run_speech.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.
# Copyright 2023, Shumin Deng
#
# 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
#
#     https://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 multiple choice (Bert, Roberta, DistilBert)."""

import sys
sys.path.append("../")

import argparse
import glob
import logging
import os
import random

import numpy as np
import torch
from torch.utils.data import DataLoader, RandomSampler, SequentialSampler, TensorDataset
from torch.utils.data.distributed import DistributedSampler
from tqdm import tqdm, trange

from transformers import (
    WEIGHTS_NAME,
    AdamW,
    PretrainedConfig,
    BERT_PRETRAINED_CONFIG_ARCHIVE_MAP,
    ROBERTA_PRETRAINED_CONFIG_ARCHIVE_MAP,
    XLNET_PRETRAINED_CONFIG_ARCHIVE_MAP,
    DISTILBERT_PRETRAINED_CONFIG_ARCHIVE_MAP,
    CAMEMBERT_PRETRAINED_CONFIG_ARCHIVE_MAP,
    # XLMROBERTA_PRETRAINED_CONFIG_ARCHIVE_MAP, 
    BertConfig,
    BertForTokenClassification,
    BertForMultipleChoice,
    BertTokenizer,
    RobertaModel,
    RobertaConfig,
    RobertaTokenizer,
    XLNetModel,
    XLNetConfig,
    XLNetTokenizer,
    CamembertConfig,
    CamembertForTokenClassification,
    CamembertTokenizer,
    DistilBertConfig,
    DistilBertForTokenClassification,
    DistilBertTokenizer,
    # XLMRobertaConfig,
    # XLMRobertaForTokenClassification,
    # XLMRobertaTokenizer,
    get_linear_schedule_with_warmup,
)

from data_utils import convert_examples_to_features, processors
from sklearn.metrics import f1_score, precision_score, recall_score, classification_report
from speech import SPEECH
from speech_roberta import SPEECH_Roberta
from speech_distilbert import SPEECH_DistilBert


logger = logging.getLogger(__name__)

ALL_MODELS = sum(
    # (tuple(conf.pretrained_config_archive_map.keys()) for conf in (BertConfig, XLNetConfig, RobertaConfig, DistilBertConfig, CamembertConfig, XLMRobertaConfig)), ()
    (tuple(conf_map.keys()) for conf_map in (BERT_PRETRAINED_CONFIG_ARCHIVE_MAP, ROBERTA_PRETRAINED_CONFIG_ARCHIVE_MAP, XLNET_PRETRAINED_CONFIG_ARCHIVE_MAP, DISTILBERT_PRETRAINED_CONFIG_ARCHIVE_MAP, CAMEMBERT_PRETRAINED_CONFIG_ARCHIVE_MAP)), 
    ()
)

MODEL_CLASSES = {
    "bert": (BertConfig, BertForTokenClassification, BertTokenizer),
    "xlnet": (XLNetConfig, XLNetModel, XLNetTokenizer),
    "roberta": (RobertaConfig, RobertaModel, RobertaTokenizer),
    "distilbert": (DistilBertConfig, DistilBertForTokenClassification, DistilBertTokenizer),
    "camembert": (CamembertConfig, CamembertForTokenClassification, CamembertTokenizer),
    # "xlmroberta": (XLMRobertaConfig, XLMRobertaForTokenClassification, XLMRobertaTokenizer),
    "speech_bert": (BertConfig, SPEECH, BertTokenizer),
    "speech_roberta": (RobertaConfig, SPEECH_Roberta, RobertaTokenizer),
    "speech_distilbert": (DistilBertConfig, SPEECH_DistilBert, DistilBertTokenizer),
    # "speech_camembert": (CamembertConfig, SPEECH, CamembertTokenizer),
    # "speech_xlnet": (XLNetConfig, SPEECH, XLNetTokenizer),
    # "speech_xlmroberta": (XLMRobertaConfig, SPEECH, XLMRobertaTokenizer),
}


def calculate_scores(preds, labels, dimE, task_type):
    """
        task_type:  "token"; "sent"; "sent_onto", "doc_all"; "doc_temporal"; "doc_causal"; "doc_sub"; "doc_corref"; "doc_joint" 
    """
    if task_type == "token":
        positive_labels = list(range(0, dimE - 1)) 
    elif task_type == "sent": 
        positive_labels = list(range(1, dimE)) 
    elif task_type == "sent_onto": 
        positive_labels = list(range(2, dimE)) 
    elif "doc" in task_type: 
        positive_labels = list(range(1, dimE)) 

    p_micro = precision_score(y_true=labels, y_pred=preds, labels=positive_labels, average='micro') 
    r_micro = recall_score(y_true=labels, y_pred=preds, labels=positive_labels, average='micro')
    f1_micro = f1_score(y_true=labels, y_pred=preds, labels=positive_labels, average='micro')
    
    return p_micro, r_micro, f1_micro


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 train(args, train_dataset, model, tokenizer):
    """ Train the model """
    args.train_batch_size = args.per_gpu_train_batch_size * max(1, args.n_gpu)
    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)

