def read_arguments():
args_ = argparse.ArgumentParser(description='Sovling GraphParser')
args_.add_argument('--dataset',
choices=['ontonotes', 'ud'],
help='Dataset',
required=True)
args_.add_argument('--domain', help='domain/language', required=True)
args_.add_argument('--rnn_mode',
choices=['RNN', 'LSTM', 'GRU'],
help='architecture of rnn',
required=True)
args_.add_argument('--gating',
action='store_true',
help='use gated mechanism')
args_.add_argument('--num_gates',
type=int,
default=0,
help='number of gates for gating mechanism')
args_.add_argument('--num_epochs',
type=int,
default=200,
help='Number of training epochs')
args_.add_argument('--batch_size',
type=int,
default=64,
help='Number of sentences in each batch')
args_.add_argument('--hidden_size',
type=int,
default=256,
help='Number of hidden units in RNN')
args_.add_argument('--arc_space',
type=int,
default=128,
help='Dimension of tag space')
args_.add_argument('--arc_tag_space',
type=int,
default=128,
help='Dimension of tag space')
args_.add_argument('--num_layers',
type=int,
default=1,
help='Number of layers of RNN')
args_.add_argument('--num_filters',
type=int,
default=50,
help='Number of filters in CNN')
args_.add_argument('--kernel_size',
type=int,
default=3,
help='Size of Kernel for CNN')
args_.add_argument('--use_pos',
action='store_true',
help='use part-of-speech embedding.')
args_.add_argument('--use_char',
action='store_true',
help='use character embedding and CNN.')
args_.add_argument('--word_dim',
type=int,
default=300,
help='Dimension of word embeddings')
args_.add_argument('--pos_dim',
type=int,
default=50,
help='Dimension of POS embeddings')
args_.add_argument('--char_dim',
type=int,
default=50,
help='Dimension of Character embeddings')
args_.add_argument('--initializer',
choices=['xavier'],
help='initialize model parameters')
args_.add_argument('--opt',
choices=['adam', 'sgd'],
help='optimization algorithm')
args_.add_argument('--momentum',
type=float,
default=0.9,
help='momentum of optimizer')
args_.add_argument('--betas',
nargs=2,
type=float,
default=[0.9, 0.9],
help='betas of optimizer')
args_.add_argument('--learning_rate',
type=float,
default=0.01,
help='Learning rate')
args_.add_argument('--decay_rate',
type=float,
default=0.05,
help='Decay rate of learning rate')
args_.add_argument('--schedule',
type=int,
help='schedule for learning rate decay')
args_.add_argument('--clip',
type=float,
default=5.0,
help='gradient clipping')
args_.add_argument('--gamma',
type=float,
default=0.0,
help='weight for regularization')
args_.add_argument('--epsilon',
type=float,
default=1e-8,
help='epsilon for adam')
args_.add_argument('--p_rnn',
nargs=2,
type=float,
required=True,
help='dropout rate for RNN')
args_.add_argument('--p_in',
type=float,
default=0.33,
help='dropout rate for input embeddings')
args_.add_argument('--p_out',
type=float,
default=0.33,
help='dropout rate for output layer')
args_.add_argument('--arc_decode',
choices=['mst', 'greedy'],
help='arc decoding algorithm',
required=True)
args_.add_argument('--unk_replace',
type=float,
default=0.,
help='The rate to replace a singleton word with UNK')
args_.add_argument('--punct_set',
nargs='+',
type=str,
help='List of punctuations')
args_.add_argument('--word_embedding',
choices=['random', 'glove', 'fasttext', 'word2vec'],
help='Embedding for words')
args_.add_argument(
'--word_path',
help=
'path for word embedding dict - in case word_embedding is not random')
args_.add_argument('--freeze_word_embeddings',
action='store_true',
help='frozen the word embedding (disable fine-tuning).')
args_.add_argument('--freeze_sequence_taggers',
action='store_true',
help='frozen the BiLSTMs of the pre-trained taggers.')
