run_finetuning_ar.py

# coding=utf-8
"""MPC-BERT-2.0 fine-tuning runner on the downstream task of addressee recognition."""

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

import os
from time import time

import tensorflow as tf

import modeling_speaker as modeling
import optimization

flags = tf.compat.v1.flags
FLAGS = flags.FLAGS

flags.DEFINE_string("train_dir", 'train.tfrecord',
                    "The input train data dir. Should contain the .tsv files (or other data files) for the task.")

flags.DEFINE_string("valid_dir", 'valid.tfrecord',
                    "The input valid data dir. Should contain the .tsv files (or other data files) for the task.")

flags.DEFINE_string("output_dir", 'output',
                    "The output directory where the model checkpoints will be written.")

flags.DEFINE_string("task_name", 'AddresseeRecognition',
                    "The name of the task to train.")

flags.DEFINE_string("bert_config_file", 'uncased_L-12_H-768_A-12/bert_config.json',
                    "The config json file corresponding to the pre-trained BERT model. "
                    "This specifies the model architecture.")

flags.DEFINE_string("vocab_file", 'uncased_L-12_H-768_A-12/vocab.txt',
                    "The vocabulary file that the BERT model was trained on.")

flags.DEFINE_string("init_checkpoint", 'uncased_L-12_H-768_A-12/bert_model.ckpt',
                    "Initial checkpoint (usually from a pre-trained BERT model).")

flags.DEFINE_bool("do_lower_case", True,
                  "Whether to lower case the input text. Should be True for uncased "
                  "models and False for cased models.")

flags.DEFINE_integer("max_seq_length", 320,
                     "The maximum total input sequence length after WordPiece tokenization. "
                     "Sequences longer than this will be truncated, and sequences shorter "
                     "than this will be padded.")

flags.DEFINE_integer("max_utr_num", 7,
                     "Maximum utterance number.")

flags.DEFINE_bool("do_train", True,
                  "Whether to run training.")

flags.DEFINE_float("warmup_proportion", 0.1,
                   "Proportion of training to perform linear learning rate warmup for. "
                   "E.g., 0.1 = 10% of training.")

flags.DEFINE_integer("train_batch_size", 12,
                     "Total batch size for training.")

flags.DEFINE_float("learning_rate", 2e-5,
                   "The initial learning rate for Adam.")

flags.DEFINE_integer("num_train_epochs", 5,
                     "Total number of training epochs to perform.")


def print_configuration_op(FLAGS):
    print('My Configurations:')
    for name, value in FLAGS.__flags.items():
        value = value.value
        if type(value) == float:
            print(' %s:\t %f' % (name, value))
        elif type(value) == int:
            print(' %s:\t %d' % (name, value))
        elif type(value) == str:
            print(' %s:\t %s' % (name, value))
        elif type(value) == bool:
            print(' %s:\t %s' % (name, value))
        else:
            print('%s:\t %s' % (name, value))
    print('End of configuration')


def count_data_size(file_name):
    sample_nums = 0
    for record in tf.compat.v1.python_io.tf_record_iterator(file_name):
        sample_nums += 1
    return sample_nums


def parse_exmp(serial_exmp):
    input_data = tf.compat.v1.parse_single_example(serial_exmp,
                                                   features={
                                                       "input_sents":
                                                           tf.compat.v1.FixedLenFeature([FLAGS.max_seq_length],
                                                                                        tf.int64),
                                                       "input_mask":
                                                           tf.compat.v1.FixedLenFeature([FLAGS.max_seq_length],
                                                                                        tf.int64),
                                                       "segment_ids":
                                                           tf.compat.v1.FixedLenFeature([FLAGS.max_seq_length],
                                                                                        tf.int64),
                                                       "speaker_ids":
                                                           tf.compat.v1.FixedLenFeature([FLAGS.max_seq_length],
                                                                                        tf.int64),
                                                       "cls_positions":
                                                           tf.compat.v1.FixedLenFeature([FLAGS.max_utr_num], tf.int64),
                                                       "label_ids":
                                                           tf.compat.v1.FixedLenFeature(
                                                               [FLAGS.max_utr_num * FLAGS.max_utr_num], tf.int64),
                                                       "label_weights":
                                                           tf.compat.v1.FixedLenFeature([FLAGS.max_utr_num],
                                                                                        tf.float32),
                                                   }
                                                   )
    # So cast all int64 to int32.
    for name in list(input_data.keys()):
        t = input_data[name]
        if t.dtype == tf.int64:
            t = tf.compat.v1.to_int32(t)
        input_data[name] = t