    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)
    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)

    # 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
        )
    # torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm)
    """Training!"""
    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
    best_valid_f1_micro = 0.0
    best_steps = 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
    for _ in train_iterator:
        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 = {
                "example_id": batch[0],
                "task_name": args.task_name,
                "doc_ere_task_type": args.ere_task_type,  
                "max_mention_size": torch.tensor([args.max_mention_size], dtype=torch.long),  
                "mention_size": batch[1],
                "pad_token_label_id": batch[2], 
                "input_ids": batch[3].view(-1, args.max_seq_length),
                "attention_mask": batch[4].view(-1, args.max_seq_length),
                "token_type_ids": batch[5].view(-1, args.max_seq_length)
                if args.model_type not in ["xlmroberta"] or (args.model_type.startswith("xlmroberta") is False)
                else None,  # XLM don't use segment_ids
                "labels4token": batch[6],
                "labels4sent": batch[7],
                "mat_rel_label": batch[8],
            }
 
            outputs = model(**inputs)
            loss = outputs[0]  # model outputs are always tuple in transformers (see doc)

            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

            if args.fp16:
                with amp.scale_loss(loss, optimizer) as scaled_loss:
                    scaled_loss.backward()
                torch.nn.utils.clip_grad_norm_(amp.master_params(optimizer), args.max_grad_norm)
            else:
                loss.backward()
                torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_grad_norm)

            tr_loss += loss.item()
            if (step + 1) % args.gradient_accumulation_steps == 0:

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

                if args.local_rank in [-1, 0] and args.logging_steps > 0 and global_step % args.logging_steps == 0:
                    # Log metrics
                    if (
                        args.local_rank == -1 and args.evaluate_during_training
                    ):  # Only evaluate when single GPU otherwise metrics may not average well
                        results = evaluate(args, model, tokenizer)
                        # central_task = "doc" # token, sent, doc, doc_temporal_joint, doc_causal_joint, doc_sub_joint, doc_corref_joint
                        central_task = args.central_task
                        if args.ere_task_type == "doc_joint":
                            central_task = "doc_temporal_joint"
                        # central_task = "sent"
                        # central_task = "token"
                        key_metric = "eval_f1_micro_" + central_task

                        if results[key_metric] > best_valid_f1_micro:
                            best_valid_f1_micro = results[key_metric]
                            best_steps = global_step
                            if args.do_test:
                                results_test = evaluate(args, model, tokenizer, test=True)
                                logger.info(
                                    "test f1_micro_" + central_task + ": %s, loss: %s, global steps: %s", 
                                    str(results_test[key_metric]), 
                                    str(results_test["eval_loss"]),
                                    str(global_step),
                                )
                    
                    logger.info(
                        "Average loss: %s at global step: %s",
                        str((tr_loss - logging_loss) / args.logging_steps),
                        str(global_step),
                    )
                    logging_loss = tr_loss

                if args.local_rank in [-1, 0] and args.save_steps > 0 and global_step % args.save_steps == 0:
                    # Save model checkpoint
                    output_dir = os.path.join(args.output_dir, "checkpoint-{}".format(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)
                    # # add
                    # basic = model.module if hasattr(model, "module") else model
                    # bert_to_save = (basic.bert.module if hasattr(basic.bert, "module") else basic.bert)
                    # tmp = os.path.join(output_dir, args.model_type)
                    # if not os.path.exists(tmp):
                    #     os.makedirs(tmp)
                    # bert_to_save.save_pretrained(tmp)
                    # # add end
                    tokenizer.save_vocabulary(output_dir)
                    torch.save(args, os.path.join(output_dir, "training_args.bin"))
                    logger.info("Saving model checkpoint to [%s]", output_dir)
                    torch.save(optimizer.state_dict(), os.path.join(output_dir, "optimizer.pt"))
                    torch.save(scheduler.state_dict(), os.path.join(output_dir, "scheduler.pt"))
                    logger.info("Saving optimizer and scheduler states to %s", output_dir)

            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

    return global_step, tr_loss / global_step, best_steps


def evaluate(args, model, tokenizer, prefix="", test=False, infer=True):
    eval_task_names = (args.task_name,)
    eval_outputs_dirs = (args.output_dir,)

    results = {}
    for eval_task, eval_output_dir in zip(eval_task_names, eval_outputs_dirs):
        eval_dataset = load_and_cache_examples(args, eval_task, tokenizer, evaluate=not test, test=test)