args_.add_argument('--char_embedding',
choices=['random'],
help='Embedding for characters',
required=True)
args_.add_argument('--char_path', help='path for character embedding dict')
args_.add_argument(
'--set_num_training_samples',
type=int,
help='downsampling training set to a fixed number of samples')
args_.add_argument('--model_path',
help='path for saving model file.',
required=True)
args_.add_argument('--load_path',
help='path for loading saved source model file.',
default=None)
args_.add_argument(
'--load_sequence_taggers_paths',
nargs='+',
help='path for loading saved sequence_tagger saved_models files.',
default=None)
args_.add_argument('--strict',
action='store_true',
help='if True loaded model state should contin '
'exactly the same keys as current model')
args_.add_argument('--eval_mode',
action='store_true',
help='evaluating model without training it')
args = args_.parse_args()
args_dict = {}
args_dict['dataset'] = args.dataset
args_dict['domain'] = args.domain
args_dict['rnn_mode'] = args.rnn_mode
args_dict['gating'] = args.gating
args_dict['num_gates'] = args.num_gates
args_dict['arc_decode'] = args.arc_decode
args_dict['splits'] = ['train', 'dev', 'test']
args_dict['model_path'] = args.model_path
if not path.exists(args_dict['model_path']):
makedirs(args_dict['model_path'])
args_dict['data_paths'] = {}
if args_dict['dataset'] == 'ontonotes':
data_path = 'data/onto_pos_ner_dp'
else:
data_path = 'data/ud_pos_ner_dp'
for split in args_dict['splits']:
args_dict['data_paths'][
split] = data_path + '_' + split + '_' + args_dict['domain']
args_dict['alphabet_data_paths'] = {}
for split in args_dict['splits']:
if args_dict['dataset'] == 'ontonotes':
args_dict['alphabet_data_paths'][
split] = data_path + '_' + split + '_' + 'all'
else:
if '_' in args_dict['domain']:
args_dict['alphabet_data_paths'][
split] = data_path + '_' + split + '_' + args_dict[
'domain'].split('_')[0]
else:
args_dict['alphabet_data_paths'][split] = args_dict[
'data_paths'][split]
args_dict['model_name'] = 'domain_' + args_dict['domain']
args_dict['full_model_name'] = path.join(args_dict['model_path'],
args_dict['model_name'])
args_dict['load_path'] = args.load_path
args_dict['load_sequence_taggers_paths'] = args.load_sequence_taggers_paths
if args_dict['load_sequence_taggers_paths'] is not None:
args_dict['gating'] = True
args_dict['num_gates'] = len(
args_dict['load_sequence_taggers_paths']) + 1
else:
if not args_dict['gating']:
args_dict['num_gates'] = 0
args_dict['strict'] = args.strict
args_dict['num_epochs'] = args.num_epochs
args_dict['batch_size'] = args.batch_size
args_dict['hidden_size'] = args.hidden_size
args_dict['arc_space'] = args.arc_space
args_dict['arc_tag_space'] = args.arc_tag_space
args_dict['num_layers'] = args.num_layers
args_dict['num_filters'] = args.num_filters
args_dict['kernel_size'] = args.kernel_size
args_dict['learning_rate'] = args.learning_rate
args_dict[
'initializer'] = nn.init.xavier_uniform_ if args.initializer == 'xavier' else None
args_dict['opt'] = args.opt
args_dict['momentum'] = args.momentum
args_dict['betas'] = tuple(args.betas)
args_dict['epsilon'] = args.epsilon
args_dict['decay_rate'] = args.decay_rate
args_dict['clip'] = args.clip
args_dict['gamma'] = args.gamma
args_dict['schedule'] = args.schedule
args_dict['p_rnn'] = tuple(args.p_rnn)
args_dict['p_in'] = args.p_in
args_dict['p_out'] = args.p_out
args_dict['unk_replace'] = args.unk_replace
args_dict['set_num_training_samples'] = args.set_num_training_samples
args_dict['punct_set'] = None
if args.punct_set is not None:
args_dict['punct_set'] = set(args.punct_set)
