bert_reranker.py

# -*- coding: utf-8 -*-
"""bert_reranker_yield.ipynb

Automatically generated by Colaboratory.

Original file is located at
    https://colab.research.google.com/drive/1Ef6bVuyVC6aUEFbKG597NF_NC5wVvMNA
"""

#!pip install transformers==2.2.1
#!pip install tensorboardX
#!pip install pandas

from __future__ import absolute_import, division, print_function
import argparse
import glob
import logging
import os
import random
import pickle
import re
import numpy as np
import pandas as pd
import torch
from torch.utils.data import (DataLoader, RandomSampler, WeightedRandomSampler, SequentialSampler,
                              TensorDataset)
from torch.utils.data.distributed import DistributedSampler
from tensorboardX import SummaryWriter
from tqdm import tqdm, trange
import sys
from transformers import (WEIGHTS_NAME, BertConfig,
                          BertForSequenceClassification, BertTokenizer,
                          XLMConfig, XLMForSequenceClassification,
                          XLMTokenizer, XLNetConfig,
                          XLNetForSequenceClassification,
                          XLNetTokenizer)

from transformers import AdamW, get_linear_schedule_with_warmup

logger = logging.getLogger(__name__)

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

MODEL_CLASSES = {
    'bert': (BertConfig, BertForSequenceClassification, BertTokenizer),
    'xlnet': (XLNetConfig, XLNetForSequenceClassification, XLNetTokenizer),
    'xlm': (XLMConfig, XLMForSequenceClassification, XLMTokenizer),
}

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 compute_AP(sorted_labels):
    num_hits = 0.0
    ap = 0.0

    if np.sum(sorted_labels) == 0:
        return 0.0

    for i, ll in enumerate(sorted_labels):
        if ll == 0:
            continue
        num_hits += 1
        ap += num_hits / (i + 1)

    return ap / np.sum(sorted_labels)


def compute_metrics(examples, preds, labels):
    logit_1 = preds[:, 1] - preds[:, 0]
    #print("logits diff: ",logit_1)
    acc = (np.argmax(preds, axis=1) == labels).mean()
    print("accuracy: ",acc)
    print(len(examples),len(preds),len(labels))
    idx_start = 0
    prev_query = examples[0].text_a

    ap_list = []
    predictions = []

    for i, example in enumerate(examples):
        #print("processing example: ",i, example.text_a,prev_query)
        if example.text_a == prev_query:
            continue

        relevant_logits = logit_1[idx_start:i]
        relevant_labels = labels[idx_start:i]
        relevant_examples = examples[idx_start:i]
        #print(relevant_logits,relevant_labels,relevant_examples)
        sorted_preds, sorted_labels, sorted_examples = zip(*sorted(zip(relevant_logits, relevant_labels,relevant_examples), key=lambda e: e[0],reverse=True))
        predictions.extend([se.guid for se in sorted_examples])
       
      #  assert(len(sorted_labels) == 4950)
        ap_list.append(compute_AP(sorted_labels))

        prev_query = example.text_a
        idx_start = i

    relevant_logits = logit_1[idx_start:]
    relevant_labels = labels[idx_start:]
    relevant_examples = examples[idx_start:]
    sorted_preds, sorted_labels, sorted_examples = zip(*sorted(zip(relevant_logits, relevant_labels, relevant_examples),key=lambda e: e[0], reverse=True))
    predictions.extend([se.guid for se in sorted_examples])
   # assert (len(sorted_labels) == 4950)
    ap_list.append(compute_AP(sorted_labels))

    return {"acc": acc, "map": np.mean(ap_list)}, predictions

class InputExample(object):
    """A single training/test example for simple sequence classification."""

    def __init__(self, guid, text_a, text_b=None, label=None):
        """Constructs a InputExample.

