import random import numpy as np import torch import os import csv import sys import logging from transformers import BertTokenizer, AutoTokenizer from torch.utils.data import (DataLoader, RandomSampler, SequentialSampler, TensorDataset) from sentence_transformers import SentenceTransformer class BERT_Loader: def __init__(self, args, base_attrs, logger_name = 'Discovery'): self.logger = logging.getLogger(logger_name) if args.method == 'SCCL' : self.tokenizer = SentenceTransformer('distilbert-base-nli-stsb-mean-tokens')[0].tokenizer else: self.tokenizer = BertTokenizer.from_pretrained(args.pretrained_bert_model, do_lower_case=True) if args.setting == 'unsupervised': self.train_examples = get_examples(args, base_attrs, 'train') self.eval_examples = get_examples(args, base_attrs, 'eval') self.train_examples = self.train_examples + self.eval_examples self.train_outputs = get_loader(self.train_examples, args, base_attrs['all_label_list'], 'train_unlabeled', self.tokenizer) self.logger.info("Number of train samples = %s", str(len(self.train_examples))) self.test_examples = get_examples(args, base_attrs, 'test') self.logger.info("Number of testing samples = %s", str(len(self.test_examples))) self.test_outputs = get_loader(self.test_examples, args, base_attrs['all_label_list'], 'test', self.tokenizer) elif args.setting == 'semi_supervised': self.train_examples, self.train_labeled_examples, self.train_unlabeled_examples = get_examples(args, base_attrs, 'train') self.logger.info("Number of labeled training samples = %s", str(len(self.train_labeled_examples))) self.logger.info("Number of unlabeled training samples = %s", str(len(self.train_unlabeled_examples))) self.eval_examples = get_examples(args, base_attrs, 'eval') self.logger.info("Number of evaluation samples = %s", str(len(self.eval_examples))) self.test_examples = get_examples(args, base_attrs, 'test') self.logger.info("Number of testing samples = %s", str(len(self.test_examples))) self.train_labeled_outputs = get_loader(self.train_labeled_examples, args, base_attrs['known_label_list'], 'train_labeled', self.tokenizer) self.train_unlabeled_outputs = get_loader(self.train_unlabeled_examples, args, base_attrs['all_label_list'], 'train_unlabeled', self.tokenizer) self.train_outputs = get_semi_loader(self.train_labeled_examples, self.train_unlabeled_examples, base_attrs, args, self.tokenizer) self.eval_outputs = get_loader(self.eval_examples, args, base_attrs['known_label_list'], 'eval', self.tokenizer) self.test_outputs = get_loader(self.test_examples, args, base_attrs['all_label_list'], 'test', self.tokenizer) if args.method == 'DTC_BERT': self.get_examples_dtc_predict(args ,base_attrs) self.num_train_examples = len(self.train_examples) def get_examples_dtc_predict(self, args ,base_attrs): num_val_cls = round(base_attrs['n_known_cls'] * 0.75 ) self.num_val_cls = num_val_cls label_val = list(np.random.choice(np.array(base_attrs['known_label_list']), num_val_cls, replace=False)) #44 label_train = [label for label in base_attrs['known_label_list'] if label not in label_val] ntrain = len(self.train_examples) train_labels = np.array([example.label for example in self.train_examples]) train_base_attrs = {} train_base_attrs['known_label_list'] = label_train train_base_attrs['data_dir'] = base_attrs['data_dir'] train_base_attrs['all_label_list'] = base_attrs['all_label_list'] self.train_examples_dtc, self.train_labeled_examples_dtc, self.train_unlabeled_examples_dtc = get_examples(args, train_base_attrs, 'train') self.logger.info("Number of labeled training samples = %s", str(len(self.train_labeled_examples_dtc))) self.logger.info("Number of unlabeled training samples = %s", str(len(self.train_unlabeled_examples_dtc))) self.eval_examples_dtc = get_examples(args, train_base_attrs, 'eval') self.logger.info("Number of evaluation samples = %s", str(len(self.eval_examples_dtc))) self.train_labeled_outputs_dtc = get_loader(self.train_labeled_examples_dtc, args, train_base_attrs['known_label_list'], 'train_labeled', self.tokenizer) self.train_unlabeled_outputs_dtc = get_loader(self.train_unlabeled_examples_dtc, args, train_base_attrs['all_label_list'], 'train_unlabeled', self.tokenizer) self.eval_outputs_dtc = get_loader(self.eval_examples_dtc, args, train_base_attrs['known_label_list'], 'eval', self.tokenizer) val_base_attrs = {} val_base_attrs['known_label_list'] = label_val val_base_attrs['data_dir'] = base_attrs['data_dir'] val_base_attrs['all_label_list'] = base_attrs['all_label_list'] self.val_examples_dtc, self.val_labeled_examples_dtc, self.val_unlabeled_examples_dtc = get_examples(args, val_base_attrs, 'train') self.val_labeled_outputs_dtc = get_loader(self.val_labeled_examples_dtc, args, val_base_attrs['known_label_list'], 'train_labeled', self.tokenizer) def get_examples(args, base_attrs, mode): processor = DatasetProcessor() ori_examples = processor.get_examples(base_attrs['data_dir'], mode) if