run_re_unidirect.py

# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors and The HuggingFace Inc. team.
# Copyright (c) 2018, NVIDIA CORPORATION.  All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
""" Finetuning the library models for sequence classification on GLUE (Bert, XLM, XLNet, RoBERTa)."""

from __future__ import absolute_import, division, print_function

import argparse
import glob
import logging
import os
import random
from collections import defaultdict
import re
import shutil

import numpy as np
import torch
from torch.utils.data import (DataLoader, RandomSampler, SequentialSampler,
                              TensorDataset)
from torch.utils.data.distributed import DistributedSampler
from tensorboardX import SummaryWriter
from tqdm import tqdm, trange
import time
from transformers import (WEIGHTS_NAME, BertConfig,
                                  BertTokenizer,
                                  RobertaConfig,
                                  RobertaTokenizer,
                                  get_linear_schedule_with_warmup,
                                  AdamW,
                                  BertForACEBothOneDropoutSub,
                                  AlbertForACEBothSub,
                                  AlbertConfig,
                                  AlbertTokenizer,
                                  AlbertForACEBothOneDropoutSub,
                                  BertForACEBothOneDropoutSubNoNer,
                                  )

from transformers import AutoTokenizer
from torch.utils.data import TensorDataset, Dataset
import json
import pickle
import numpy as np
import unicodedata
import itertools

from tqdm import tqdm

logger = logging.getLogger(__name__)


ALL_MODELS = sum((tuple(conf.pretrained_config_archive_map.keys()) for conf in (BertConfig,  AlbertConfig)), ())

MODEL_CLASSES = {
    'bertsub': (BertConfig, BertForACEBothOneDropoutSub, BertTokenizer),
}

task_ner_labels = {
    'ace04': ['FAC', 'WEA', 'LOC', 'VEH', 'GPE', 'ORG', 'PER'],
    'ace05': ['FAC', 'WEA', 'LOC', 'VEH', 'GPE', 'ORG', 'PER'],
    'scierc': ['Method', 'OtherScientificTerm', 'Task', 'Generic', 'Material', 'Metric'],
}

task_rel_labels = {
    'ace04': ['PER-SOC', 'OTHER-AFF', 'ART', 'GPE-AFF', 'EMP-ORG', 'PHYS'],
    'ace05': ['PER-SOC', 'ART', 'ORG-AFF', 'GEN-AFF', 'PHYS', 'PART-WHOLE'], 
    'scierc': ['PART-OF', 'USED-FOR', 'FEATURE-OF', 'CONJUNCTION', 'EVALUATE-FOR', 'HYPONYM-OF', 'COMPARE'],
}



class ACEDataset(Dataset):
    def __init__(self, tokenizer, args=None, evaluate=False, do_test=False, max_pair_length=None):

        if not evaluate:
            file_path = os.path.join(args.data_dir, args.train_file)
        else:
            if do_test:
                if args.test_file.find('models')==-1:
                    file_path = os.path.join(args.data_dir, args.test_file)
                else:
                    file_path = args.test_file
            else:
                if args.dev_file.find('models')==-1:
                    file_path = os.path.join(args.data_dir, args.dev_file)
                else:
                    file_path = args.dev_file

        assert os.path.isfile(file_path)

        self.file_path = file_path
                
        self.tokenizer = tokenizer
        self.max_seq_length = args.max_seq_length
        self.max_pair_length = max_pair_length
        self.max_entity_length = self.max_pair_length*2

        self.max_mention_length = args.max_mention_length
        self.evaluate = evaluate

        self.local_rank = args.local_rank
        self.args = args
        self.model_type = args.model_type
        self.no_sym = args.no_sym

        if args.data_dir.find('ace05')!=-1:
            self.ner_label_list = ['NIL', 'FAC', 'WEA', 'LOC', 'VEH', 'GPE', 'ORG', 'PER']

            if args.no_sym:
                label_list = ['PER-SOC', 'ART', 'ORG-AFF', 'GEN-AFF', 'PHYS', 'PART-WHOLE']
                self.sym_labels = ['NIL']
                self.label_list = self.sym_labels + label_list
            else:
                label_list = ['ART', 'ORG-AFF', 'GEN-AFF', 'PHYS',  'PART-WHOLE']
                self.sym_labels = ['NIL', 'PER-SOC']
                self.label_list = self.sym_labels + label_list

        elif args.data_dir.find('ace04')!=-1:
            self.ner_label_list = ['NIL', 'FAC', 'WEA', 'LOC', 'VEH', 'GPE', 'ORG', 'PER']

            if args.no_sym:
                label_list = ['PER-SOC', 'OTHER-AFF', 'ART', 'GPE-AFF', 'EMP-ORG', 'PHYS']
                self.sym_labels = ['NIL']
                self.label_list = self.sym_labels + label_list
            else:
                label_list = ['OTHER-AFF', 'ART', 'GPE-AFF', 'EMP-ORG', 'PHYS']
                self.sym_labels = ['NIL', 'PER-SOC']
                self.label_list = self.sym_labels + label_list

        elif args.data_dir.find('scierc')!=-1:      
            self.ner_label_list = ['NIL', 'Method', 'OtherScientificTerm', 'Task', 'Generic', 'Material', 'Metric']

            if args.no_sym:
                label_list = ['CONJUNCTION', 'COMPARE', 'PART-OF', 'USED-FOR', 'FEATURE-OF',  'EVALUATE-FOR', 'HYPONYM-OF']
                self.sym_labels = ['NIL']
                self.label_list = self.sym_labels + label_list
            else:
                label_list = ['PART-OF', 'USED-FOR', 'FEATURE-OF',  'EVALUATE-FOR', 'HYPONYM-OF']
                self.sym_labels = ['NIL', 'CONJUNCTION', 'COMPARE']
                self.label_list = self.sym_labels + label_list

        else:
            assert (False)  

