train_es.py

import utils
import tf_utils
from build_data import build_data
import numpy as np
import tensorflow as tf
import sys
import os.path

'Train the model on the train set and evaluate on the evaluation and test sets until ' \
'(1) maximum epochs limit or (2) early stopping break'
def checkInputs():
    if (len(sys.argv) <= 3) or os.path.isfile(sys.argv[0])==False :
        raise ValueError(
            'The configuration file and the timestamp should be specified.')


if __name__ == "__main__":

    checkInputs()

    config=build_data(sys.argv[1])
    

    train_data = utils.HeadData(config.train_id_docs, np.arange(len(config.train_id_docs)))
    dev_data = utils.HeadData(config.dev_id_docs, np.arange(len(config.dev_id_docs)))
    test_data = utils.HeadData(config.test_id_docs, np.arange(len(config.test_id_docs)))

    tf.reset_default_graph()
    tf.set_random_seed(1)

    utils.printParameters(config)

    with tf.Session() as sess:
        embedding_matrix = tf.get_variable('embedding_matrix', shape=config.wordvectors.shape, dtype=tf.float32,
                                           trainable=False).assign(config.wordvectors)
        emb_mtx = sess.run(embedding_matrix)

        model = tf_utils.model(config,emb_mtx,sess)

        obj, m_op, predicted_op_ner, actual_op_ner, predicted_op_rel, actual_op_rel, score_op_rel = model.run()

        train_step = model.get_train_op(obj)

        operations=tf_utils.operations(train_step,obj, m_op, predicted_op_ner, actual_op_ner, predicted_op_rel, actual_op_rel, score_op_rel)


        sess.run(tf.global_variables_initializer())

        best_score=0
        nepoch_no_imprv = 0  # for early stopping

        for iter in range(config.nepochs+1):

            model.train(train_data,operations,iter)

            dev_score=model.evaluate(dev_data,operations,'dev')

            model.evaluate(test_data, operations,'test')

            if dev_score>=best_score:
                nepoch_no_imprv = 0
                best_score = dev_score

                print ("- Best dev score {} so far in {} epoch".format(dev_score,iter))

            else:
                nepoch_no_imprv += 1
                if nepoch_no_imprv >= config.nepoch_no_imprv:

                    print ("- early stopping {} epochs without " \
                                     "improvement".format(nepoch_no_imprv))

                    with open(sys.argv[3]+"/es_"+sys.argv[2]+".txt", "w+") as myfile:
                        myfile.write(str(iter))
                        myfile.close()

                    break