logger.info(
"punctuations(%d): %s" %
(len(args_dict['punct_set']), ' '.join(args_dict['punct_set'])))
args_dict['freeze_word_embeddings'] = args.freeze_word_embeddings
args_dict['freeze_sequence_taggers'] = args.freeze_sequence_taggers
args_dict['word_embedding'] = args.word_embedding
args_dict['word_path'] = args.word_path
args_dict['use_char'] = args.use_char
args_dict['char_embedding'] = args.char_embedding
args_dict['char_path'] = args.char_path
args_dict['use_pos'] = args.use_pos
args_dict['pos_dim'] = args.pos_dim
args_dict['word_dict'] = None
args_dict['word_dim'] = args.word_dim
if args_dict['word_embedding'] != 'random' and args_dict['word_path']:
args_dict['word_dict'], args_dict[
'word_dim'] = load_word_embeddings.load_embedding_dict(
args_dict['word_embedding'], args_dict['word_path'])
args_dict['char_dict'] = None
args_dict['char_dim'] = args.char_dim
if args_dict['char_embedding'] != 'random':
args_dict['char_dict'], args_dict[
'char_dim'] = load_word_embeddings.load_embedding_dict(
args_dict['char_embedding'], args_dict['char_path'])
args_dict['pos_dict'] = None
args_dict['alphabet_path'] = path.join(
args_dict['model_path'],
'alphabets' + '_src_domain_' + args_dict['domain'] + '/')
args_dict['model_name'] = path.join(args_dict['model_path'],
args_dict['model_name'])
args_dict['eval_mode'] = args.eval_mode
args_dict['device'] = torch.device(
'cuda' if torch.cuda.is_available() else 'cpu')
args_dict[
'word_status'] = 'frozen' if args.freeze_word_embeddings else 'fine tune'
args_dict['char_status'] = 'enabled' if args.use_char else 'disabled'
args_dict['pos_status'] = 'enabled' if args.use_pos else 'disabled'
logger.info("Saving arguments to file")
save_args(args, args_dict['full_model_name'])
logger.info("Creating Alphabets")
alphabet_dict = creating_alphabets(args_dict['alphabet_path'],
args_dict['alphabet_data_paths'],
args_dict['word_dict'])
args_dict = {**args_dict, **alphabet_dict}
ARGS = namedtuple('ARGS', args_dict.keys())
my_args = ARGS(**args_dict)
return my_args
def read_arguments():
args_ = argparse.ArgumentParser(description='Sovling SequenceTagger')
args_.add_argument('--dataset',
choices=['ontonotes', 'ud'],
help='Dataset',
required=True)
args_.add_argument('--domain', help='domain', required=True)
args_.add_argument('--rnn_mode',
choices=['RNN', 'LSTM', 'GRU'],
help='architecture of rnn',
required=True)
args_.add_argument('--task', default='distance_from_the_root', choices=['distance_from_the_root', 'number_of_children',\
'relative_pos_based', 'language_model','add_label','add_head_coarse_pos','Multitask_POS_predict','Multitask_case_predict',\
'Multitask_label_predict','Multitask_coarse_predict','MRL_Person','MRL_Gender','MRL_case','MRL_POS','MRL_no','MRL_label',\
'predict_coarse_of_modifier','predict_ma_tag_of_modifier','add_head_ma','predict_case_of_modifier'], help='sequence_tagger task')
args_.add_argument('--num_epochs',
type=int,
default=200,
help='Number of training epochs')
args_.add_argument('--batch_size',
type=int,
default=64,
help='Number of sentences in each batch')
args_.add_argument('--hidden_size',
type=int,
default=256,
help='Number of hidden units in RNN')
args_.add_argument('--tag_space',
type=int,
default=128,
help='Dimension of tag space')
args_.add_argument('--num_layers',
type=int,
default=1,
help='Number of layers of RNN')
args_.add_argument('--num_filters',
type=int,
default=50,
help='Number of filters in CNN')
args_.add_argument('--kernel_size',
type=int,
default=3,
help='Size of Kernel for CNN')
args_.add_argument('--use_pos',
action='store_true',
help='use part-of-speech embedding.')