    input_sents = input_data["input_sents"]
    input_mask = input_data["input_mask"]
    segment_ids = input_data["segment_ids"]
    speaker_ids = input_data["speaker_ids"]
    cls_positions = input_data["cls_positions"]
    labels = input_data['label_ids']
    label_weights = input_data['label_weights']
    return input_sents, input_mask, segment_ids, speaker_ids, cls_positions, labels, label_weights


def gather_indexes(sequence_tensor, positions):
    """Gathers the vectors at the specific positions over a minibatch."""
    # sequence_tensor = [batch_size, seq_length, width]
    # positions = [batch_size, max_utr_num]

    sequence_shape = modeling.get_shape_list(sequence_tensor, expected_rank=3)
    batch_size = sequence_shape[0]
    seq_length = sequence_shape[1]
    width = sequence_shape[2]

    flat_offsets = tf.reshape(
        tf.range(0, batch_size, dtype=tf.int32) * seq_length, [-1, 1])  # [batch_size, 1]
    flat_positions = tf.reshape(positions + flat_offsets, [-1])  # [batch_size*max_utr_num, ]
    flat_sequence_tensor = tf.reshape(sequence_tensor,
                                      [batch_size * seq_length, width])
    output_tensor = tf.gather(flat_sequence_tensor, flat_positions)  # [batch_size*max_utr_num, width]
    return output_tensor


def create_model(bert_config, is_training, input_ids, input_mask, segment_ids, speaker_ids, cls_positions, labels,
                 label_weights,
                 use_one_hot_embeddings):
    """Creates a classification model."""
    model = modeling.BertModel(
        config=bert_config,
        is_training=is_training,
        input_ids=input_ids,
        input_mask=input_mask,
        token_type_ids=segment_ids,
        speaker_ids=speaker_ids,
        use_one_hot_embeddings=use_one_hot_embeddings)

    positions_shape = modeling.get_shape_list(cls_positions, expected_rank=2)
    max_utr_num = positions_shape[-1]

    input_tensor = gather_indexes(model.get_sequence_output(), cls_positions)  # [batch_size*max_utr_num, dim]
    input_shape = modeling.get_shape_list(input_tensor, expected_rank=2)
    width = input_shape[-1]

    with tf.compat.v1.variable_scope("cls/addressee_recognize"):
        # We apply one more non-linear transformation before the output layer.
        with tf.compat.v1.variable_scope("transform"):
            input_tensor = tf.compat.v1.layers.dense(
                input_tensor,
                units=bert_config.hidden_size,
                activation=modeling.get_activation(bert_config.hidden_act),
                kernel_initializer=modeling.create_initializer(bert_config.initializer_range))
            input_tensor = modeling.layer_norm(input_tensor)  # [batch_size*max_utr_num, dim]

        input_tensor = tf.reshape(input_tensor, [-1, max_utr_num, width])  # [batch_size, max_utr_num, dim]
        output_weights = tf.compat.v1.get_variable(
            "output_weights",
            shape=[width, width],
            initializer=modeling.create_initializer(bert_config.initializer_range))
        logits = tf.matmul(tf.einsum('aij,jk->aik', input_tensor, output_weights),
                           input_tensor, transpose_b=True)  # [batch_size, max_utr_num, max_utr_num]

        # mask = [[0. 0. 0. 0. 0.]
        #         [1. 0. 0. 0. 0.]
        #         [1. 1. 0. 0. 0.]
        #         [1. 1. 1. 0. 0.]
        #         [1. 1. 1. 1. 0.]]
        mask = tf.compat.v1.matrix_band_part(tf.ones((max_utr_num, max_utr_num)), -1, 0) - tf.compat.v1.matrix_band_part(
            tf.ones((max_utr_num, max_utr_num)), 0, 0)
        logits = logits * mask + -1e9 * (1 - mask)  # [batch_size, max_utr_num, max_utr_num]
        log_probs = tf.nn.log_softmax(logits, axis=-1)  # [batch_size, max_utr_num, max_utr_num]
        one_hot_labels = tf.reshape(labels, [-1, max_utr_num, max_utr_num])
        one_hot_labels = tf.cast(one_hot_labels, "float")