        # self.eval_pred_ent = args.eval_pred_ent
        self.global_predicted_ners = {}
        self.initialize()
 
    def initialize(self, eval_cands=None):
        tokenizer = self.tokenizer
        vocab_size = tokenizer.vocab_size
        max_num_subwords = self.max_seq_length - 4  # for two marker
        label_map = {label: i for i, label in enumerate(self.label_list)}
        ner_label_map = {label: i for i, label in enumerate(self.ner_label_list)}

        def tokenize_word(text):
            if (
                isinstance(tokenizer, RobertaTokenizer)
                and (text[0] != "'")
                and (len(text) != 1 or not self.is_punctuation(text))
            ):
                return tokenizer.tokenize(text, add_prefix_space=True)
            return tokenizer.tokenize(text)

        f = open(self.file_path, "r", encoding='utf-8')
        self.ner_tot_recall = 0
        self.tot_recall = 0
        self.data = []
        self.ner_golden_labels = set([])
        self.golden_labels = set([])
        self.golden_labels_withner = set([])
        maxR = 0
        maxL = 0
        for l_idx, line in enumerate(f):
            data = json.loads(line)

            if self.args.output_dir.find('test')!=-1:
                if len(self.data) > 100:
                    break

            sentences = data['sentences']
            if 'predicted_ner' in data: 
               ners = data['predicted_ner']               
            else:
               ners = data['ner']

            std_ners = data['ner']

         
            relations = data['relations']

            for sentence_relation in relations:
                for x in sentence_relation:
                    if x[4] in self.sym_labels[1:]:
                        self.tot_recall += 2
                    else: 
                        self.tot_recall +=  1

            sentence_boundaries = [0]
            words = []
            L = 0
            for i in range(len(sentences)):
                L += len(sentences[i])
                sentence_boundaries.append(L)
                words += sentences[i]


            tokens = [tokenize_word(w) for w in words]
            subwords = [w for li in tokens for w in li]
            maxL = max(maxL, len(subwords))
            subword2token = list(itertools.chain(*[[i] * len(li) for i, li in enumerate(tokens)]))
            token2subword = [0] + list(itertools.accumulate(len(li) for li in tokens))
            subword_start_positions = frozenset(token2subword)
            subword_sentence_boundaries = [sum(len(li) for li in tokens[:p]) for p in sentence_boundaries]

            for n in range(len(subword_sentence_boundaries) - 1):

                sentence_ners = ners[n]
                sentence_relations = relations[n]
                std_ner = std_ners[n]

                # entity_labels = {}
                # for start, end, label in sentence_ners:
                #     entity_labels[(token2subword[start], token2subword[end+1])] = ner_label_map[label]    
                std_entity_labels = {}
                self.ner_tot_recall += len(std_ner)

                for start, end, label in std_ner:
                    std_entity_labels[(start, end)] = label
                    self.ner_golden_labels.add( ((l_idx, n), (start, end), label) )

                self.global_predicted_ners[(l_idx, n)] = list(sentence_ners)


                doc_sent_start, doc_sent_end = subword_sentence_boundaries[n : n + 2]

                left_length = doc_sent_start
                right_length = len(subwords) - doc_sent_end
                sentence_length = doc_sent_end - doc_sent_start
                half_context_length = int((max_num_subwords - sentence_length) / 2)

                if sentence_length < max_num_subwords:

                    if left_length < right_length:
                        left_context_length = min(left_length, half_context_length)
                        right_context_length = min(right_length, max_num_subwords - left_context_length - sentence_length)
                    else:
                        right_context_length = min(right_length, half_context_length)
                        left_context_length = min(left_length, max_num_subwords - right_context_length - sentence_length)


                doc_offset = doc_sent_start - left_context_length
                target_tokens = subwords[doc_offset : doc_sent_end + right_context_length]
                target_tokens = [tokenizer.cls_token] + target_tokens[ : self.max_seq_length - 4] + [tokenizer.sep_token] 
                assert(len(target_tokens) <= self.max_seq_length - 2)
                
                pos2label = {}
                for x in sentence_relations:
                    pos2label[(x[0],x[1],x[2],x[3])] = label_map[x[4]]
                    self.golden_labels.add(((l_idx, n), (x[0],x[1]), (x[2],x[3]), x[4]))
                    self.golden_labels_withner.add(((l_idx, n), (x[0],x[1], std_entity_labels[(x[0], x[1])]), (x[2],x[3], std_entity_labels[(x[2], x[3])]), x[4]))
                    if x[4] in self.sym_labels[1:]:
                        self.golden_labels.add(((l_idx, n),  (x[2],x[3]), (x[0],x[1]), x[4]))
                        self.golden_labels_withner.add(((l_idx, n), (x[2],x[3], std_entity_labels[(x[2], x[3])]), (x[0],x[1], std_entity_labels[(x[0], x[1])]), x[4]))


                if eval_cands:
                    if (l_idx, n) in eval_cands:
                        entities = eval_cands[(l_idx, n)]
                    else:
                        continue
                else:
                    entities = list(sentence_ners)


                for x in sentence_relations:
                    w = (x[2],x[3],x[0],x[1])
                    if w not in pos2label:
                        if x[4] in self.sym_labels[1:]:
                            pos2label[w] = label_map[x[4]]  

                if not self.evaluate:
                    entities.append((10000, 10000, 'NIL')) # only for NER

                for sub in entities:    
                    cur_ins = []

                    if sub[0] < 10000:
                        sub_s = token2subword[sub[0]] - doc_offset + 1
                        sub_e = token2subword[sub[1]+1] - doc_offset
                        l_m = '[unused0]'
                        r_m = '[unused1]'
                        
                        sub_tokens = target_tokens[:sub_s] + [l_m] + target_tokens[sub_s:sub_e+1] + [r_m] + target_tokens[sub_e+1: ]
                        sub_e += 2
                        sub_label = ner_label_map[sub[2]]
                    else:
                        sub_s = len(target_tokens)
                        sub_e = len(target_tokens)+1
                        sub_tokens = target_tokens + ['[unused0]',  '[unused1]']
                        sub_label = -1

                    if sub_e >= self.max_seq_length-1:
                        continue

                    for start, end, obj_label in sentence_ners:
                        if self.model_type.endswith('nersub'):
                            if start==sub[0] and end==sub[1]:
                                continue