args_.add_argument('--use_char',
action='store_true',
help='use character embedding and CNN.')
args_.add_argument('--word_dim',
type=int,
default=300,
help='Dimension of word embeddings')
args_.add_argument('--pos_dim',
type=int,
default=50,
help='Dimension of POS embeddings')
args_.add_argument('--char_dim',
type=int,
default=50,
help='Dimension of Character embeddings')
args_.add_argument('--initializer',
choices=['xavier'],
help='initialize model parameters')
args_.add_argument('--opt',
choices=['adam', 'sgd'],
help='optimization algorithm')
args_.add_argument('--momentum',
type=float,
default=0.9,
help='momentum of optimizer')
args_.add_argument('--betas',
nargs=2,
type=float,
default=[0.9, 0.9],
help='betas of optimizer')
args_.add_argument('--learning_rate',
type=float,
default=0.01,
help='Learning rate')
args_.add_argument('--decay_rate',
type=float,
default=0.05,
help='Decay rate of learning rate')
args_.add_argument('--schedule',
type=int,
help='schedule for learning rate decay')
args_.add_argument('--clip',
type=float,
default=5.0,
help='gradient clipping')
args_.add_argument('--gamma',
type=float,
default=0.0,
help='weight for regularization')
args_.add_argument('--epsilon',
type=float,
default=1e-8,
help='epsilon for adam')
args_.add_argument('--p_rnn',
nargs=2,
type=float,
required=True,
help='dropout rate for RNN')
args_.add_argument('--p_in',
type=float,
default=0.33,
help='dropout rate for input embeddings')
args_.add_argument('--p_out',
type=float,
default=0.33,
help='dropout rate for output layer')
args_.add_argument('--unk_replace',
type=float,
default=0.,
help='The rate to replace a singleton word with UNK')
args_.add_argument('--punct_set',
nargs='+',
type=str,
help='List of punctuations')
args_.add_argument('--word_embedding',
choices=['random', 'glove', 'fasttext', 'word2vec'],
help='Embedding for words')
args_.add_argument(
'--word_path',
help=
'path for word embedding dict - in case word_embedding is not random')
args_.add_argument('--freeze_word_embeddings',
action='store_true',
help='frozen the word embedding (disable fine-tuning).')
args_.add_argument('--char_embedding',
choices=['random', 'hellwig'],
help='Embedding for characters',
required=True)
args_.add_argument('--pos_embedding',
choices=['random', 'one_hot'],
help='Embedding for pos',
required=True)
args_.add_argument('--char_path', help='path for character embedding dict')
args_.add_argument('--pos_path', help='path for pos embedding dict')
args_.add_argument('--use_unlabeled_data',
action='store_true',
help='flag to use unlabeled data.')
args_.add_argument('--use_labeled_data',
action='store_true',
help='flag to use labeled data.')