        # loss
        per_example_loss = - tf.reduce_sum(log_probs * one_hot_labels, axis=[-1])  # [batch_size, max_utr_num]
        numerator = tf.reduce_sum(label_weights * per_example_loss)  # [1, ]
        denominator = tf.reduce_sum(label_weights) + 1e-5  # [1, ]
        mean_loss = numerator / denominator

        # accuracy
        predictions = tf.argmax(log_probs, axis=-1, output_type=tf.int32)  # [batch_size, max_utr_num]
        predictions_one_hot = tf.one_hot(predictions, depth=max_utr_num,
                                         dtype=tf.float32)  # [batch_size, max_utr_num, max_utr_num]
        correct_prediction_utr = tf.reduce_sum(predictions_one_hot * one_hot_labels, -1)  # [batch_size, max_utr_num]
        numerator_utr = tf.reduce_sum(label_weights * correct_prediction_utr)  # [1, ]
        accuracy_utr = numerator_utr / denominator

        correct_prediction_sess = tf.equal(tf.reduce_sum(label_weights * correct_prediction_utr, -1),
                                           tf.reduce_sum(label_weights, -1))  # [batch_size, ]
        accuracy_sess = tf.reduce_mean(tf.cast(correct_prediction_sess, "float"), name="accuracy")

        return mean_loss, logits, log_probs, accuracy_utr, accuracy_sess, tf.reduce_sum(label_weights)


def run_epoch(epoch, sess, training, logits, accuracy_utr, accuracy_sess, mean_loss, train_opt):
    step = 0
    t0 = time()

    try:
        while True:
            step += 1
            batch_logits, batch_loss, _, batch_accuracy_utr, batch_accuracy_sess = sess.run(
                [logits, mean_loss, train_opt, accuracy_utr, accuracy_sess], feed_dict={training: True})

            if step % 1000 == 0:
                tf.compat.v1.logging.info(
                    "Epoch: %i, Step: %d, Time (min): %.2f, Loss: %.4f, Accuracy_utr: %.2f, Accuracy_sess: %.2f" %
                    (
                        epoch, step, (time() - t0) / 60.0, batch_loss, 100 * batch_accuracy_utr,
                        100 * batch_accuracy_sess))

    except tf.errors.OutOfRangeError:
        tf.compat.v1.logging.info("Epoch: %i, Step: %d, Time (min): %.2f, Loss: %.4f, Accuracy_utr: %.2f, "
                                  "Accuracy_sess: %.2f" %
                        (epoch, step, (time() - t0) / 60.0, batch_loss, 100 * batch_accuracy_utr,
                         100 * batch_accuracy_sess))
        pass


best_score = 0.0


def run_test(epoch, op_name, sess, training, prob, accuracy_utr, accuracy_sess, num_utr, saver, dir_path):
    step = 0
    t0 = time()
    num_test_utr = 0
    num_correct_utr = 0.0
    test_accuracy_utr = 0
    num_test_sess = 0
    num_correct_sess = 0.0
    test_accuracy_sess = 0

    try:
        while True:
            step += 1
            batch_accuracy_utr, batch_accuracy_sess, predicted_prob, batch_num_utr = sess.run(
                [accuracy_utr, accuracy_sess, prob, num_utr], feed_dict={training: False})

            num_test_utr += int(batch_num_utr)
            num_correct_utr += batch_num_utr * batch_accuracy_utr

            num_test_sess += len(predicted_prob)
            num_correct_sess += len(predicted_prob) * batch_accuracy_sess

            if step % 100 == 0:
                tf.compat.v1.logging.info("Epoch: %i, Step: %d, Time (min): %.2f" % (epoch, step, (time() - t0) / 60.0))

    except tf.errors.OutOfRangeError:
        test_accuracy_utr = num_correct_utr / num_test_utr
        print('num_test_utterance: {}, test_accuracy_utr: {}'.format(num_test_utr, test_accuracy_utr))
        test_accuracy_sess = num_correct_sess / num_test_sess
        print('num_test_session: {}, test_accuracy_sess: {}'.format(num_test_sess, test_accuracy_sess))

        global best_score
        if op_name == 'valid' and test_accuracy_sess > best_score:
            best_score = test_accuracy_sess
            dir_path = os.path.join(dir_path, "epoch_{}".format(epoch))
            saver.save(sess, dir_path)
            tf.compat.v1.logging.info(">> Save model!")