                        doc_entity_start = token2subword[start]
                        doc_entity_end = token2subword[end+1]
                        left = doc_entity_start - doc_offset + 1
                        right = doc_entity_end - doc_offset

                        obj = (start, end)
                        if obj[0] >= sub[0]:
                            left += 1
                            if obj[0] > sub[1]:
                                left += 1

                        if obj[1] >= sub[0]:   
                            right += 1
                            if obj[1] > sub[1]:
                                right += 1
    
                        label = pos2label.get((sub[0], sub[1], obj[0], obj[1]), 0)

                        if right >= self.max_seq_length-1:
                            continue

                        cur_ins.append(((left, right, ner_label_map[obj_label]), label, obj))


                    maxR = max(maxR, len(cur_ins))
                    dL = self.max_pair_length
                    if self.args.shuffle:
                        np.random.shuffle(cur_ins)
                    for i in range(0, len(cur_ins), dL):
                        examples = cur_ins[i : i + dL]
                        item = {
                            'index': (l_idx, n),
                            'sentence': sub_tokens,
                            'examples': examples,
                            'sub': (sub, (sub_s, sub_e), sub_label), #(sub[0], sub[1], sub_label),
                        }                
                        
                        self.data.append(item)                    
        logger.info('maxR: %s', maxR)
        logger.info('maxL: %s', maxL)
                
    def __len__(self):
        return len(self.data)

    def __getitem__(self, idx):
        entry = self.data[idx]
        sub, sub_position, sub_label = entry['sub']
        input_ids = self.tokenizer.convert_tokens_to_ids(entry['sentence'])

        L = len(input_ids)
        input_ids += [self.tokenizer.pad_token_id] * (self.max_seq_length - len(input_ids))


        attention_mask = torch.zeros((self.max_entity_length+self.max_seq_length, self.max_entity_length+self.max_seq_length), dtype=torch.int64)
        attention_mask[:L, :L] = 1
        
        if self.model_type.startswith('albert'):
            input_ids = input_ids + [30002] * (len(entry['examples'])) + [self.tokenizer.pad_token_id] * (self.max_pair_length - len(entry['examples']))
            input_ids = input_ids + [30003] * (len(entry['examples'])) + [self.tokenizer.pad_token_id] * (self.max_pair_length - len(entry['examples'])) # for debug
        else:
            input_ids = input_ids + [3] * (len(entry['examples'])) + [self.tokenizer.pad_token_id] * (self.max_pair_length - len(entry['examples']))
            input_ids = input_ids + [4] * (len(entry['examples'])) + [self.tokenizer.pad_token_id] * (self.max_pair_length - len(entry['examples'])) # for debug

        labels = []
        ner_labels = []
        mention_pos = []
        mention_2 = []
        position_ids = list(range(self.max_seq_length)) + [0] * self.max_entity_length 
        num_pair = self.max_pair_length

        for x_idx, obj in enumerate(entry['examples']):
            m2 = obj[0]
            label = obj[1]

            mention_pos.append((m2[0], m2[1]))
            mention_2.append(obj[2])

            w1 = x_idx  
            w2 = w1 + num_pair

            w1 += self.max_seq_length
            w2 += self.max_seq_length
            
            position_ids[w1] = m2[0]
            position_ids[w2] = m2[1]

            for xx in [w1, w2]:
                for yy in [w1, w2]:
                    attention_mask[xx, yy] = 1
                attention_mask[xx, :L] = 1

            labels.append(label)
            ner_labels.append(m2[2])

        pair_L = len(entry['examples'])
        if self.args.att_left:
            attention_mask[self.max_seq_length : self.max_seq_length+pair_L, self.max_seq_length : self.max_seq_length+pair_L] = 1
        if self.args.att_right:
            attention_mask[self.max_seq_length+num_pair : self.max_seq_length+num_pair+pair_L, self.max_seq_length+num_pair : self.max_seq_length+num_pair+pair_L] = 1


        mention_pos += [(0, 0)] * (num_pair - len(mention_pos))
        labels += [-1] * (num_pair - len(labels))
        ner_labels += [-1] * (num_pair - len(ner_labels))

        item = [torch.tensor(input_ids),
                attention_mask,
                torch.tensor(position_ids),
                torch.tensor(sub_position),
                torch.tensor(mention_pos),
                torch.tensor(labels, dtype=torch.int64),
                torch.tensor(ner_labels, dtype=torch.int64),
                torch.tensor(sub_label, dtype=torch.int64)
        ]

        if self.evaluate:
            item.append(entry['index'])
            item.append(sub)
            item.append(mention_2)

        return item

    @staticmethod
    def collate_fn(batch):
        fields = [x for x in zip(*batch)]

        num_metadata_fields = 3
        stacked_fields = [torch.stack(field) for field in fields[:-num_metadata_fields]]  # don't stack metadata fields
        stacked_fields.extend(fields[-num_metadata_fields:])  # add them as lists not torch tensors

        return stacked_fields

 


def set_seed(args):
    random.seed(args.seed)
    np.random.seed(args.seed)
    torch.manual_seed(args.seed)
    if args.n_gpu > 0:
        torch.cuda.manual_seed_all(args.seed)



def _rotate_checkpoints(args, checkpoint_prefix, use_mtime=False):
    if not args.save_total_limit:
        return
    if args.save_total_limit <= 0:
        return

