ass_train.py

from __future__ import (division, absolute_import,
                        print_function, unicode_literals)
import os
import sys
import argparse
import logging
from ass_theano import train

def main():
    parser = argparse.ArgumentParser(
        description='Abstractive sentence summariser'
    )
    #parser.add_argument('--log',
    #                    default='DEBUG',
    #                    choices=['DEBUG', 'WARNING', 'ERROR'],
    #                    help='log level for Python logging')
    parser.add_argument('--context-encoder',
                        default='baseline',
                        choices=['baseline', 'attention'],
                        help='context encoder name')
    parser.add_argument('--corpus', required=True,
                        help='directory of the corpus e.g data/wikipedia/')

    # optimiser
    parser.add_argument('--optimizer', 
                        default='adam',
                        choices=['adam', 'adadelta', 'rmsprop', 'sgd'],
                        help='optimizing algorithm')
    parser.add_argument('--learning-rate', type=float,
                        default=0.001,
                        help='learning rate for the optimizer')

    # model params
    parser.add_argument('--embed-full-text-by',
                        choices=['word', 'sentence'],
                        default='word',
                        help='embed full text by word or sentence')
    parser.add_argument('--seq-maxlen', type=int,
                        default=500,
                        help='max length of input full text')
    parser.add_argument('--summary-maxlen', type=int,
                        default=200,
                        help='max length of each summary')
    parser.add_argument('--summary-context-length', type=int,
                        default=5,
                        help='summary context length used for training')
    parser.add_argument('--internal-representation-dim', type=int,
                        default=2000,
                        help='internal representation dimension')
    parser.add_argument('--attention-weight-max-roll', type=int,
                        default=1,
                        help='max roll for the attention weight vector in attention encoder')

    # training params
    parser.add_argument('--l2-penalty-coeff', type=float,
                        default=0.00,
                        help='penalty coefficient to the L2-norms of the model params')
    parser.add_argument('--train-split', type=float,
                        default=0.75,
                        help='weight of training corpus in the entire corpus, the rest for validation')
    parser.add_argument('--epochs', type=int,
                        default=10000,
                        help='number of epochs for training')
    parser.add_argument('--minibatch-size', type=int,
                        default=20,
                        help='mini batch size')
    parser.add_argument('--seed', type=int,
                        default=None,
                        help='seed for the random stream')
    parser.add_argument('--dropout-rate', type=float,
                        default=None,
                        help='dropout rate in (0,1)')

    # model load/save
    parser.add_argument('--save-params',
                        default='ass_params.pkl',
                        help='file for saving params')
    parser.add_argument('--save-params-every', type=int,
                        default=5,
                        help='save params every <k> epochs')
    parser.add_argument('--validate-every', type=int,
                        default=5,
                        help='validate every <k> epochs')
    parser.add_argument('--print-every', type=int,
                        default=5,
                        help='print info every <k> batches')

    # summary generation on the validation set
    parser.add_argument('--generate-summary',
                        action='store_true',
                        default=False,
                        help='whether to generate summaries when validating')
    parser.add_argument('--summary-search-beam-size', type=int,
                        default=2,
                        help='beam size for the summary search')
    
    args = parser.parse_args()

    #logging.basicConfig(level=args.log.upper())
    assert args.learning_rate > 0
    assert args.seq_maxlen > 0
    assert args.summary_maxlen > 0
    assert args.summary_context_length > 0
    assert args.internal_representation_dim > 0
    assert args.attention_weight_max_roll >= 0
    assert args.l2_penalty_coeff >= 0
    assert (args.train_split > 0 and args.train_split <= 1)
    assert args.epochs >= 0
    assert args.minibatch_size > 0
    assert (args.seed is None or args.seed >= 0)
    assert (args.dropout_rate is None
            or (args.dropout_rate > 0 and args.dropout_rate < 1))
    assert args.save_params_every > 0
    assert args.validate_every > 0
    assert args.print_every > 0
    assert args.summary_search_beam_size > 0
   
    args_dict = vars(args)
    print('Args', args_dict)
    train(**args_dict)

    #train(
    #    model=args.model,
    #    corpus=args.corpus,
    #    optimizer=args.optimizer,
    #    learning_rate=args.learning_rate,
    #    embed_full_text_by=args.embed_full_text_by,
        
    #    summary_maxlen=args.summary_maxlen,

        
        
        
    #    summary_context_length=args.summary_context_length,
    #    l2_penalty_coeff=args.L2_penalty_coeff,
    #    minibatch_size=args.minibatch_size,
    #    epochs=args.epochs,
    #    train_split=args.train_split,
    #    seed=args.seed,
    #    dropout_rate=args.dropout_rate,
    #            internal_representation_dim=args.internal_representation_dim,
    #    attention_weight_max_roll=args.attention_weight_max_roll,
    #    load_params=args.load_params,
    #    save_params=args.save_params,
    #    save_params_every=args.save_params_every,
    #    validate_every=args.validate_every,
    #    generate_summary=args.generate_summary,
    #           summary_search_beam_size=args.summary_search_beam_size
    #)

if __name__ == '__main__':
    main()