    return test_accuracy_sess


def main(_):
    tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.INFO)
    print_configuration_op(FLAGS)

    bert_config = modeling.BertConfig.from_json_file(FLAGS.bert_config_file)

    root_path = FLAGS.output_dir
    if not os.path.exists(root_path):
        os.makedirs(root_path)
    timestamp = str(int(time()))
    root_path = os.path.join(root_path, timestamp)
    tf.compat.v1.logging.info('root_path: {}'.format(root_path))
    if not os.path.exists(root_path):
        os.makedirs(root_path)

    train_data_size = count_data_size(FLAGS.train_dir)
    tf.compat.v1.logging.info('train data size: {}'.format(train_data_size))
    valid_data_size = count_data_size(FLAGS.valid_dir)
    tf.compat.v1.logging.info('valid data size: {}'.format(valid_data_size))

    num_train_steps = train_data_size // FLAGS.train_batch_size * FLAGS.num_train_epochs
    num_warmup_steps = int(num_train_steps * FLAGS.warmup_proportion)

    filenames = tf.compat.v1.placeholder(tf.string, shape=[None])
    shuffle_size = tf.compat.v1.placeholder(tf.int64)
    dataset = tf.data.TFRecordDataset(filenames)
    dataset = dataset.map(parse_exmp)  # Parse the record into tensors.
    dataset = dataset.repeat(1)
    # buffer_size 100
    dataset = dataset.shuffle(shuffle_size)
    dataset = dataset.batch(FLAGS.train_batch_size)
    iterator = dataset.make_initializable_iterator()
    input_sents, input_mask, segment_ids, speaker_ids, cls_positions, labels, label_weights = iterator.get_next()

    training = tf.compat.v1.placeholder(tf.bool)
    mean_loss, logits, log_probs, accuracy_utr, accuracy_sess, num_utr = create_model(bert_config=bert_config,
                                                                                      is_training=training,
                                                                                      input_ids=input_sents,
                                                                                      input_mask=input_mask,
                                                                                      segment_ids=segment_ids,
                                                                                      speaker_ids=speaker_ids,
                                                                                      cls_positions=cls_positions,
                                                                                      labels=labels,
                                                                                      label_weights=label_weights,
                                                                                      use_one_hot_embeddings=False)

    # init model with pre-training
    tvars = tf.compat.v1.trainable_variables()
    if FLAGS.init_checkpoint:
        (assignment_map, initialized_variable_names) = modeling.get_assignment_map_from_checkpoint(tvars,
                                                                                                   FLAGS.init_checkpoint)
        tf.compat.v1.train.init_from_checkpoint(FLAGS.init_checkpoint, assignment_map)

    tf.compat.v1.logging.info("**** Trainable Variables ****")
    for var in tvars:
        init_string = ""
        if var.name in initialized_variable_names:
            init_string = ", *INIT_FROM_CKPT*"
        tf.compat.v1.logging.info("  name = %s, shape = %s%s", var.name, var.shape, init_string)

    train_opt = optimization.create_optimizer(mean_loss, FLAGS.learning_rate, num_train_steps, num_warmup_steps, False)

    config = tf.compat.v1.ConfigProto(allow_soft_placement=True)
    config.gpu_options.allow_growth = True
    saver = tf.compat.v1.train.Saver()

    if FLAGS.do_train:
        with tf.compat.v1.Session(config=config) as sess:
            sess.run(tf.compat.v1.global_variables_initializer())

            for epoch in range(FLAGS.num_train_epochs):
                tf.compat.v1.logging.info('Train begin epoch {}'.format(epoch))
                sess.run(iterator.initializer,
                         feed_dict={filenames: [FLAGS.train_dir], shuffle_size: 1024})
                run_epoch(epoch, sess, training, logits, accuracy_utr, accuracy_sess, mean_loss, train_opt)

                tf.compat.v1.logging.info('Valid begin')
                sess.run(iterator.initializer,
                         feed_dict={filenames: [FLAGS.valid_dir], shuffle_size: 1})
                run_test(epoch, "valid", sess, training, log_probs, accuracy_utr, accuracy_sess, num_utr, saver,
                         root_path)


if __name__ == "__main__":
    tf.compat.v1.app.run()