        Args:
            guid: Unique id for the example.
            text_a: string. The untokenized text of the first sequence. For single
            sequence tasks, only this sequence must be specified.
            text_b: (Optional) string. The untokenized text of the second sequence.
            Only must be specified for sequence pair tasks.
            label: (Optional) string. The label of the example. This should be
            specified for train and dev examples, but not for test examples.
        """
        self.guid = guid
        self.text_a = text_a
        self.text_b = text_b
        self.label = label


class InputFeatures(object):
    """A single set of features of data."""

    def __init__(self, input_ids, input_mask, segment_ids, label_id):
        self.input_ids = input_ids
        self.input_mask = input_mask
        self.segment_ids = segment_ids
        self.label_id = label_id

# String used to indicate a blank
BLANK_STR = "___"


# Create a hypothesis statement from the the input fill-in-the-blank statement and answer choice.
def create_hypothesis(fitb, choice):
    if ". " + BLANK_STR in fitb or fitb.startswith(BLANK_STR):
        choice = choice[0].upper() + choice[1:]
    else:
        choice = choice.lower()
    if not fitb.endswith(BLANK_STR):
        choice = choice.rstrip(".")
    hypothesis = re.sub("__+", choice.strip(), fitb)
    return hypothesis
	
# Identify the wh-word in the question and replace with a blank
def replace_wh_word_with_blank(question_str):
    wh_word_offset_matches = []
    wh_words = ["which", "what", "where", "when", "how", "who", "why"]
    for wh in wh_words:
        m = re.search(wh + "\?[^\.]*[\. ]*$", question_str.lower())
        if m:
            wh_word_offset_matches = [(wh, m.start())]
            break
        else:
            m = re.search(wh + "[ ,][^\.]*[\. ]*$", question_str.lower())
            if m:
                wh_word_offset_matches.append((wh, m.start()))

    if len(wh_word_offset_matches):
       
        wh_word_offset_matches.sort(key=lambda x: x[1])
        wh_word_found = wh_word_offset_matches[0][0]
        wh_word_start_offset = wh_word_offset_matches[0][1]
        question_str = re.sub("\?$", ".", question_str.strip())
        fitb_question = (question_str[:wh_word_start_offset] + BLANK_STR +
                         question_str[wh_word_start_offset + len(wh_word_found):])
        
        return fitb_question.replace(BLANK_STR + " of the following", BLANK_STR)
    elif re.match(".*[^\.\?] *$", question_str):
        
        return question_str + " " + BLANK_STR
    else:
        return re.sub(" this[ \?]", " ___ ", question_str)
		
def get_fitb_from_question(question_text):
    fitb = replace_wh_word_with_blank(question_text)
    if not re.match(".*_+.*", fitb):
        print("Can't create hypothesis from: '{}'. Appending {} !".format(question_text, BLANK_STR))
        fitb = re.sub("[\.\? ]*$", "", question_text.strip()) + BLANK_STR
    return fitb

class TG2019RerankProcessor:
    def get_train_examples(self, args):
        """Gets a collection of `InputExample`s for the train set."""
        df_questions = pd.read_csv(args.train_questions_file, sep='\t').dropna(subset=["explanation"]).reset_index()
        return self._create_examples(df_questions, args.mcq_choices, args.facts_file)

    def get_dev_examples(self, args):
        """Gets a collection of `InputExample`s for the dev set."""
        df_questions = pd.read_csv(args.dev_questions_file, sep='\t').dropna(subset=["explanation"]).reset_index()
        return self._create_examples(df_questions, args.mcq_choices, args.facts_file)

    def get_test_examples(self, args):
        """Gets a collection of `InputExample`s for the test set."""
        df_questions = pd.read_csv(args.test_questions_file, sep='\t')
        df_questions = df_questions[df_questions['flags'].str.lower().isin(('success', 'ready'))]
        return self._create_examples(df_questions, args.mcq_choices, args.facts_file)

    def get_labels(self):
        """Gets the list of labels for this data set."""
        return ["0", "1"]


    def _create_examples(self, df_questions, mcq_choices, facts_file):
        # Remove wrong choices
        def remove_wrong_answer_choices(row, choices):
            
            delimiters="(A)","(B)","(C)","(D)","(E)","(1)","(2)","(3)","(4)"
            regexPattern = '|'.join(map(re.escape, delimiters))
            ques=row['question']
            qa=re.split(regexPattern, ques)
            #print(qa)
            ans=row['AnswerKey']
		#print(qa)
            if ans=="A" or ans=="1":
                ch=qa[1]
            elif ans=="B" or ans=="2":
                ch=qa[2]
            elif ans=="C" or ans=="3":
                ch=qa[3]
            elif ans=="D" or ans=="4":
                ch=qa[4]
            else:
                ch=qa[5]
            fitb_q=get_fitb_from_question(qa[0])
            processed_q=create_hypothesis(fitb_q,ch)
            return processed_q
			