    # Check if we should delete older checkpoint(s)
    glob_checkpoints = glob.glob(os.path.join(args.output_dir, '{}-*'.format(checkpoint_prefix)))
    if len(glob_checkpoints) <= args.save_total_limit:
        return

    ordering_and_checkpoint_path = []
    for path in glob_checkpoints:
        if use_mtime:
            ordering_and_checkpoint_path.append((os.path.getmtime(path), path))
        else:
            regex_match = re.match('.*{}-([0-9]+)'.format(checkpoint_prefix), path)
            if regex_match and regex_match.groups():
                ordering_and_checkpoint_path.append((int(regex_match.groups()[0]), path))

    checkpoints_sorted = sorted(ordering_and_checkpoint_path)
    checkpoints_sorted = [checkpoint[1] for checkpoint in checkpoints_sorted]
    number_of_checkpoints_to_delete = max(0, len(checkpoints_sorted) - args.save_total_limit)
    checkpoints_to_be_deleted = checkpoints_sorted[:number_of_checkpoints_to_delete]
    for checkpoint in checkpoints_to_be_deleted:
        logger.info("Deleting older checkpoint [{}] due to args.save_total_limit".format(checkpoint))
        shutil.rmtree(checkpoint)

def train(args, model, tokenizer):
    """ Train the model """
    if args.local_rank in [-1, 0]:
        tb_writer = SummaryWriter("logs/"+args.data_dir[max(args.data_dir.rfind('/'),0):]+"_re_logs/"+args.output_dir[args.output_dir.rfind('/'):])

    args.train_batch_size = args.per_gpu_train_batch_size * max(1, args.n_gpu)

    train_dataset = ACEDataset(tokenizer=tokenizer, args=args, max_pair_length=args.max_pair_length)

    train_sampler = RandomSampler(train_dataset) if args.local_rank == -1 else DistributedSampler(train_dataset)
    train_dataloader = DataLoader(train_dataset, sampler=train_sampler, batch_size=args.train_batch_size, num_workers=4*int(args.output_dir.find('test')==-1))

    if args.max_steps > 0:
        t_total = args.max_steps
        args.num_train_epochs = args.max_steps // (len(train_dataloader) // args.gradient_accumulation_steps) + 1
    else:
        t_total = len(train_dataloader) // args.gradient_accumulation_steps * args.num_train_epochs

    # Prepare optimizer and schedule (linear warmup and decay)
    no_decay = ['bias', 'LayerNorm.weight']
    optimizer_grouped_parameters = [
        {'params': [p for n, p in model.named_parameters() if not any(nd in n for nd in no_decay)], 'weight_decay': args.weight_decay},
        {'params': [p for n, p in model.named_parameters() if any(nd in n for nd in no_decay)], 'weight_decay': 0.0}
        ]
    
    optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
    if args.warmup_steps==-1:
        scheduler = get_linear_schedule_with_warmup(
            optimizer, num_warmup_steps=int(0.1*t_total), num_training_steps=t_total
        )
    else:
        scheduler = get_linear_schedule_with_warmup(
            optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total
        )

    if args.fp16:
        try:
            from apex import amp
        except ImportError:
            raise ImportError("Please install apex from https://www.github.com/nvidia/apex to use fp16 training.")
        model, optimizer = amp.initialize(model, optimizer, opt_level=args.fp16_opt_level)
    # ori_model = model
    # multi-gpu training (should be after apex fp16 initialization)
    if args.n_gpu > 1:
        model = torch.nn.DataParallel(model)

    # Distributed training (should be after apex fp16 initialization)
    if args.local_rank != -1:
        model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[args.local_rank],
                                                          output_device=args.local_rank,
                                                          find_unused_parameters=True)

    # Train!
    logger.info("***** Running training *****")
    logger.info("  Num examples = %d", len(train_dataset))
    logger.info("  Num Epochs = %d", args.num_train_epochs)
    logger.info("  Instantaneous batch size per GPU = %d", args.per_gpu_train_batch_size)
    logger.info("  Total train batch size (w. parallel, distributed & accumulation) = %d",
                   args.train_batch_size * args.gradient_accumulation_steps * (torch.distributed.get_world_size() if args.local_rank != -1 else 1))
    logger.info("  Gradient Accumulation steps = %d", args.gradient_accumulation_steps)
    logger.info("  Total optimization steps = %d", t_total)

    global_step = 0
    tr_loss, logging_loss = 0.0, 0.0
    tr_ner_loss, logging_ner_loss = 0.0, 0.0
    tr_re_loss, logging_re_loss = 0.0, 0.0

    model.zero_grad()
    train_iterator = trange(int(args.num_train_epochs), desc="Epoch", disable=args.local_rank not in [-1, 0])
    set_seed(args)  # Added here for reproductibility (even between python 2 and 3)
    best_f1 = -1



    for _ in train_iterator:
        if args.shuffle and _ > 0:
            train_dataset.initialize()
        epoch_iterator = tqdm(train_dataloader, desc="Iteration", disable=args.local_rank not in [-1, 0])
        for step, batch in enumerate(epoch_iterator):

            model.train()
            batch = tuple(t.to(args.device) for t in batch)

            inputs = {'input_ids':      batch[0],
                      'attention_mask': batch[1],
                      'position_ids':   batch[2],
                      'labels':         batch[5],
                      'ner_labels':     batch[6],
                      }

            # if args.model_type.endswith('sub'):
            inputs['sub_positions'] = batch[3]
            if args.model_type.find('span')!=-1:
                inputs['mention_pos'] = batch[4]
            if args.model_type.endswith('nersub'):
                inputs['sub_ner_labels'] = batch[7]

            outputs = model(**inputs)
            loss = outputs[0]  # model outputs are always tuple in pytorch-transformers (see doc)
            re_loss = outputs[1]
            ner_loss = outputs[2]

            if args.n_gpu > 1:
                loss = loss.mean() # mean() to average on multi-gpu parallel training
            if args.gradient_accumulation_steps > 1:
                loss = loss / args.gradient_accumulation_steps
                re_loss = re_loss / args.gradient_accumulation_steps
                ner_loss = ner_loss / args.gradient_accumulation_steps

            if args.fp16:
                with amp.scale_loss(loss, optimizer) as scaled_loss:
                    scaled_loss.backward()
            else:
                loss.backward()

            tr_loss += loss.item()
            if re_loss > 0:
                tr_re_loss += re_loss.item()
            if ner_loss > 0:
                tr_ner_loss += ner_loss.item()


            if (step + 1) % args.gradient_accumulation_steps == 0:
                if args.max_grad_norm > 0:
                    if args.fp16:
                        torch.nn.utils.clip_grad_norm_(amp.master_params(optimizer), args.max_grad_norm)
                    else:
                        torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_grad_norm)

                optimizer.step()
                scheduler.step()  # Update learning rate schedule
                model.zero_grad()
                global_step += 1