args.setting == 'unsupervised': return ori_examples elif args.setting == 'semi_supervised': if mode == 'train': train_labels = np.array([example.label for example in ori_examples]) train_labeled_ids = [] for label in base_attrs['known_label_list']: num = round(len(train_labels[train_labels == label]) * args.labeled_ratio) pos = list(np.where(train_labels == label)[0]) train_labeled_ids.extend(random.sample(pos, num)) labeled_examples, unlabeled_examples = [], [] for idx, example in enumerate(ori_examples): if idx in train_labeled_ids: labeled_examples.append(example) else: unlabeled_examples.append(example) return ori_examples, labeled_examples, unlabeled_examples elif mode == 'eval': examples = [] for example in ori_examples: if (example.label in base_attrs['known_label_list']): examples.append(example) return examples elif mode == 'test': return ori_examples def get_loader(examples, args, label_list, mode, tokenizer): features = convert_examples_to_features(examples, label_list, args.max_seq_length, tokenizer) input_ids = torch.tensor([f.input_ids for f in features], dtype=torch.long) input_mask = torch.tensor([f.input_mask for f in features], dtype=torch.long) segment_ids = torch.tensor([f.segment_ids for f in features], dtype=torch.long) if mode == 'train_unlabeled': label_ids = torch.tensor([-1 for f in features], dtype=torch.long) else: label_ids = torch.tensor([f.label_id for f in features], dtype=torch.long) datatensor = TensorDataset(input_ids, input_mask, segment_ids, label_ids) if mode == 'train_labeled': sampler = RandomSampler(datatensor) dataloader = DataLoader(datatensor, sampler=sampler, batch_size = args.train_batch_size, num_workers = args.num_workers, pin_memory = True) #, num_workers = args.num_workers, pin_memory = True else: sampler = SequentialSampler(datatensor) if mode == 'train_unlabeled': dataloader = DataLoader(datatensor, sampler=sampler, batch_size = args.train_batch_size, num_workers = args.num_workers, pin_memory = True) elif mode == 'eval': dataloader = DataLoader(datatensor, sampler=sampler, batch_size = args.eval_batch_size, num_workers = args.num_workers, pin_memory = True) elif mode == 'test': dataloader = DataLoader(datatensor, sampler=sampler, batch_size = args.test_batch_size, num_workers = args.num_workers, pin_memory = True) outputs = { 'loader': dataloader, 'input_ids': input_ids, 'input_mask': input_mask, 'segment_ids': segment_ids, 'label_ids': label_ids, 'data': datatensor } return outputs def get_semi_loader(labeled_examples, unlabeled_examples, base_attrs, args, tokenizer): labeled_features = convert_examples_to_features(labeled_examples, base_attrs['known_label_list'], args.max_seq_length, tokenizer) unlabeled_features = convert_examples_to_features(unlabeled_examples, base_attrs['all_label_list'], args.max_seq_length, tokenizer) labeled_input_ids = torch.tensor([f.input_ids for f in labeled_features], dtype=torch.long) labeled_input_mask = torch.tensor([f.input_mask for f in labeled_features], dtype=torch.long) labeled_segment_ids = torch.tensor([f.segment_ids for f in labeled_features], dtype=torch.long) labeled_label_ids = torch.tensor([f.label_id for f in labeled_features], dtype=torch.long) unlabeled_input_ids = torch.tensor([f.input_ids for f in unlabeled_features], dtype=torch.long) unlabeled_input_mask = torch.tensor([f.input_mask for f in unlabeled_features], dtype=torch.long) unlabeled_segment_ids = torch.tensor([f.segment_ids for f in unlabeled_features], dtype=torch.long) unlabeled_label_ids = torch.tensor([-1 for f in unlabeled_features], dtype=torch.long) semi_input_ids = torch.cat([labeled_input_ids, unlabeled_input_ids]) semi_input_mask = torch.cat([labeled_input_mask, unlabeled_input_mask]) semi_segment_ids = torch.cat([labeled_segment_ids, unlabeled_segment_ids]) semi_label_ids = torch.cat([labeled_label_ids, unlabeled_label_ids]) semi_data = TensorDataset(semi_input_ids, semi_input_mask, semi_segment_ids, semi_label_ids) semi_sampler = SequentialSampler(semi_data) semi_dataloader = DataLoader(semi_data, sampler=semi_sampler, batch_size = args.train_batch_size, num_workers = args.num_workers, pin_memory = True)#args.train_batch_size) outputs = { 'loader': semi_dataloader, 'input_ids': semi_input_ids, 'input_mask': semi_input_mask, 'segment_ids': semi_segment_ids, 'label_ids': semi_label_ids, 'semi_data' : semi_data } return outputs class InputExample(object): """A single training/test example for simple sequence classification.""" def __init__(self, guid, text_a, text_b=None, label=None): """Constructs a InputExample. Args: guid: Unique id for the example. text_a: string. The untokenized text of the first sequence. For single sequence tasks, only this sequence must be specified. text_b: (Optional) string. The untokenized text of the second sequence. Only must be specified for sequence pair tasks. label: (Optional) string. The label