        if mcq_choices != "all":
            df_questions["ProcessedQuestion"] = df_questions.apply(remove_wrong_answer_choices, 1,
                                                                   choices=mcq_choices)
        else:
            df_questions["ProcessedQuestion"] = df_questions["question"]
        #for q in df_questions["ProcessedQuestion"]:
         #   print(q)
        df_facts = pd.read_csv(facts_file, sep='\t', index_col="uid")

        examples = []
        for i_q, question in df_questions.iterrows():
            if i_q % 10 ==0:
                print(i_q)
            qid = question["QuestionID"]
            explanations = [e.split('|')[0] for e in question["explanation"].split(' ')]
            for uid, fact in df_facts.iterrows():
                guid = '###'.join([qid, uid])
                examples.append(InputExample(guid=guid, text_a=question["ProcessedQuestion"], text_b=fact["text"],
                                             label="1" if uid in explanations else "0"))
        return examples

def _truncate_seq_pair(tokens_a, tokens_b, max_length):
    """Truncates a sequence pair in place to the maximum length."""

    # This is a simple heuristic which will always truncate the longer sequence
    # one token at a time. This makes more sense than truncating an equal percent
    # of tokens from each, since if one sequence is very short then each token
    # that's truncated likely contains more information than a longer sequence.
    while True:
        total_length = len(tokens_a) + len(tokens_b)
        if total_length <= max_length:
            break
        if len(tokens_a) > len(tokens_b):
            tokens_a.pop()
        else:
            tokens_b.pop()


def check_sampling_index(num_samples,examples):
    
    label_list=["0", "1"]

    label_map = {label: i for i, label in enumerate(label_list)}
    count=0
    li=[]
    print('creating label array...')
    for example in examples:
        label_id = label_map[example.label]
        if(label_id)==0 and count<=10000000:
            li.append(label_id)
            count+=1 
        elif (label_id)==0 and count>10000000:
            continue        
        else:
            li.append(label_id)
    print("number of examples being used: ",len(li))            
    label_array = np.asarray(li, dtype=np.int32)
    class_pos=np.sum(label_array==1)
    class_neg=np.sum(label_array==0)

    print('number of samples in both classes:')
    print(class_pos,class_neg)
    sampler_weights = np.zeros(len(label_array))
    label_map = {label:i for i, label in enumerate(label_list)}
    #print(label_map)
    for label in label_list:
        sampler_weights[label_array == label_map[label]] = 1. / np.mean(label_array == label_map[label])
    #print(sampler_weights)
    print("sampler weightes created..")
    train_sampler = WeightedRandomSampler(sampler_weights, num_samples=num_samples, replacement= False)
    sampling_indices=list(train_sampler)
    print("sampler indices created...")
    with open('sampled_indices.pkl','wb') as wr:
        pickle.dump(sampling_indices,wr)
    return set(sampling_indices)

def get_sampled_indices():
    with open('sampled_indices.pkl','rb') as wr:
        indices=pickle.load(wr)
    return set(indices)
	
def convert_examples_to_features(filename,batch_size,examples, label_list, max_seq_length,
                                 tokenizer, output_mode,
                                 cls_token_at_end=False, pad_on_left=False,
                                 cls_token='[CLS]', sep_token='[SEP]', pad_token=0,
                                 sequence_a_segment_id=0, sequence_b_segment_id=1,
                                 cls_token_segment_id=1, pad_token_segment_id=0,
                                 mask_padding_with_zero=True):
    """ Loads a data file into a list of `InputBatch`s
        `cls_token_at_end` define the location of the CLS token:
            - False (Default, BERT/XLM pattern): [CLS] + A + [SEP] + B + [SEP]
            - True (XLNet/GPT pattern): A + [SEP] + B + [SEP] + [CLS]
        `cls_token_segment_id` define the segment id associated to the CLS token (0 for BERT, 2 for XLNet)
    """
    
    if "train" in filename:
        total_samples_to_generate=500000
        if (os.path.exists('sampled_indices.pkl')):
            sampling_indices=get_sampled_indices()
        else:
            sampling_indices=check_sampling_index(total_samples_to_generate,examples)