                # if args.model_type.endswith('rel') :
                #     ori_model.bert.encoder.layer[args.add_coref_layer].attention.self.relative_attention_bias.weight.data[0].zero_() # 可以手动乘个mask

                if args.local_rank in [-1, 0] and args.logging_steps > 0 and global_step % args.logging_steps == 0:
                    # Log metrics
                    tb_writer.add_scalar('lr', scheduler.get_lr()[0], global_step)
                    tb_writer.add_scalar('loss', (tr_loss - logging_loss)/args.logging_steps, global_step)
                    logging_loss = tr_loss

                    tb_writer.add_scalar('RE_loss', (tr_re_loss - logging_re_loss)/args.logging_steps, global_step)
                    logging_re_loss = tr_re_loss

                    tb_writer.add_scalar('NER_loss', (tr_ner_loss - logging_ner_loss)/args.logging_steps, global_step)
                    logging_ner_loss = tr_ner_loss


                if args.local_rank in [-1, 0] and args.save_steps > 0 and global_step % args.save_steps == 0: # valid for bert/spanbert
                    update = True
                    # Save model checkpoint
                    if args.evaluate_during_training:  # Only evaluate when single GPU otherwise metrics may not average well
                        results = evaluate(args, model, tokenizer)
                        f1 = results['f1_with_ner']
                        tb_writer.add_scalar('f1_with_ner', f1, global_step)

                        # f1 = results['f1']
                        # tb_writer.add_scalar('f1', f1, global_step)

                        if f1 > best_f1:
                            best_f1 = f1
                            print ('Best F1', best_f1)
                        else:
                            update = False

                    if update:
                        checkpoint_prefix = 'checkpoint'
                        output_dir = os.path.join(args.output_dir, '{}-{}'.format(checkpoint_prefix, global_step))
                        if not os.path.exists(output_dir):
                            os.makedirs(output_dir)
                        model_to_save = model.module if hasattr(model, 'module') else model  # Take care of distributed/parallel training

                        model_to_save.save_pretrained(output_dir)

                        torch.save(args, os.path.join(output_dir, 'training_args.bin'))
                        logger.info("Saving model checkpoint to %s", output_dir)

                        _rotate_checkpoints(args, checkpoint_prefix)

            if args.max_steps > 0 and global_step > args.max_steps:
                epoch_iterator.close()
                break
        if args.max_steps > 0 and global_step > args.max_steps:
            train_iterator.close()
            break

    if args.local_rank in [-1, 0]:
        tb_writer.close()


    return global_step, tr_loss / global_step, best_f1

def to_list(tensor):
    return tensor.detach().cpu().tolist()

def evaluate(args, model, tokenizer, prefix="", do_test=False):

    eval_output_dir = args.output_dir

    eval_dataset = ACEDataset(tokenizer=tokenizer, args=args, evaluate=True, do_test=do_test, max_pair_length=args.max_pair_length)
    golden_labels = set(eval_dataset.golden_labels)
    golden_labels_withner = set(eval_dataset.golden_labels_withner)
    label_list = list(eval_dataset.label_list)
    sym_labels = list(eval_dataset.sym_labels)
    tot_recall = eval_dataset.tot_recall

    if not os.path.exists(eval_output_dir) and args.local_rank in [-1, 0]:
        os.makedirs(eval_output_dir)

    args.eval_batch_size = args.per_gpu_eval_batch_size * max(1, args.n_gpu)


    scores = defaultdict(dict)
    # ner_pred = not args.model_type.endswith('noner')
    example_subs = set([])
    num_label = len(label_list)

    logger.info("***** Running evaluation {} *****".format(prefix))
    logger.info("  Batch size = %d", args.eval_batch_size)

    model.eval()

    eval_sampler = SequentialSampler(eval_dataset) 
    eval_dataloader = DataLoader(eval_dataset, sampler=eval_sampler, batch_size=args.eval_batch_size,  collate_fn=ACEDataset.collate_fn, num_workers=4*int(args.output_dir.find('test')==-1))

    # Eval!
    logger.info("  Num examples = %d", len(eval_dataset))

    for batch in tqdm(eval_dataloader, desc="Evaluating"):
        indexs = batch[-3]
        subs = batch[-2]
        batch_m2s = batch[-1]
        ner_labels = batch[6]

        batch = tuple(t.to(args.device) for t in batch[:-3])

        with torch.no_grad():
            inputs = {'input_ids':      batch[0],
                    'attention_mask': batch[1],
                    'position_ids':   batch[2],
                    #   'labels':         batch[4],
                    #   'ner_labels':     batch[5],
                    }
            # if args.model_type.endswith('sub'):
            inputs['sub_positions'] = batch[3]
            if args.model_type.find('span')!=-1:
                inputs['mention_pos'] = batch[4]

            outputs = model(**inputs)

            logits = outputs[0]
            if args.eval_logsoftmax:  # perform a bit better
                logits = torch.nn.functional.log_softmax(logits, dim=-1)

            elif args.eval_softmax:
                logits = torch.nn.functional.softmax(logits, dim=-1)

            if args.use_ner_results:                 
                ner_preds = ner_labels
            else:
                ner_preds = torch.argmax(outputs[1], dim=-1)
            logits = logits.cpu().numpy()
            ner_preds = ner_preds.cpu().numpy()
            for i in range(len(indexs)):
                index = indexs[i]
                sub = subs[i]
                m2s = batch_m2s[i]
                example_subs.add(((index[0], index[1]), (sub[0], sub[1])))
                for j in range(len(m2s)):
                    obj = m2s[j]
                    ner_label = eval_dataset.ner_label_list[ner_preds[i,j]]
                    scores[(index[0], index[1])][( (sub[0], sub[1]), (obj[0], obj[1]))] = (logits[i, j].tolist(), ner_label)

    cor = 0 
    tot_pred = 0
    cor_with_ner = 0
    global_predicted_ners = eval_dataset.global_predicted_ners
    ner_golden_labels = eval_dataset.ner_golden_labels
    ner_cor = 0 
    ner_tot_pred = 0
    ner_ori_cor = 0