        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)
        # Note that DistributedSampler samples randomly
        eval_sampler = SequentialSampler(eval_dataset)
        eval_dataloader = DataLoader(eval_dataset, sampler=eval_sampler, batch_size=args.eval_batch_size)

        """Evaluation!"""
        logger.info("***** Running evaluation {} *****".format(prefix))
        logger.info("  Num examples = %d", len(eval_dataset))
        logger.info("  Batch size = %d", args.eval_batch_size)
        eval_loss = 0.0
        nb_eval_steps = 0
        preds_token = None
        preds_sent = None
        preds_doc = None
        preds_doc_temp = None
        preds_doc_causal = None
        preds_doc_sub = None
        preds_doc_corref = None
        out_label4token_ids = None
        out_label4sent_ids = None
        out_label4doc_ids = None
        out_label4doc_temp_ids = None 
        out_label4doc_causal_ids = None  
        out_label4doc_sub_ids = None  
        out_label4doc_corref_ids = None   
        for batch in tqdm(eval_dataloader, desc="Evaluating"):
            model.eval()
            batch = tuple(t.to(args.device) for t in batch)

            with torch.no_grad():
                inputs = {
                    "example_id": batch[0],
                    "task_name": args.task_name,
                    "doc_ere_task_type": args.ere_task_type, 
                    "max_mention_size": torch.tensor([args.max_mention_size], dtype=torch.long),  
                    "mention_size": batch[1],
                    "pad_token_label_id": batch[2], 
                    "input_ids": batch[3].view(-1, args.max_seq_length),
                    "attention_mask": batch[4].view(-1, args.max_seq_length),
                    "token_type_ids": batch[5].view(-1, args.max_seq_length)
                    if args.model_type not in ["xlmroberta"] or (args.model_type.startswith("xlmroberta") is False)
                    else None,  # XLM don't use segment_ids
                    "labels4token": batch[6],
                    "labels4sent": batch[7],
                    "mat_rel_label": batch[8],
                }
                outputs = model(**inputs)
                if "_ere" not in args.model_type and "_ec" not in args.model_type:
                    if args.ere_task_type != "doc_joint":
                        tmp_eval_loss, logits_doc, label_doc, logits_sent, label_sent, logits_token, label_tokens = outputs[:7]
                    else:
                        if args.task_name == "maven-ere":
                            tmp_eval_loss, logits_doc_temp, label_doc_temp, logits_doc_causal, label_doc_causal, logits_doc_sub, label_doc_sub, logits_doc_corref, label_doc_corref, logits_sent, label_sent, logits_token, label_tokens = outputs[:13]
                        else:
                            tmp_eval_loss, logits_doc_temp, label_doc_temp, logits_doc_causal, label_doc_causal, logits_doc_sub, label_doc_sub, logits_sent, label_sent, logits_token, label_tokens = outputs[:11]
                elif "_ere" in args.model_type and "_ec" not in args.model_type and "_ed" not in args.model_type:
                    if args.ere_task_type != "doc_joint":
                        tmp_eval_loss, logits_doc, label_doc = outputs[:3]
                    else:
                        if args.task_name == "maven-ere":
                            tmp_eval_loss, logits_doc_temp, label_doc_temp, logits_doc_causal, label_doc_causal, logits_doc_sub, label_doc_sub, logits_doc_corref, label_doc_corref = outputs[:9]
                        else:
                            tmp_eval_loss, logits_doc_temp, label_doc_temp, logits_doc_causal, label_doc_causal, logits_doc_sub, label_doc_sub = outputs[:7] 
                eval_loss += tmp_eval_loss.mean().item()
            nb_eval_steps += 1