    label_map = {label: i for i, label in enumerate(label_list)}
    print("number of examples to be processed: ",len(examples))
 #   with open(filename,'ab') as p:
    for (ex_index, example) in enumerate(examples):
        #feat=[]

        dict={}
        if "train" in filename:
            if ex_index not in sampling_indices:
                continue 
        if ex_index % 50000 == 0:
            logger.info("Writing example %d of %d" % (ex_index, len(examples)))

        tokens_a = tokenizer.tokenize(example.text_a)

        tokens_b = None
        if example.text_b:
            tokens_b = tokenizer.tokenize(example.text_b)
            # Modifies `tokens_a` and `tokens_b` in place so that the total
            # length is less than the specified length.
            # Account for [CLS], [SEP], [SEP] with "- 3"
            _truncate_seq_pair(tokens_a, tokens_b, max_seq_length - 3)
        else:
             # Account for [CLS] and [SEP] with "- 2"
            if len(tokens_a) > max_seq_length - 2:
                tokens_a = tokens_a[:(max_seq_length - 2)]


        tokens = tokens_a + [sep_token]
        segment_ids = [sequence_a_segment_id] * len(tokens)

        if tokens_b:
            tokens += tokens_b + [sep_token]
            segment_ids += [sequence_b_segment_id] * (len(tokens_b) + 1)

        if cls_token_at_end:
            tokens = tokens + [cls_token]
            segment_ids = segment_ids + [cls_token_segment_id]
        else:
            tokens = [cls_token] + tokens
            segment_ids = [cls_token_segment_id] + segment_ids

        input_ids = tokenizer.convert_tokens_to_ids(tokens)

        # The mask has 1 for real tokens and 0 for padding tokens. Only real
        # tokens are attended to.
        input_mask = [1 if mask_padding_with_zero else 0] * len(input_ids)

        # Zero-pad up to the sequence length.
        padding_length = max_seq_length - len(input_ids)
        if pad_on_left:
            input_ids = ([pad_token] * padding_length) + input_ids
            input_mask = ([0 if mask_padding_with_zero else 1] * padding_length) + input_mask
            segment_ids = ([pad_token_segment_id] * padding_length) + segment_ids
        else:
            input_ids = input_ids + ([pad_token] * padding_length)
            input_mask = input_mask + ([0 if mask_padding_with_zero else 1] * padding_length)
            segment_ids = segment_ids + ([pad_token_segment_id] * padding_length)

        assert len(input_ids) == max_seq_length
        assert len(input_mask) == max_seq_length
        assert len(segment_ids) == max_seq_length

        if output_mode == "classification":
            label_id = label_map[example.label]
        elif output_mode == "regression":
            label_id = float(example.label)
        else:
            raise KeyError(output_mode)

        if ex_index < 5:
            logger.info("*** Example ***")
            logger.info("guid: %s" % (example.guid))
            logger.info("tokens: %s" % " ".join(
                    [str(x) for x in tokens]))
            logger.info("input_ids: %s" % " ".join([str(x) for x in input_ids]))
            logger.info("input_mask: %s" % " ".join([str(x) for x in input_mask]))
            logger.info("segment_ids: %s" % " ".join([str(x) for x in segment_ids]))
            logger.info("label: %s (id = %d)" % (example.label, label_id))
        #feat.append(InputFeatures(input_ids=input_ids,input_mask=input_mask,segment_ids=segment_ids,label_id=label_id))
        dict['input_ids']= input_ids
        dict['input_mask']=input_mask
        dict['segment_ids']=segment_ids
        dict['label_id']=label_id		
      
        with open(filename,'ab') as p:
            pickle.dump(dict,p)
          

   
				
from torch.utils.data import IterableDataset
class MyDataset(IterableDataset):

    def __init__(self,args,filename):
        self.args=args
        self.filename=filename

					
    def process(self,filename):
        with open(self.filename, "rb") as f:
            while True:
                try:
                    yield pickle.load(f)
                except EOFError:
                    break 
        
      
    def __iter__(self):
        dataset=self.process(self.filename)       
        return dataset