args_.add_argument('--model_path',
help='path for saving model file.',
required=True)
args_.add_argument('--parser_path',
help='path for loading parser files.',
default=None)
args_.add_argument('--load_path',
help='path for loading saved source model file.',
default=None)
args_.add_argument('--strict',
action='store_true',
help='if True loaded model state should contain '
'exactly the same keys as current model')
args_.add_argument('--eval_mode',
action='store_true',
help='evaluating model without training it')
args = args_.parse_args()
args_dict = {}
args_dict['dataset'] = args.dataset
args_dict['domain'] = args.domain
args_dict['task'] = args.task
args_dict['rnn_mode'] = args.rnn_mode
args_dict['load_path'] = args.load_path
args_dict['strict'] = args.strict
args_dict['model_path'] = args.model_path
if not path.exists(args_dict['model_path']):
makedirs(args_dict['model_path'])
args_dict['parser_path'] = args.parser_path
args_dict['model_name'] = 'domain_' + args_dict['domain']
args_dict['full_model_name'] = path.join(args_dict['model_path'],
args_dict['model_name'])
args_dict['use_unlabeled_data'] = args.use_unlabeled_data
args_dict['use_labeled_data'] = args.use_labeled_data
print(args_dict['parser_path'])
if args_dict['task'] == 'number_of_children':
args_dict['data_paths'] = write_extra_labels.add_number_of_children(
args_dict['model_path'],
args_dict['parser_path'],
args_dict['domain'],
args_dict['domain'],
use_unlabeled_data=args_dict['use_unlabeled_data'],
use_labeled_data=args_dict['use_labeled_data'])
elif args_dict['task'] == 'distance_from_the_root':
args_dict[
'data_paths'] = write_extra_labels.add_distance_from_the_root(
args_dict['model_path'],
args_dict['parser_path'],
args_dict['domain'],
args_dict['domain'],
use_unlabeled_data=args_dict['use_unlabeled_data'],
use_labeled_data=args_dict['use_labeled_data'])
elif args_dict['task'] == 'Multitask_label_predict':
args_dict['data_paths'] = write_extra_labels.Multitask_label_predict(
args_dict['model_path'],
args_dict['parser_path'],
args_dict['domain'],
args_dict['domain'],
use_unlabeled_data=args_dict['use_unlabeled_data'],
use_labeled_data=args_dict['use_labeled_data'])
elif args_dict['task'] == 'Multitask_coarse_predict':
args_dict['data_paths'] = write_extra_labels.Multitask_coarse_predict(
args_dict['model_path'],
args_dict['parser_path'],
args_dict['domain'],
args_dict['domain'],
use_unlabeled_data=args_dict['use_unlabeled_data'],
use_labeled_data=args_dict['use_labeled_data'])
elif args_dict['task'] == 'Multitask_POS_predict':
args_dict['data_paths'] = write_extra_labels.Multitask_POS_predict(
args_dict['model_path'],
args_dict['parser_path'],
args_dict['domain'],
args_dict['domain'],
use_unlabeled_data=args_dict['use_unlabeled_data'],
use_labeled_data=args_dict['use_labeled_data'])
elif args_dict['task'] == 'relative_pos_based':
args_dict['data_paths'] = write_extra_labels.add_relative_pos_based(
args_dict['model_path'],
args_dict['parser_path'],
args_dict['domain'],
args_dict['domain'],
use_unlabeled_data=args_dict['use_unlabeled_data'],
use_labeled_data=args_dict['use_labeled_data'])
elif args_dict['task'] == 'add_label':
args_dict['data_paths'] = write_extra_labels.add_label(
args_dict['model_path'],
args_dict['parser_path'],
args_dict['domain'],
args_dict['domain'],
use_unlabeled_data=args_dict['use_unlabeled_data'],
use_labeled_data=args_dict['use_labeled_data'])
elif args_dict['task'] == 'add_relative_TAG':
args_dict['data_paths'] = write_extra_labels.add_relative_TAG(
args_dict['model_path'],
args_dict['parser_path'],
args_dict['domain'],
args_dict['domain'],
use_unlabeled_data=args_dict['use_unlabeled_data'],