    for example_index, pair_dict in sorted(scores.items(), key=lambda x:x[0]):  
        sentence_results = []
        for k1, (v1, v2_ner_label) in pair_dict.items():

            if v2_ner_label=='NIL':
                continue
            v1 = list(v1)
            m1 = k1[0]
            m2 = k1[1]
            if m1 == m2:
                continue

            pred_label = np.argmax(v1)
            if pred_label>0:
                pred_score = v1[pred_label]
                sentence_results.append( (pred_score, m1, m2, pred_label, None, v2_ner_label) )

        sentence_results.sort(key=lambda x: -x[0])
        no_overlap = []
        def is_overlap(m1, m2):
            if m2[0]<=m1[0] and m1[0]<=m2[1]:
                return True
            if m1[0]<=m2[0] and m2[0]<=m1[1]:
                return True
            return False

        output_preds = []

        for item in sentence_results:
            m1 = item[1]
            m2 = item[2]
            overlap = False
            for x in no_overlap:
                _m1 = x[1]
                _m2 = x[2]
                if item[3]==x[3] and (is_overlap(m1, _m1) and is_overlap(m2, _m2)):
                    overlap = True
                    break

            pred_label = label_list[item[3]]

            output_preds.append((m1, m2, pred_label))

            if not overlap:
                no_overlap.append(item)

        pos2ner = {}
        predpos2ner = {}
        ner_results = list(global_predicted_ners[example_index])
        for start, end, label in ner_results:
            predpos2ner[(start, end)] = label

        for item in no_overlap:
            m1 = item[1]
            m2 = item[2]
            pred_label = label_list[item[3]]
            tot_pred += 1

            if (example_index, m1, m2, pred_label) in golden_labels:
                cor += 1        

            if m1 not in pos2ner:
                pos2ner[m1] = predpos2ner[m1]#item[4]

            if m2 not in pos2ner:
                pos2ner[m2] = item[5]

            if (example_index, (m1[0], m1[1], pos2ner[m1]), (m2[0], m2[1], pos2ner[m2]), pred_label) in golden_labels_withner:
                cor_with_ner += 1      
        
        # refine NER results
        ner_results = list(global_predicted_ners[example_index])
        for i in range(len(ner_results)):
            start, end, label = ner_results[i] 
            if (example_index, (start, end), label) in ner_golden_labels:
                ner_ori_cor += 1
            if (start, end) in pos2ner:
                label = pos2ner[(start, end)]
            if (example_index, (start, end), label) in ner_golden_labels:
                ner_cor += 1
            ner_tot_pred += 1
        
    ner_ori_p = ner_ori_cor / ner_tot_pred if ner_tot_pred > 0 else 0 
    ner_p = ner_cor / ner_tot_pred if ner_tot_pred > 0 else 0 
    ner_r = ner_cor / len(ner_golden_labels) 
    ner_f1 = 2 * (ner_p * ner_r) / (ner_p + ner_r) if ner_cor > 0 else 0.0

    p = cor / tot_pred if tot_pred > 0 else 0 
    r = cor / tot_recall 
    f1 = 2 * (p * r) / (p + r) if cor > 0 else 0.0
    assert(tot_recall==len(golden_labels))

    p_with_ner = cor_with_ner / tot_pred if tot_pred > 0 else 0 
    r_with_ner = cor_with_ner / tot_recall
    assert(tot_recall==len(golden_labels_withner))
    f1_with_ner = 2 * (p_with_ner * r_with_ner) / (p_with_ner + r_with_ner) if cor_with_ner > 0 else 0.0

    results = {'f1':  f1,  'f1_with_ner': f1_with_ner, 'ner_f1': ner_f1}

    logger.info("Result: %s", json.dumps(results))

    return results



def main():
    parser = argparse.ArgumentParser()

    ## Required parameters
    parser.add_argument("--data_dir", default='ace_data', type=str, required=True,
                        help="The input data dir. Should contain the .tsv files (or other data files) for the task.")
    parser.add_argument("--model_type", default=None, type=str, required=True,
                        help="Model type selected in the list: " + ", ".join(MODEL_CLASSES.keys()))
    parser.add_argument("--model_name_or_path", default=None, type=str, required=True,
                        help="Path to pre-trained model or shortcut name selected in the list: " + ", ".join(ALL_MODELS))
    parser.add_argument("--output_dir", default=None, type=str, required=True,
                        help="The output directory where the model predictions and checkpoints will be written.")

    ## Other parameters
    parser.add_argument("--config_name", default="", type=str,
                        help="Pretrained config name or path if not the same as model_name")
    parser.add_argument("--tokenizer_name", default="", type=str,
                        help="Pretrained tokenizer name or path if not the same as model_name")
    parser.add_argument("--cache_dir", default="", type=str,
                        help="Where do you want to store the pre-trained models downloaded from s3")
    parser.add_argument("--max_seq_length", default=384, type=int,
                        help="The maximum total input sequence length after tokenization. Sequences longer "
                             "than this will be truncated, sequences shorter will be padded.")
    parser.add_argument("--do_train", action='store_true',
                        help="Whether to run training.")
    parser.add_argument("--do_eval", action='store_true',
                        help="Whether to run eval on the dev set.")
    parser.add_argument("--evaluate_during_training", action='store_true',
                        help="Rul evaluation during training at each logging step.")
    parser.add_argument("--do_lower_case", action='store_true',
                        help="Set this flag if you are using an uncased model.")