def load_examples(args, task, mode='train'):
    print("current mode: ",mode)
    processor = TG2019RerankProcessor()
    output_mode = "classification"
    # Load data features from cache or dataset file
    cached_examples_file = os.path.join(args.data_dir, 'examples_{}_{}_{}_{}'.format(
        mode,
        list(filter(None, args.model_name_or_path.split('/'))).pop(),
        str(args.max_seq_length),
        str(task)))
    if args.cached_examples_file:
        cached_examples_file = args.cached_examples_file
  

    if os.path.exists(cached_examples_file):
        logger.info("Loading examples from cached file %s", cached_examples_file)
        examples = torch.load(cached_examples_file)
    else:
        if mode == 'train':
            logger.info("Creating examples from dataset file at %s", args.train_questions_file)
            examples = processor.get_train_examples(args)
        elif mode == 'dev':
            logger.info("Creating examples from dataset file at %s", args.dev_questions_file)
            examples = processor.get_dev_examples(args)
        elif mode == 'test':
            logger.info("Creating examples from dataset file at %s", args.test_questions_file)
            examples = processor.get_test_examples(args)
        else:
            raise ValueError("Unhandled mode {}".format(mode))
        if args.local_rank in [-1, 0]:
            logger.info("Saving examples into cached file %s", cached_examples_file)
            torch.save(examples, cached_examples_file)
    return examples

#import _pickle as cPickle


def train(args, train_dataset, model, tokenizer, label_list, num_samples):
    """ Train the model """
    if args.local_rank in [-1, 0]:
        tb_writer = SummaryWriter()

    args.train_batch_size = args.per_gpu_train_batch_size * max(1, args.n_gpu)
 
    #train_dataloader = DataLoader(train_dataset, sampler=train_sampler, batch_size=args.train_batch_size)
    
    #train_dataloader_iter = iter(train_dataloader)

    train_dataloader = DataLoader(train_dataset, batch_size=args.train_batch_size)
    num_batches=num_samples//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
        args.num_train_epochs = args.max_steps //( num_batches // args.gradient_accumulation_steps) + 1
    else:
      #  t_total = len(train_dataloader) // args.gradient_accumulation_steps * args.num_train_epochs
         t_total =  num_batches// 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.to('cuda'), optimizer, opt_level=args.fp16_opt_level)
  #  print("len of train dataloader: ",len(train_dataloader))
   # print("len of train dataset: ",len(train_dataset))
    # Train!
    logger.info("***** Running training *****")
    logger.info("  Num examples = %d", num_samples)
    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
    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)
    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):
        #while True:
         #   step=0
          #  batch = next(train_dataloader_iter)
            model.train()
            #batch = tuple(t.to(args.device) for t in batch)
            a=torch.stack(list(map(torch.stack, zip(*(batch['input_ids']))))).to(args.device)
            b=torch.stack(list(map(torch.stack, zip(*(batch['input_mask']))))).to(args.device)
            c=torch.stack(list(map(torch.stack, zip(*(batch['segment_ids']))))).to(args.device)
            d=batch['label_id'].to(args.device)

            inputs = {'input_ids':      a,
                      'attention_mask': b,
                      'token_type_ids': c if args.model_type in ['bert', 'xlnet'] else None,  # XLM don't use segment_ids
                      'labels':         d}
         
            outputs = model(**inputs)
            loss = outputs[0]  # model outputs are always tuple in pytorch-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:
                scheduler.step()  # Update learning rate schedule
                optimizer.step()
                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
                    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

                if args.local_rank == -1 and args.evaluate_during_training and global_step % args.evaluation_steps == 0:
                    # Only evaluate when single GPU otherwise metrics may not average well
                    results = evaluate(args, model, tokenizer, label_list)
                    for key, value in results.items():
                        tb_writer.add_scalar('eval_{}'.format(key), value, global_step)

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

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

    return global_step, tr_loss / global_step

def evaluate(args, model, tokenizer, label_list, prefix=""):
    eval_task = args.task_name
    eval_output_dir = args.output_dir

    results = {}

    eval_examples=load_examples(args,eval_task,mode='dev')
    mode='dev'
    cached_features_file = os.path.join(args.data_dir,'features_{}_{}_{}_{}'.format(mode,list(filter(None, args.model_name_or_path.split('/'))).pop(),str(args.max_seq_length),str(eval_task)))
    if args.cached_features_file:
        cached_features_file = args.cached_features_file