            if "_ere" not in args.model_type: # or, only for ERE task 
                if "_ec" not in args.model_type: # or, only for EC task 
                    # for token-level task
                    if preds_token is None:
                        preds_token = logits_token.detach().cpu().numpy()
                        out_label4token_ids = label_tokens.detach().cpu().numpy()
                    else:
                        preds_token = np.append(preds_token, logits_token.detach().cpu().numpy(), axis=0)
                        out_label4token_ids = np.append(out_label4token_ids, label_tokens.detach().cpu().numpy(), axis=0)
                # for sentence-level task
                if preds_sent is None:
                    preds_sent = logits_sent.detach().cpu().numpy()
                    out_label4sent_ids = label_sent.detach().cpu().numpy()
                else:
                    preds_sent = np.append(preds_sent, logits_sent.detach().cpu().numpy(), axis=0)
                    out_label4sent_ids = np.append(out_label4sent_ids, label_sent.detach().cpu().numpy(), axis=0)
            # for document-level task
            if args.ere_task_type != "doc_joint":
                if preds_doc is None:
                    preds_doc = logits_doc.detach().cpu().numpy()
                    out_label4doc_ids = label_doc.detach().cpu().numpy()
                else:
                    preds_doc = np.append(preds_doc, logits_doc.detach().cpu().numpy(), axis=0)
                    out_label4doc_ids = np.append(out_label4doc_ids, label_doc.detach().cpu().numpy(), axis=0)
            else: 
                if preds_doc_temp is None:
                    preds_doc_temp = logits_doc_temp.detach().cpu().numpy()
                    out_label4doc_temp_ids = label_doc_temp.detach().cpu().numpy()
                else:
                    preds_doc_temp = np.append(preds_doc_temp, logits_doc_temp.detach().cpu().numpy(), axis=0)
                    out_label4doc_temp_ids = np.append(out_label4doc_temp_ids, label_doc_temp.detach().cpu().numpy(), axis=0)

                if preds_doc_causal is None:
                    preds_doc_causal = logits_doc_causal.detach().cpu().numpy()
                    out_label4doc_causal_ids = label_doc_causal.detach().cpu().numpy()
                else:
                    preds_doc_causal = np.append(preds_doc_causal, logits_doc_causal.detach().cpu().numpy(), axis=0)
                    out_label4doc_causal_ids = np.append(out_label4doc_causal_ids, label_doc_causal.detach().cpu().numpy(), axis=0)

                if preds_doc_sub is None:
                    preds_doc_sub = logits_doc_sub.detach().cpu().numpy()
                    out_label4doc_sub_ids = label_doc_sub.detach().cpu().numpy()
                else:
                    preds_doc_sub = np.append(preds_doc_sub, logits_doc_sub.detach().cpu().numpy(), axis=0)
                    out_label4doc_sub_ids = np.append(out_label4doc_sub_ids, label_doc_sub.detach().cpu().numpy(), axis=0)
                
                if args.task_name == "maven-ere":
                    if preds_doc_corref is None:
                        preds_doc_corref = logits_doc_corref.detach().cpu().numpy()
                        out_label4doc_corref_ids = label_doc_corref.detach().cpu().numpy()
                    else:
                        preds_doc_corref = np.append(preds_doc_corref, logits_doc_corref.detach().cpu().numpy(), axis=0)
                        out_label4doc_corref_ids = np.append(out_label4doc_corref_ids, label_doc_corref.detach().cpu().numpy(), axis=0)

        eval_loss = eval_loss / nb_eval_steps
        if "_ere" not in args.model_type:
            preds_token = np.argmax(preds_token, axis=1)
            preds_sent = np.argmax(preds_sent, axis=1)
        if args.ere_task_type != "doc_joint": 
            preds_doc = np.argmax(preds_doc, axis=1)
        else:
            preds_doc_temp = np.argmax(preds_doc_temp, axis=1)
            preds_doc_causal = np.argmax(preds_doc_causal, axis=1)
            preds_doc_sub = np.argmax(preds_doc_sub, axis=1)
            if args.task_name == "maven-ere": 
                preds_doc_corref = np.argmax(preds_doc_corref, axis=1)

        if "_ere" not in args.model_type and "_ec" not in args.model_type: # or, only for ECE or EC task
            p_micro_token, r_micro_token, f1_micro_token = calculate_scores(preds_token, out_label4token_ids, len(processors[eval_task]().get_labels4tokens()), "token")
            if args.task_name == "ontoevent-doc":
                p_micro_sent, r_micro_sent, f1_micro_sent = calculate_scores(preds_sent, out_label4sent_ids, len(processors[eval_task]().get_labels4sent()), "sent_onto")
            elif args.task_name == "maven-ere":
                p_micro_sent, r_micro_sent, f1_micro_sent = calculate_scores(preds_sent, out_label4sent_ids, len(processors[eval_task]().get_labels4sent()), "sent")
        elif "_ec" in args.model_type: # or, only for EC task
            p_micro_token = None 
            r_micro_token = None 
            f1_micro_token = None
            if args.task_name == "ontoevent-doc":
                p_micro_sent, r_micro_sent, f1_micro_sent = calculate_scores(preds_sent, out_label4sent_ids, len(processors[eval_task]().get_labels4sent()), "sent_onto")
            elif args.task_name == "maven-ere":
                p_micro_sent, r_micro_sent, f1_micro_sent = calculate_scores(preds_sent, out_label4sent_ids, len(processors[eval_task]().get_labels4sent()), "sent")
            p_micro_doc = None 
            r_micro_doc = None 
            f1_micro_doc = None  
        else: # only for ECE task
            p_micro_token = None 
            r_micro_token = None 
            f1_micro_token = None 
            p_micro_sent = None 
            r_micro_sent = None 
            f1_micro_sent = None 