    parser.add_argument("--per_gpu_train_batch_size", default=8, type=int,
                        help="Batch size per GPU/CPU for training.")
    parser.add_argument("--per_gpu_eval_batch_size", default=8, type=int,
                        help="Batch size per GPU/CPU for evaluation.")
    parser.add_argument('--gradient_accumulation_steps', type=int, default=1,
                        help="Number of updates steps to accumulate before performing a backward/update pass.")
    parser.add_argument("--learning_rate", default=2e-5, type=float,
                        help="The initial learning rate for Adam.")
    parser.add_argument("--weight_decay", default=0.0, type=float,
                        help="Weight deay if we apply some.")
    parser.add_argument("--adam_epsilon", default=1e-8, type=float,
                        help="Epsilon for Adam optimizer.")
    parser.add_argument("--max_grad_norm", default=1.0, type=float,
                        help="Max gradient norm.")
    parser.add_argument("--num_train_epochs", default=10.0, type=float,
                        help="Total number of training epochs to perform.")
    parser.add_argument("--max_steps", default=-1, type=int,
                        help="If > 0: set total number of training steps to perform. Override num_train_epochs.")
    parser.add_argument("--warmup_steps", default=-1, type=int,
                        help="Linear warmup over warmup_steps.")
    parser.add_argument('--logging_steps', type=int, default=5,
                        help="Log every X updates steps.")
    parser.add_argument('--save_steps', type=int, default=1000,
                        help="Save checkpoint every X updates steps.")
    parser.add_argument("--eval_all_checkpoints", action='store_true',
                        help="Evaluate all checkpoints starting with the same prefix as model_name ending and ending with step number")
    parser.add_argument("--no_cuda", action='store_true',
                        help="Avoid using CUDA when available")
    parser.add_argument('--overwrite_output_dir', action='store_true',
                        help="Overwrite the content of the output directory")
    parser.add_argument('--overwrite_cache', action='store_true',
                        help="Overwrite the cached training and evaluation sets")
    parser.add_argument('--seed', type=int, default=42,
                        help="random seed for initialization")

    parser.add_argument('--fp16', action='store_true',
                        help="Whether to use 16-bit (mixed) precision (through NVIDIA apex) instead of 32-bit")
    parser.add_argument('--fp16_opt_level', type=str, default='O1',
                        help="For fp16: Apex AMP optimization level selected in ['O0', 'O1', 'O2', and 'O3']."
                             "See details at https://nvidia.github.io/apex/amp.html")
    parser.add_argument("--local_rank", type=int, default=-1,
                        help="For distributed training: local_rank")
    parser.add_argument('--server_ip', type=str, default='', help="For distant debugging.")
    parser.add_argument('--server_port', type=str, default='', help="For distant debugging.")
    parser.add_argument('--save_total_limit', type=int, default=1,
                        help='Limit the total amount of checkpoints, delete the older checkpoints in the output_dir, does not delete by default')

    parser.add_argument("--train_file",  default="train.json", type=str)
    parser.add_argument("--dev_file",  default="dev.json", type=str)
    parser.add_argument("--test_file",  default="ent_pred_test.json", type=str)
    parser.add_argument('--max_pair_length', type=int, default=128,  help="")
    parser.add_argument("--eval_pred_ent", action='store_true', help="")
    parser.add_argument('--max_mention_length', type=int, default=12,  help="")
    parser.add_argument("--alpha", default=1.0, type=float)
    parser.add_argument('--save_results', action='store_true')
    parser.add_argument('--no_test', action='store_true')
    parser.add_argument('--eval_logsoftmax', action='store_true')
    parser.add_argument('--eval_softmax', action='store_true')
    parser.add_argument('--shuffle', action='store_true')
    parser.add_argument('--lminit', action='store_true')
    parser.add_argument('--no_sym', action='store_true')
    parser.add_argument('--att_left', action='store_true')
    parser.add_argument('--att_right', action='store_true')
    parser.add_argument('--use_ner_results', action='store_true')

    args = parser.parse_args()

    if os.path.exists(args.output_dir) and os.listdir(args.output_dir) and args.do_train and not args.overwrite_output_dir:
        raise ValueError("Output directory ({}) already exists and is not empty. Use --overwrite_output_dir to overcome.".format(args.output_dir))

    def create_exp_dir(path, scripts_to_save=None):
        if args.output_dir.endswith("test"):
            return
        if not os.path.exists(path):
            os.mkdir(path)

        print('Experiment dir : {}'.format(path))
        if scripts_to_save is not None:
            if not os.path.exists(os.path.join(path, 'scripts')):
                os.mkdir(os.path.join(path, 'scripts'))
            for script in scripts_to_save:
                dst_file = os.path.join(path, 'scripts', os.path.basename(script))
                shutil.copyfile(script, dst_file)

    if args.do_train and args.local_rank in [-1, 0] and args.output_dir.find('test')==-1:
        create_exp_dir(args.output_dir, scripts_to_save=['run_re_unidrect.py', 'transformers/src/transformers/modeling_bert.py', 'transformers/src/transformers/modeling_albert.py'])


    # Setup distant debugging if needed
    if args.server_ip and args.server_port:
        # Distant debugging - see https://code.visualstudio.com/docs/python/debugging#_attach-to-a-local-script
        import ptvsd
        print("Waiting for debugger attach")
        ptvsd.enable_attach(address=(args.server_ip, args.server_port), redirect_output=True)
        ptvsd.wait_for_attach()

    # Setup CUDA, GPU & distributed training
    if args.local_rank == -1 or args.no_cuda:
        device = torch.device("cuda" if torch.cuda.is_available() and not args.no_cuda else "cpu")
        args.n_gpu = torch.cuda.device_count()
    else:  # Initializes the distributed backend which will take care of sychronizing nodes/GPUs
        torch.cuda.set_device(args.local_rank)
        device = torch.device("cuda", args.local_rank)
        torch.distributed.init_process_group(backend='nccl')
        args.n_gpu = 1
    args.device = device

    # Setup logging
    logging.basicConfig(format = '%(asctime)s - %(levelname)s - %(name)s -   %(message)s',
                        datefmt = '%m/%d/%Y %H:%M:%S',
                        level = logging.INFO if args.local_rank in [-1, 0] else logging.WARN)
    logger.warning("Process rank: %s, device: %s, n_gpu: %s, distributed training: %s, 16-bits training: %s",
                    args.local_rank, device, args.n_gpu, bool(args.local_rank != -1), args.fp16)