of the example. This should be specified for train and dev examples, but not for test examples. """ self.guid = guid self.text_a = text_a self.text_b = text_b self.label = label class InputFeatures(object): """A single set of features of data.""" def __init__(self, input_ids, input_mask, segment_ids, label_id): self.input_ids = input_ids self.input_mask = input_mask self.segment_ids = segment_ids self.label_id = label_id class DataProcessor(object): """Base class for data converters for sequence classification data sets.""" def get_train_examples(self, data_dir): """Gets a collection of `InputExample`s for the train set.""" raise NotImplementedError() def get_dev_examples(self, data_dir): """Gets a collection of `InputExample`s for the dev set.""" raise NotImplementedError() def get_labels(self): """Gets the list of labels for this data set.""" raise NotImplementedError() @classmethod def _read_tsv(cls, input_file, quotechar=None): """Reads a tab separated value file.""" with open(input_file, "r") as f: reader = csv.reader(f, delimiter="\t", quotechar=quotechar) lines = [] for line in reader: if sys.version_info[0] == 2: line = list(unicode(cell, 'utf-8') for cell in line) lines.append(line) return lines class DatasetProcessor(DataProcessor): def get_examples(self, data_dir, mode): if mode == 'train': return self._create_examples( self._read_tsv(os.path.join(data_dir, "train.tsv")), "train") elif mode == 'eval': return self._create_examples( self._read_tsv(os.path.join(data_dir, "dev.tsv")), "train") elif mode == 'test': return self._create_examples( self._read_tsv(os.path.join(data_dir, "test.tsv")), "test") def _create_examples(self, lines, set_type): """Creates examples for the training and dev sets.""" examples = [] for (i, line) in enumerate(lines): if i == 0: continue if len(line) != 2: continue guid = "%s-%s" % (set_type, i) text_a = line[0] label = line[1] examples.append( InputExample(guid=guid, text_a=text_a, text_b=None, label=label)) return examples def convert_examples_to_features(examples, label_list, max_seq_length, tokenizer): """Loads a data file into a list of `InputBatch`s.""" label_map = {} for i, label in enumerate(label_list): label_map[label] = i features = [] for (ex_index, example) in enumerate(examples): tokens_a = tokenizer.tokenize(example.text_a) tokens_b = None if example.text_b: tokens_b = tokenizer.tokenize(example.text_b) # Modifies `tokens_a` and `tokens_b` in place so that the total # length is less than the specified length. # Account for [CLS], [SEP], [SEP] with "- 3" _truncate_seq_pair(tokens_a, tokens_b, max_seq_length - 3) else: # Account for [CLS] and [SEP] with "- 2" if len(tokens_a) > max_seq_length - 2: tokens_a = tokens_a[:(max_seq_length - 2)] # The convention in BERT is: # (a) For sequence pairs: # tokens: [CLS] is this jack ##son ##ville ? [SEP] no it is not . [SEP] # type_ids: 0 0 0 0 0 0 0 0 1 1 1 1 1 1 # (b) For single sequences: # tokens: [CLS] the dog is hairy . [SEP] # type_ids: 0 0 0 0 0 0 0 # # Where "type_ids" are used to indicate whether this is the first # sequence or the second sequence. The embedding vectors for `type=0` and # `type=1` were learned during pre-training and are added to the wordpiece # embedding vector (and position vector). This is not *strictly* necessary # since the [SEP] token unambigiously separates the sequences, but it makes # it easier for the model to learn the concept of sequences. # # For classification tasks, the first vector (corresponding to [CLS]) is # used as as the "sentence vector". Note that this only makes sense because # the entire model is fine-tuned. tokens = ["[CLS]"] + tokens_a + ["[SEP]"] segment_ids = [0] * len(tokens) if tokens_b: tokens += tokens_b + ["[SEP]"] segment_ids += [1] * (len(tokens_b) + 1) 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] * len(input_ids) # Zero-pad up to the sequence length. padding = [0] * (max_seq_length - len(input_ids)) input_ids += padding input_mask += padding segment_ids += padding assert len(input_ids) == max_seq_length assert len(input_mask) == max_seq_length assert len(segment_ids) == max_seq_length label_id = label_map[example.label] features.append( InputFeatures(input_ids=input_ids, input_mask=input_mask, segment_ids=segment_ids, label_id=label_id)) return features def _truncate_seq_pair(tokens_a, tokens_b, max_length): """Truncates a sequence pair in place to the maximum length.""" # This is a simple heuristic which will always truncate the longer sequence # one token at a time. This makes more sense than truncating an equal percent # of tokens from each, since if one sequence is very short then each token # that's truncated likely contains more information than a longer sequence. while True: total_length = len(tokens_a) + len(tokens_b) if total_length <= max_length: break if len(tokens_a) > len(tokens_b): tokens_a.pop(0) # For dialogue context else: tokens_b.pop()