        if args.ere_task_type == "doc_all": 
            p_micro_doc, r_micro_doc, f1_micro_doc = calculate_scores(preds_doc, out_label4doc_ids, len(processors[eval_task]().get_labels4doc()), args.ere_task_type) 
        else:
            if args.task_name == "ontoevent-doc":
                if args.ere_task_type == "doc_temporal":  
                    p_micro_doc, r_micro_doc, f1_micro_doc = calculate_scores(preds_doc, out_label4doc_ids, 1+3, args.ere_task_type)
                elif args.ere_task_type == "doc_causal":  
                    p_micro_doc, r_micro_doc, f1_micro_doc = calculate_scores(preds_doc, out_label4doc_ids, 1+2, args.ere_task_type)
                elif args.ere_task_type == "doc_sub":  
                    p_micro_doc, r_micro_doc, f1_micro_doc = calculate_scores(preds_doc, out_label4doc_ids, 1+3, args.ere_task_type)
                elif args.ere_task_type == "doc_joint":  
                    p_micro_temporal_joint, r_micro_temporal_joint, f1_micro_temporal_joint  = calculate_scores(preds_doc_temp, out_label4doc_temp_ids, 1+3, "doc_temporal")
                    p_micro_causal_joint, r_micro_causal_joint, f1_micro_causal_joint = calculate_scores(preds_doc_causal, out_label4doc_causal_ids, 1+2, "doc_causal")
                    p_micro_sub_joint, r_micro_sub_joint, f1_micro_sub_joint = calculate_scores(preds_doc_sub, out_label4doc_sub_ids, 1+3, "doc_sub")
            elif args.task_name == "maven-ere":
                if args.ere_task_type == "doc_temporal":  
                    p_micro_doc, r_micro_doc, f1_micro_doc = calculate_scores(preds_doc, out_label4doc_ids, 1+6, args.ere_task_type)
                elif args.ere_task_type == "doc_causal":  
                    p_micro_doc, r_micro_doc, f1_micro_doc = calculate_scores(preds_doc, out_label4doc_ids, 1+2, args.ere_task_type)
                elif args.ere_task_type == "doc_sub":  
                    p_micro_doc, r_micro_doc, f1_micro_doc = calculate_scores(preds_doc, out_label4doc_ids, 1+1, args.ere_task_type)
                elif args.ere_task_type == "doc_corref":  
                    p_micro_doc, r_micro_doc, f1_micro_doc = calculate_scores(preds_doc, out_label4doc_ids, 1+1, args.ere_task_type)
                elif args.ere_task_type == "doc_joint":  
                    p_micro_temporal_joint, r_micro_temporal_joint, f1_micro_temporal_joint  = calculate_scores(preds_doc_temp, out_label4doc_temp_ids, 1+6, "doc_temporal")
                    p_micro_causal_joint, r_micro_causal_joint, f1_micro_causal_joint = calculate_scores(preds_doc_causal, out_label4doc_causal_ids, 1+2, "doc_causal")
                    p_micro_sub_joint, r_micro_sub_joint, f1_micro_sub_joint = calculate_scores(preds_doc_sub, out_label4doc_sub_ids, 1+1, "doc_sub")
                    p_micro_corref_joint, r_micro_corref_joint, f1_micro_corref_joint = calculate_scores(preds_doc_corref, out_label4doc_corref_ids, 1+1, "doc_corref") 
                
        if infer:
            if "_ere" not in args.model_type: # or, only for document-level task
                np.save(os.path.join(eval_output_dir, str(prefix) + "_preds-token.npy"), preds_token)
                np.save(os.path.join(eval_output_dir, str(prefix) + "_preds-sentence.npy"), preds_sent)
            if args.ere_task_type != "doc_joint":   
                np.save(os.path.join(eval_output_dir, str(prefix) + "_preds-document.npy"), preds_doc)
            else:
                if args.task_name == "ontoevent-doc":
                    np.save(os.path.join(eval_output_dir, str(prefix) + "_preds-document_temporal.npy"), preds_doc_temp)
                    np.save(os.path.join(eval_output_dir, str(prefix) + "_preds-document_causal.npy"), preds_doc_causal)
                elif args.task_name == "maven-ere": 
                    np.save(os.path.join(eval_output_dir, str(prefix) + "_preds-document_temporal.npy"), preds_doc_temp)
                    np.save(os.path.join(eval_output_dir, str(prefix) + "_preds-document_causal.npy"), preds_doc_causal)
                    np.save(os.path.join(eval_output_dir, str(prefix) + "_preds-document_subevent.npy"), preds_doc_sub)