    if os.path.exists(cached_features_file):
        logger.info("Loading features from cached file %s", cached_features_file)	
    else:
        convert_examples_to_features(cached_features_file,args.per_gpu_eval_batch_size,eval_examples, label_list, args.max_seq_length, tokenizer, args.output_mode,
            cls_token_at_end=bool(args.model_type in ['xlnet']),            # xlnet has a cls token at the end
            cls_token=tokenizer.cls_token,
            sep_token=tokenizer.sep_token,
            cls_token_segment_id=2 if args.model_type in ['xlnet'] else 1,
            pad_on_left=bool(args.model_type in ['xlnet']),                 # pad on the left for xlnet
            pad_token_segment_id=4 if args.model_type in ['xlnet'] else 0) 
        print("written features to disk, now loading file...")
			
    eval_dataset=MyDataset(args,cached_features_file)
    
 
    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) if args.local_rank == -1 else DistributedSampler(eval_dataset)
    #eval_dataloader = DataLoader(eval_dataset, sampler=eval_sampler, batch_size=args.eval_batch_size)
    eval_dataloader = DataLoader(eval_dataset, batch_size=args.eval_batch_size)

    # Eval!
    logger.info("***** Running evaluation {} *****".format(prefix))
    logger.info("  Num examples = %d", len(eval_examples))
    logger.info("  Batch size = %d", args.eval_batch_size)
    eval_loss = 0.0
    nb_eval_steps = 0
    preds = None
    out_label_ids = None
    eval_file=os.path.join(eval_output_dir, "tg2020_eval_preds{}.npy".format(prefix))
    with open(eval_file,'ab') as f:
        for batch in tqdm(eval_dataloader, desc="Evaluating"):
            model.eval()
        #batch = tuple(t.to(args.device) for t in batch)
            a=torch.stack(list(map(torch.stack, zip(*(batch['input_ids']))))).to(args.device)
            b=torch.stack(list(map(torch.stack, zip(*(batch['input_mask']))))).to(args.device)
            c=torch.stack(list(map(torch.stack, zip(*(batch['segment_ids']))))).to(args.device)
            d=batch['label_id'].to(args.device)

            with torch.no_grad():
           
                inputs = {'input_ids':      a,
                      'attention_mask': b,
                      'token_type_ids': c if args.model_type in ['bert', 'xlnet'] else None,  # XLM don't use segment_ids
                      'labels':         d}
                outputs = model(**inputs)
                tmp_eval_loss, logits = outputs[:2]

                eval_loss += tmp_eval_loss.mean().item()
            nb_eval_steps += 1
            preds = logits.detach().cpu().numpy()
            np.save(f, preds)

            out_label_ids = inputs['labels'].detach().cpu().numpy()

    eval_loss = eval_loss / nb_eval_steps
  #  print("predictions: ",preds)
   # print("output label id",out_label_ids)

    return results

def predict(args, model, tokenizer, label_list, prefix=""):
    eval_task = args.task_name
    eval_output_dir = args.output_dir

    results = {}

    mode='test'
    eval_examples=load_examples(args,eval_task,mode='test')
    cached_features_file = os.path.join(args.data_dir,'features_{}_{}_{}_{}'.format(mode,list(filter(None, args.model_name_or_path.split('/'))).pop(),str(args.max_seq_length),str(eval_task)))
    if args.cached_features_file:
        cached_features_file = args.cached_features_file

    if os.path.exists(cached_features_file):
        logger.info("Loading features from cached file %s", cached_features_file)
    else:
        logger.info("Creating features from cached examples file.... ")

        convert_examples_to_features(cached_features_file,args.per_gpu_train_batch_size,eval_examples, label_list, args.max_seq_length, tokenizer, args.output_mode,
            cls_token_at_end=bool(args.model_type in ['xlnet']),            # xlnet has a cls token at the end
            cls_token=tokenizer.cls_token,
            sep_token=tokenizer.sep_token,
            cls_token_segment_id=2 if args.model_type in ['xlnet'] else 1,
            pad_on_left=bool(args.model_type in ['xlnet']),                 # pad on the left for xlnet
            pad_token_segment_id=4 if args.model_type in ['xlnet'] else 0) 
        print("written features to disk, now loading file...")

    eval_dataset=MyDataset(args,cached_features_file)
    