use_labeled_data=args_dict['use_labeled_data'])
elif args_dict['task'] == 'add_head_coarse_pos':
args_dict['data_paths'] = write_extra_labels.add_head_coarse_pos(
args_dict['model_path'],
args_dict['parser_path'],
args_dict['domain'],
args_dict['domain'],
use_unlabeled_data=args_dict['use_unlabeled_data'],
use_labeled_data=args_dict['use_labeled_data'])
elif args_dict['task'] == 'predict_ma_tag_of_modifier':
args_dict[
'data_paths'] = write_extra_labels.predict_ma_tag_of_modifier(
args_dict['model_path'],
args_dict['parser_path'],
args_dict['domain'],
args_dict['domain'],
use_unlabeled_data=args_dict['use_unlabeled_data'],
use_labeled_data=args_dict['use_labeled_data'])
elif args_dict['task'] == 'Multitask_case_predict':
args_dict['data_paths'] = write_extra_labels.Multitask_case_predict(
args_dict['model_path'],
args_dict['parser_path'],
args_dict['domain'],
args_dict['domain'],
use_unlabeled_data=args_dict['use_unlabeled_data'],
use_labeled_data=args_dict['use_labeled_data'])
elif args_dict['task'] == 'predict_coarse_of_modifier':
args_dict[
'data_paths'] = write_extra_labels.predict_coarse_of_modifier(
args_dict['model_path'],
args_dict['parser_path'],
args_dict['domain'],
args_dict['domain'],
use_unlabeled_data=args_dict['use_unlabeled_data'],
use_labeled_data=args_dict['use_labeled_data'])
elif args_dict['task'] == 'predict_case_of_modifier':
args_dict['data_paths'] = write_extra_labels.predict_case_of_modifier(
args_dict['model_path'],
args_dict['parser_path'],
args_dict['domain'],
args_dict['domain'],
use_unlabeled_data=args_dict['use_unlabeled_data'],
use_labeled_data=args_dict['use_labeled_data'])
elif args_dict['task'] == 'add_head_ma':
args_dict['data_paths'] = write_extra_labels.add_head_ma(
args_dict['model_path'],
args_dict['parser_path'],
args_dict['domain'],
args_dict['domain'],
use_unlabeled_data=args_dict['use_unlabeled_data'],
use_labeled_data=args_dict['use_labeled_data'])
elif args_dict['task'] == 'MRL_case':
args_dict['data_paths'] = write_extra_labels.MRL_case(
args_dict['model_path'],
args_dict['parser_path'],
args_dict['domain'],
args_dict['domain'],
use_unlabeled_data=args_dict['use_unlabeled_data'],
use_labeled_data=args_dict['use_labeled_data'])
elif args_dict['task'] == 'MRL_POS':
args_dict['data_paths'] = write_extra_labels.MRL_POS(
args_dict['model_path'],
args_dict['parser_path'],
args_dict['domain'],
args_dict['domain'],
use_unlabeled_data=args_dict['use_unlabeled_data'],
use_labeled_data=args_dict['use_labeled_data'])
elif args_dict['task'] == 'MRL_no':
args_dict['data_paths'] = write_extra_labels.MRL_no(
args_dict['model_path'],
args_dict['parser_path'],
args_dict['domain'],
args_dict['domain'],
use_unlabeled_data=args_dict['use_unlabeled_data'],
use_labeled_data=args_dict['use_labeled_data'])
elif args_dict['task'] == 'MRL_label':
args_dict['data_paths'] = write_extra_labels.MRL_label(
args_dict['model_path'],
args_dict['parser_path'],
args_dict['domain'],
args_dict['domain'],
use_unlabeled_data=args_dict['use_unlabeled_data'],
use_labeled_data=args_dict['use_labeled_data'])
elif args_dict['task'] == 'MRL_Person':
args_dict['data_paths'] = write_extra_labels.MRL_Person(
args_dict['model_path'],
args_dict['parser_path'],
args_dict['domain'],
args_dict['domain'],
use_unlabeled_data=args_dict['use_unlabeled_data'],
use_labeled_data=args_dict['use_labeled_data'])
elif args_dict['task'] == 'MRL_Gender':
args_dict['data_paths'] = write_extra_labels.MRL_Gender(
args_dict['model_path'],
args_dict['parser_path'],
args_dict['domain'],
args_dict['domain'],
use_unlabeled_data=args_dict['use_unlabeled_data'],
use_labeled_data=args_dict['use_labeled_data'])
else: #args_dict['task'] == 'language_model':