        if args.ere_task_type != "doc_joint":  
            result = {
                "eval_p_micro_token": p_micro_token, "eval_r_micro_token": r_micro_token, "eval_f1_micro_token": f1_micro_token, 
                "eval_p_micro_sent": p_micro_sent, "eval_r_micro_sent": r_micro_sent, "eval_f1_micro_sent": f1_micro_sent, 
                "eval_p_micro_doc": p_micro_doc, "eval_r_micro_doc": r_micro_doc, "eval_f1_micro_doc": f1_micro_doc, 
                "eval_loss": eval_loss
            }
        else: 
            if args.task_name == "ontoevent-doc":
                result = {
                    "eval_p_micro_token": p_micro_token, "eval_r_micro_token": r_micro_token, "eval_f1_micro_token": f1_micro_token, 
                    "eval_p_micro_sent": p_micro_sent, "eval_r_micro_sent": r_micro_sent, "eval_f1_micro_sent": f1_micro_sent, 
                    "eval_p_micro_doc_temporal_joint": p_micro_temporal_joint, "eval_r_micro_doc_temporal_joint": r_micro_temporal_joint, "eval_f1_micro_doc_temporal_joint": f1_micro_temporal_joint, 
                    "eval_p_micro_doc_causal_joint": p_micro_causal_joint, "eval_r_micro_doc_causal_joint": r_micro_causal_joint, "eval_f1_micro_doc_causal_joint": f1_micro_causal_joint, 
                    "eval_loss": eval_loss
                }
            elif args.task_name == "maven-ere":
                result = {
                    "eval_p_micro_token": p_micro_token, "eval_r_micro_token": r_micro_token, "eval_f1_micro_token": f1_micro_token, 
                    "eval_p_micro_sent": p_micro_sent, "eval_r_micro_sent": r_micro_sent, "eval_f1_micro_sent": f1_micro_sent, 
                    "eval_p_micro_doc_temporal_joint": p_micro_temporal_joint, "eval_r_micro_doc_temporal_joint": r_micro_temporal_joint, "eval_f1_micro_doc_temporal_joint": f1_micro_temporal_joint, 
                    "eval_p_micro_doc_causal_joint": p_micro_causal_joint, "eval_r_micro_doc_causal_joint": r_micro_causal_joint, "eval_f1_micro_doc_causal_joint": f1_micro_causal_joint, 
                    "eval_p_micro_doc_sub_joint": p_micro_sub_joint, "eval_r_micro_doc_sub_joint": r_micro_sub_joint, "eval_f1_micro_doc_sub_joint": f1_micro_sub_joint,   
                    "eval_loss": eval_loss
                } 
        
        results.update(result)

        output_eval_file = os.path.join(eval_output_dir, "is_test_" + str(test).lower() + "_eval_results.txt")

        with open(output_eval_file, "w") as writer:
            logger.info("***** Eval results {} *****".format(str(prefix) + "are test: " + str(test)))
            writer.write("model            = %s\n" % str(args.model_name_or_path))
            writer.write(
                "total batch size = %d\n"
                % (
                    args.per_gpu_train_batch_size
                    * args.gradient_accumulation_steps
                    * (torch.distributed.get_world_size() if args.local_rank != -1 else 1)
                )
            )
            writer.write("train num epochs = %d\n" % args.num_train_epochs)
            writer.write("fp16             = %s\n" % args.fp16)
            writer.write("max seq length   = %d\n" % args.max_seq_length)
            for key in sorted(result.keys()):
                logger.info("%s = %s", key, str(result[key]))
                writer.write("%s = %s\n" % (key, str(result[key])))
    return results


def load_and_cache_examples(args, task, tokenizer, evaluate=False, test=False):
    if args.local_rank not in [-1, 0]:
        torch.distributed.barrier()  # Make sure only the first process in distributed training process the dataset, and the others will use the cache