    # Set seed
    set_seed(args)
    # assert (not args.no_sym)
    if args.data_dir.find('ace')!=-1:
        num_ner_labels = 8
        num_labels = 7
        # if args.no_sym:
        #     num_labels = 7 + 7 - 1
        # else:
        #     num_labels = 7 + 7 - 2
    elif args.data_dir.find('scierc')!=-1:
        num_ner_labels = 7
        num_labels = 8
        # if args.no_sym:
        #     num_labels = 8 + 8 - 1
        # else:
        #     num_labels = 8 + 8 - 3
    else:
        assert (False)
    # Load pretrained model and tokenizer
    if args.local_rank not in [-1, 0]:
        torch.distributed.barrier()  # Make sure only the first process in distributed training will download model & vocab


    args.model_type = args.model_type.lower()

    config_class, model_class, tokenizer_class = MODEL_CLASSES[args.model_type]

    config = config_class.from_pretrained(args.config_name if args.config_name else args.model_name_or_path, num_labels=num_labels)
    tokenizer = tokenizer_class.from_pretrained(args.model_name_or_path,  do_lower_case=args.do_lower_case)

    config.max_seq_length = args.max_seq_length
    config.alpha = args.alpha
    config.num_ner_labels = num_ner_labels

    model = model_class.from_pretrained(args.model_name_or_path, from_tf=bool('.ckpt' in args.model_name_or_path), config=config)


    if args.model_type.startswith('albert'):
        special_tokens_dict = {'additional_special_tokens': ['[unused' + str(x) + ']' for x in range(4)]}
        tokenizer.add_special_tokens(special_tokens_dict)
        model.albert.resize_token_embeddings(len(tokenizer))

    if args.do_train:
        subject_id = tokenizer.encode('subject', add_special_tokens=False)
        assert(len(subject_id)==1)
        subject_id = subject_id[0]
        object_id = tokenizer.encode('object', add_special_tokens=False)
        assert(len(object_id)==1)
        object_id = object_id[0]

        mask_id = tokenizer.encode('[MASK]', add_special_tokens=False)
        assert(len(mask_id)==1)
        mask_id = mask_id[0]

        logger.info(" subject_id = %s, object_id = %s, mask_id = %s", subject_id, object_id, mask_id)

        if args.lminit:
            if args.model_type.startswith('albert'):
                word_embeddings = model.albert.embeddings.word_embeddings.weight.data
                subs = 30000
                sube = 30001
                objs = 30002
                obje = 30003
            else:
                word_embeddings = model.bert.embeddings.word_embeddings.weight.data
                subs = 1
                sube = 2
                objs = 3
                obje = 4

            word_embeddings[subs].copy_(word_embeddings[mask_id])   # subject
            word_embeddings[sube].copy_(word_embeddings[subject_id])    

            word_embeddings[objs].copy_(word_embeddings[mask_id])   # object
            word_embeddings[obje].copy_(word_embeddings[object_id])      


    if args.local_rank == 0:
        torch.distributed.barrier()  # Make sure only the first process in distributed training will download model & vocab

    model.to(args.device)

    logger.info("Training/evaluation parameters %s", args)
    best_f1 = 0
    # Training
    if args.do_train:
        # train_dataset = load_and_cache_examples(args,  tokenizer, evaluate=False)
        global_step, tr_loss, best_f1 = train(args, model, tokenizer)
        logger.info(" global_step = %s, average loss = %s", global_step, tr_loss)

    # Saving best-practices: if you use defaults names for the model, you can reload it using from_pretrained()
    if args.do_train and (args.local_rank == -1 or torch.distributed.get_rank() == 0):
        # Create output directory if needed
        if not os.path.exists(args.output_dir) and args.local_rank in [-1, 0]:
            os.makedirs(args.output_dir)
        update = True
        if args.evaluate_during_training:
            results = evaluate(args, model, tokenizer)
            f1 = results['f1_with_ner']
            # f1 = results['f1']

            if f1 > best_f1:
                best_f1 = f1
                print ('Best F1', best_f1)
            else:
                update = False

        if update:
            checkpoint_prefix = 'checkpoint'
            output_dir = os.path.join(args.output_dir, '{}-{}'.format(checkpoint_prefix, global_step))
            if not os.path.exists(output_dir):
                os.makedirs(output_dir)
            model_to_save = model.module if hasattr(model, 'module') else model  # Take care of distributed/parallel training

            model_to_save.save_pretrained(output_dir)

            torch.save(args, os.path.join(output_dir, 'training_args.bin'))
            logger.info("Saving model checkpoint to %s", output_dir)
            _rotate_checkpoints(args, checkpoint_prefix)

        tokenizer.save_pretrained(args.output_dir)

        torch.save(args, os.path.join(args.output_dir, 'training_args.bin'))


    # Evaluation
    results = {'dev_best_f1': best_f1}
    if args.do_eval and args.local_rank in [-1, 0]:
        checkpoints = [args.output_dir]

        WEIGHTS_NAME = 'pytorch_model.bin'

        if args.eval_all_checkpoints:
            checkpoints = list(os.path.dirname(c) for c in sorted(glob.glob(args.output_dir + '/**/' + WEIGHTS_NAME, recursive=True)))

        # logging.getLogger("pytorch_transformers.modeling_utils").setLevel(logging.WARN)  # Reduce logging
        logger.info("Evaluate on test set")

        logger.info("Evaluate the following checkpoints: %s", checkpoints)
        for checkpoint in checkpoints:
            global_step = checkpoint.split('-')[-1] if len(checkpoints) > 1 else ""

            model = model_class.from_pretrained(checkpoint, config=config)

            model.to(args.device)
            result = evaluate(args, model, tokenizer, prefix=global_step, do_test=not args.no_test)
            # print (global_step, result)
            result = dict((k + '_{}'.format(global_step), v) for k, v in result.items())
            results.update(result)
        print (results)

        if args.no_test:
            bestv = 0
            k = 0
            for k, v in results.items():
                if v > bestv:
                    bestk = k
            print (bestk)

        output_eval_file = os.path.join(args.output_dir, "results.json")
        json.dump(results, open(output_eval_file, "w"))

if __name__ == "__main__":
    main()