    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) if args.local_rank == -1 else DistributedSampler(eval_dataset)
    #eval_dataloader = DataLoader(eval_dataset, sampler=eval_sampler, batch_size=args.eval_batch_size)
    eval_dataloader = DataLoader(eval_dataset, batch_size=args.eval_batch_size)

    # Predict!
    logger.info("***** Running prediction {} *****".format(prefix))
    logger.info("  Num examples = %d", len(eval_examples))
    logger.info("  Batch size = %d", args.eval_batch_size)
    nb_eval_steps = 0
    preds = None
    eval_file=os.path.join(eval_output_dir, "tg2020_test_preds{}.npy".format(prefix))

    for batch in tqdm(eval_dataloader, desc="Evaluating"):
        model.eval()
        #batch = tuple(t.to(args.device) for t in batch)
        a=torch.stack(list(map(torch.stack, zip(*(batch['input_ids']))))).to(args.device)
        b=torch.stack(list(map(torch.stack, zip(*(batch['input_mask']))))).to(args.device)
        c=torch.stack(list(map(torch.stack, zip(*(batch['segment_ids']))))).to(args.device)

            
        with torch.no_grad():
            inputs = {'input_ids':      a,
                      'attention_mask': b,
                      'token_type_ids': c if args.model_type in ['bert', 'xlnet'] else None,  # XLM don't use segment_ids
                      }
            outputs = model(**inputs)
            logits = outputs[0]

        nb_eval_steps += 1
        preds = logits.detach().cpu().numpy()
        with open(eval_file,'ab') as f:
            np.save(f, preds)

base_path=''
class Args1:
    TRAIN_QUESTIONS_FILE= base_path + "questions/questions.train.tsv"
    DEV_QUESTIONS_FILE= base_path +"questions/questions.dev.tsv"
    TEST_QUESTIONS_FILE= base_path +"questions/questions.test.tsv"
    FACTS_FILE= base_path +"questions/explanations.tsv"


    DATA_DIR=base_path +"questions"

    RERANK_TRAIN_SEQ=72
    RERANK_PRED_SEQ=140
    RERANK_OUTPUT_DIR= base_path +"outputs/bert_rerank_correctchoices_unweighted/"

    PATH_RANK_TRAIN_SEQ=90
    PATH_RANK_TRAIN_TFIDF=25
    PATH_RANK_PRED_SEQ=140
    PATH_RANK_PRED_TFIDF=50
    PATH_RANK_OUTPUT_DIR= base_path + "outputs/bert_path_rank_correctchoices_unweighted_k25_1e/"

    model_type= "bert"
    model_name_or_path= "bert-base-uncased" 
    task_name= "TG2020"
    do_train=1
    do_eval=1
    do_lower_case=1
    data_dir=DATA_DIR
    max_seq_length=RERANK_TRAIN_SEQ
    per_gpu_eval_batch_size=64
    per_gpu_train_batch_size=64
    learning_rate= 2e-5 
    num_train_epochs= 3.0 
    logging_steps= 5000 
    evaluate_during_training= 1 
    evaluation_steps= 25000 
    save_steps= 5000 
    output_dir=RERANK_OUTPUT_DIR
    train_questions_file=TRAIN_QUESTIONS_FILE
    dev_questions_file=DEV_QUESTIONS_FILE
    facts_file=FACTS_FILE
    mcq_choices= "correct"
    do_predict = 0
    ## Other parameters

    weighted_sampling=0
    config_name=""
    tokenizer_name=""
    cached_examples_file=""
    cached_features_file=""
    cache_dir=""
    gradient_accumulation_steps = 1
    weight_decay = 0.0
    adam_epsilon = 1e-8
    max_grad_norm = 1.0
    overwrite_cache= 1
    overwrite_output_dir= 1
    no_cuda= 1
    logging_steps= 50
    eval_all_checkpoints = 1
    seed = 42
    warmup_steps = 0
    max_steps = -1
    fp16= 0
    fp16_opt_level= "O1"
    local_rank = -1


class Args2:
    TEST_QUESTIONS_FILE=base_path + "questions/questions.test.tsv"
    FACTS_FILE=base_path + "questions/explanations.tsv"
    DATA_DIR=base_path + "questions"
    RERANK_PRED_SEQ=140
    RERANK_OUTPUT_DIR=base_path + "outputs/bert_rerank_correctchoices_unweighted/"