    processor = processors[task]()
    # Load data features from cache or dataset file
    if evaluate:
        cached_mode = "valid"
    elif test:
        cached_mode = "test"
    else:
        cached_mode = "train"
    assert not (evaluate and test)
    cached_features_file = os.path.join(
        args.data_dir,
        "Cached_{}_{}_{}_{}_{}".format(
            cached_mode,
            list(filter(None, args.model_name_or_path.split("/"))).pop(),
            str(args.max_seq_length),
            str(args.max_mention_size),
            str(task),
        ),
    )
    if os.path.exists(cached_features_file) and not args.overwrite_cache:
        logger.info("Loading features from cached file %s", cached_features_file)
        features = torch.load(cached_features_file)
    else:
        logger.info("Creating features from dataset file at %s", args.data_dir)
        label4sent_list = processor.get_labels4sent()
        label4token_list = processor.get_labels4tokens()
        label4rel_list = processor.get_labels4doc() 
        if evaluate:
            examples = processor.get_valid_examples(args.data_dir)
        elif test:
            examples = processor.get_test_examples(args.data_dir)
        else:
            examples = processor.get_train_examples(args.data_dir)      
        logger.info("Training number: %s", str(len(examples)))
        
        features = convert_examples_to_features(
            examples,
            label4token_list,
            label4sent_list,
            label4rel_list,
            args.max_seq_length,
            args.max_mention_size,
            tokenizer,
            cls_token_at_end=bool(args.model_type.startswith("xlnet")), # xlnet has a cls token at the end,
            cls_token=tokenizer.cls_token,
            cls_token_segment_id=2 if args.model_type.startswith("xlnet") else 0,
            sep_token=tokenizer.sep_token,
            sep_token_extra=bool(args.model_type.startswith("roberta")), # roberta uses an extra separator b/w pairs of sentences, cf. github.com/pytorch/fairseq/commit/1684e166e3da03f5b600dbb7855cb98ddfcd0805
            pad_on_left=bool(args.model_type.startswith("xlnet")), # pad on the left for xlnet
            pad_token=tokenizer.convert_tokens_to_ids([tokenizer.pad_token])[0],
            pad_token_segment_id=4 if args.model_type.startswith("xlnet") else 0,
            model_name=args.model_name_or_path,
            task_name=args.task_name
        )
        if args.local_rank in [-1, 0]:
            logger.info("Saving features into cached file [%s]", cached_features_file)
            torch.save(features, cached_features_file)

    if args.local_rank == 0:
        torch.distributed.barrier()  # Make sure only the first process in distributed training process the dataset, and the others will use the cache

    # Convert to Tensors and build dataset
    all_example_id = torch.tensor([f.example_id for f in features], dtype=torch.long)
    all_mention_size = torch.tensor([f.mention_size for f in features], dtype=torch.long)
    all_pad_token_label_id = torch.tensor([f.pad_token_label_id for f in features], dtype=torch.long) 
    all_list_input_ids = torch.tensor([f.list_input_ids for f in features], dtype=torch.long)
    all_list_input_mask = torch.tensor([f.list_input_mask for f in features], dtype=torch.long)
    all_list_segment_ids = torch.tensor([f.list_segment_ids for f in features], dtype=torch.long)
    all_list_label4token_ids = torch.tensor([f.list_token_labels for f in features], dtype=torch.long)
    all_list_label4sent_ids = torch.tensor([f.list_sent_label for f in features], dtype=torch.long)
    all_mat_rel_label_ids = torch.tensor([f.mat_rel_label for f in features], dtype=torch.long)
    
    dataset = TensorDataset(all_example_id, all_mention_size, all_pad_token_label_id, all_list_input_ids, all_list_input_mask, all_list_segment_ids, all_list_label4token_ids, all_list_label4sent_ids, all_mat_rel_label_ids)
    return dataset


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

    # Required parameters
    parser.add_argument(
        "--data_dir",
        default=None,
        type=str,
        required=True,
        help="The input data directory.",
    )
    parser.add_argument(
        "--model_type",
        default=None,
        type=str,
        required=True,
        help="Model type selected in the list: " + ", ".join(MODEL_CLASSES.keys()),
    ) # "speech_bert, speech_roberta, speech_distilbert"
    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),
    ) # "bert-base-uncased, roberta-base, distilbert-base-uncased"
    parser.add_argument(
        "--task_name",
        default=None,
        type=str,
        required=True,
        help="The name of the task to train selected in the list: " + ", ".join(processors.keys()),
    ) # "ontoevent-doc, maven-ere"
    parser.add_argument(
        "--central_task",
        default=None,
        type=str,
        required=True,
        help="The central task for optimization: " + 
            "token, " + "sent, " + "doc",
    ) # "token, sent, doc"
    parser.add_argument(
        "--ere_task_type",
        default=None,
        type=str,
        required=True,
        help="The type of doc ere task to train selected in the list: " + 
            "doc_all, " + "doc_joint, " + "doc_temporal, " +  "doc_causal, " +  "doc_sub, " +  "doc_corref" + 
            "doc_all: for \"All Joint\" in the paper; " + 
            "doc_joint: for each ERE subtask +joint; " + 
            "doc_temporal/doc_causal/doc_sub: for each ERE subtask only",
    )
    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=128,
        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(
        "--max_mention_size",
        default=50,
        type=int,
        help="The maximum size of event mentions in on document. The event mention size of each document should not larger than it, documents 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 valid set.")
    parser.add_argument("--do_test", action="store_true", help="Whether to run test on the test set") 
    # Note that for maven-ere datasets, we only evaluate on the valid set, thus "--do_test" should be dismissed for experiments on maven-ere 
    parser.add_argument(
        "--evaluate_during_training", action="store_true", help="Run 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=1, type=int, help="Batch size per GPU/CPU for training.")
    parser.add_argument(
        "--per_gpu_eval_batch_size", default=1, 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=5e-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=3.0, 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=0, type=int, help="Linear warmup over warmup_steps.")

