest_json[0]["text"] # 取最大概率值，作为最终的预测答案 all_nbest_json[example.qas_id] = nbest_json with open(output_prediction_file, "w") as writer: writer.write(json.dumps(all_predictions, indent=4) + "\n") with open(output_nbest_file, "w") as writer: writer.write(json.dumps(all_nbest_json, indent=4) + "\n") # 对logits的概率值排序，并获取前个n_best_size的概率值 def get_best_indexes(logits, n_best_size): 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 # 计算softmax def compute_softmax(scores): 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