  
  # Get predictions from bert-reranker (base predictions)
    model_type= "bert" 
    model_name_or_path= "bert-base-uncased"
    task_name= "TG2020"
    do_predict= 1
    do_lower_case= 1
    data_dir = DATA_DIR
    max_seq_length = RERANK_PRED_SEQ
    output_dir = RERANK_OUTPUT_DIR
    test_questions_file = TEST_QUESTIONS_FILE
    facts_file= FACTS_FILE
    mcq_choices= "correct"

    #Other parameters
    do_train=0
    do_eval=0
    evaluation_steps=50
    overwrite_cache= 1
    overwrite_output_dir= 1
    save_steps= 50
    logging_steps= 50
    eval_all_checkpoints = 1
    seed = 42
    warmup_steps = 0
    max_steps = -1
    fp16= 0
    fp16_opt_level= "O1"
    local_rank = -1
    per_gpu_train_batch_size = 64
    per_gpu_eval_batch_size = 64
    learning_rate = 5e-5
    evaluate_during_training = 1
    num_train_epochs = 3.0
    no_cuda=0
    weighted_sampling=0
    config_name=""
    tokenizer_name=""
    cached_examples_file=""
    cached_features_file=""
    cache_dir=""
    gradient_accumulation_steps = 1
    weight_decay = 0.0
    adam_epsilon = 1e-8
    max_grad_norm = 1.0

def main():
    mode=sys.argv[1]
    if mode=='train':
        args = Args1()
    if mode=='test':
        args = Args2()

    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 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)

    # Prepare task
    args.task_name = args.task_name.lower()
    processor = TG2019RerankProcessor()
    args.output_mode = "classification"
    label_list = processor.get_labels()
    num_labels = len(label_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_labels, finetuning_task=args.task_name)
    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)
    model = model_class.from_pretrained(args.model_name_or_path, from_tf=bool('.ckpt' in args.model_name_or_path), config=config)

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

    # Distributed and parallel training
    model.to(args.device)
    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)
    elif args.n_gpu > 1:
        model = torch.nn.DataParallel(model)

    logger.info("Training/evaluation parameters %s", args)

    # Training
    if args.do_train:
        task=args.task_name
        mode='train'
        train_examples=load_examples(args,task,mode)
        cached_features_file = os.path.join(args.data_dir,'features_{}_{}_{}_{}'.format(mode,list(filter(None, args.model_name_or_path.split('/'))).pop(),str(args.max_seq_length),str(task)))
        if args.cached_features_file:
            cached_features_file = args.cached_features_file

        if os.path.exists(cached_features_file):
            logger.info("Loading features from cached file %s", cached_features_file)	
        else:
            convert_examples_to_features(cached_features_file,args.per_gpu_train_batch_size,train_examples, label_list, args.max_seq_length, tokenizer, args.output_mode,
            cls_token_at_end=bool(args.model_type in ['xlnet']),            # xlnet has a cls token at the end
            cls_token=tokenizer.cls_token,
            sep_token=tokenizer.sep_token,
            cls_token_segment_id=2 if args.model_type in ['xlnet'] else 1,
            pad_on_left=bool(args.model_type in ['xlnet']),                 # pad on the left for xlnet
            pad_token_segment_id=4 if args.model_type in ['xlnet'] else 0) 
            print("written features to disk, now loading file...")
			
        train_dataset=MyDataset(args,cached_features_file)
        #print("type of object from mydataset: ",type(out))
        #print(type(train_dataset))  
        
        num_samples=500000
        print("number of training samples: ",num_samples)
        global_step, tr_loss = train(args, train_dataset, model, tokenizer, label_list, num_samples)
        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]:
        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("pytorch_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 ""
            model = model_class.from_pretrained(checkpoint)
            model.to(args.device)
            result = evaluate(args, model, tokenizer, label_list, prefix=global_step)
            result = dict((k + '_{}'.format(global_step), v) for k, v in result.items())
            results.update(result)

    # Prediction
    if args.do_predict and args.local_rank in [-1, 0]:
        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("pytorch_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 ""
            model = model_class.from_pretrained(checkpoint)
            model.to(args.device)
            predict(args, model, tokenizer, label_list, prefix=global_step)

    return results


if __name__ == "__main__":
    main()