    parser.add_argument("--logging_steps", type=int, default=50, help="Log every X updates steps.")
    parser.add_argument("--save_steps", type=int, default=50, 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.")
    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
            )
        )

    # 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:0" if torch.cuda.is_available() and not args.no_cuda else "cpu")
        args.n_gpu = 0 if args.no_cuda else 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)

    # Prepare for task
    args.task_name = args.task_name.lower()
    if args.task_name not in processors:
        raise ValueError("Task not found: %s" % (args.task_name))
    processor = processors[args.task_name]()
    label4token_list = processor.get_labels4tokens()
    num_labels4token = len(label4token_list) 
    label4sent_list = processor.get_labels4sent()
    num_labels4sent = len(label4sent_list)
    label4doc_list = processor.get_labels4doc()
    num_labels4doc = len(label4doc_list) 

    # 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_labels4token,
        finetuning_task=args.task_name,
        cache_dir=args.cache_dir if args.cache_dir else None,
    )
    tokenizer = tokenizer_class.from_pretrained(
        args.tokenizer_name if args.tokenizer_name else args.model_name_or_path,
        do_lower_case=args.do_lower_case,
        cache_dir=args.cache_dir if args.cache_dir else None,
    )
    model = model_class.from_pretrained(
        args.model_name_or_path,
        from_tf=bool(".ckpt" in args.model_name_or_path),
        config=config,
        cache_dir=args.cache_dir if args.cache_dir else None,
    )

    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_steps = 0

    """Training"""
    if args.do_train:
        train_dataset = load_and_cache_examples(args, args.task_name, tokenizer, evaluate=False)
        global_step, tr_loss, best_steps = train(args, train_dataset, 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)

        logger.info("Saving model checkpoint to [%s]", args.output_dir)
        # Save a trained model, configuration and tokenizer using `save_pretrained()`.
        # They can then be reloaded using `from_pretrained()`
        model_to_save = (
            model.module if hasattr(model, "module") else model
        )  # Take care of distributed/parallel training
        model_to_save.save_pretrained(args.output_dir)
        tokenizer.save_pretrained(args.output_dir)

        # Good practice: save your training arguments together with the trained model
        torch.save(args, os.path.join(args.output_dir, "training_args.bin"))

        # Load a trained model and vocabulary that you have fine-tuned
        model = model_class.from_pretrained(args.output_dir)
        tokenizer = tokenizer_class.from_pretrained(args.output_dir)
        model.to(args.device)

    """Evaluation"""
    results = {}
    if args.do_eval and args.local_rank in [-1, 0]:
        if not args.do_train:
            args.output_dir = args.model_name_or_path
        checkpoints = [args.output_dir]
        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("transformers.modeling_utils").setLevel(logging.WARN)  # Reduce logging
        logger.info("Evaluate the following checkpoints: [%s]", checkpoints)
        for checkpoint in checkpoints:
            global_step = checkpoint.split("-")[-1] if len(checkpoints) > 1 else ""
            prefix = checkpoint.split("/")[-1] if checkpoint.find("checkpoint") != -1 else ""

            model = model_class.from_pretrained(checkpoint)
            model.to(args.device)
            result = evaluate(args, model, tokenizer, prefix=prefix)
            result = dict((k + "_{}".format(global_step), v) for k, v in result.items())
            results.update(result)

    if args.do_test and args.local_rank in [-1, 0]:
        if not args.do_train:
            args.output_dir = args.model_name_or_path
        checkpoints = [args.output_dir]

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

            model = model_class.from_pretrained(checkpoint)
            model.to(args.device)
            result = evaluate(args, model, tokenizer, prefix=prefix, test=True)
            result = dict((k + "_{}".format(global_step), v) for k, v in result.items())
            results.update(result)

    if best_steps:
        logger.info("best steps of eval f1 is the following checkpoints: %s", best_steps)
    return results


if __name__ == "__main__":
    main()