args_dict['data_paths'] = write_extra_labels.add_language_model(
args_dict['model_path'],
args_dict['parser_path'],
args_dict['domain'],
args_dict['domain'],
use_unlabeled_data=args_dict['use_unlabeled_data'],
use_labeled_data=args_dict['use_labeled_data'])
args_dict['splits'] = args_dict['data_paths'].keys()
alphabet_data_paths = deepcopy(args_dict['data_paths'])
if args_dict['dataset'] == 'ontonotes':
data_path = 'data/onto_pos_ner_dp'
else:
data_path = 'data/ud_pos_ner_dp'
# Adding more resources to make sure equal alphabet size for all domains
for split in args_dict['splits']:
if args_dict['dataset'] == 'ontonotes':
alphabet_data_paths['additional_' +
split] = data_path + '_' + split + '_' + 'all'
else:
if '_' in args_dict['domain']:
alphabet_data_paths[
split] = data_path + '_' + split + '_' + args_dict[
'domain'].split('_')[0]
else:
alphabet_data_paths[split] = args_dict['data_paths'][split]
args_dict['alphabet_data_paths'] = alphabet_data_paths
args_dict['num_epochs'] = args.num_epochs
args_dict['batch_size'] = args.batch_size
args_dict['hidden_size'] = args.hidden_size
args_dict['tag_space'] = args.tag_space
args_dict['num_layers'] = args.num_layers
args_dict['num_filters'] = args.num_filters
args_dict['kernel_size'] = args.kernel_size
args_dict['learning_rate'] = args.learning_rate
args_dict[
'initializer'] = nn.init.xavier_uniform_ if args.initializer == 'xavier' else None
args_dict['opt'] = args.opt
args_dict['momentum'] = args.momentum
args_dict['betas'] = tuple(args.betas)
args_dict['epsilon'] = args.epsilon
args_dict['decay_rate'] = args.decay_rate
args_dict['clip'] = args.clip
args_dict['gamma'] = args.gamma
args_dict['schedule'] = args.schedule
args_dict['p_rnn'] = tuple(args.p_rnn)
args_dict['p_in'] = args.p_in
args_dict['p_out'] = args.p_out
args_dict['unk_replace'] = args.unk_replace
args_dict['punct_set'] = None
if args.punct_set is not None:
args_dict['punct_set'] = set(args.punct_set)
logger.info(
"punctuations(%d): %s" %
(len(args_dict['punct_set']), ' '.join(args_dict['punct_set'])))
args_dict['freeze_word_embeddings'] = args.freeze_word_embeddings
args_dict['word_embedding'] = args.word_embedding
args_dict['word_path'] = args.word_path
args_dict['use_char'] = args.use_char
args_dict['char_embedding'] = args.char_embedding
args_dict['pos_embedding'] = args.pos_embedding
args_dict['char_path'] = args.char_path
args_dict['pos_path'] = args.pos_path
args_dict['use_pos'] = args.use_pos
args_dict['pos_dim'] = args.pos_dim
args_dict['word_dict'] = None
args_dict['word_dim'] = args.word_dim
if args_dict['word_embedding'] != 'random' and args_dict['word_path']:
args_dict['word_dict'], args_dict[
'word_dim'] = load_word_embeddings.load_embedding_dict(
args_dict['word_embedding'], args_dict['word_path'])
args_dict['char_dict'] = None
args_dict['char_dim'] = args.char_dim
if args_dict['char_embedding'] != 'random':
args_dict['char_dict'], args_dict[
'char_dim'] = load_word_embeddings.load_embedding_dict(
args_dict['char_embedding'], args_dict['char_path'])
args_dict['pos_dict'] = None
if args_dict['pos_embedding'] != 'random':
args_dict['pos_dict'], args_dict[
'pos_dim'] = load_word_embeddings.load_embedding_dict(
args_dict['pos_embedding'], args_dict['pos_path'])
args_dict['alphabet_path'] = path.join(
args_dict['model_path'],
'alphabets' + '_src_domain_' + args_dict['domain'] + '/')
args_dict['alphabet_parser_path'] = path.join(
args_dict['parser_path'],
'alphabets' + '_src_domain_' + args_dict['domain'] + '/')
args_dict['model_name'] = path.join(args_dict['model_path'],
args_dict['model_name'])
args_dict['eval_mode'] = args.eval_mode
args_dict['device'] = torch.device(
'cuda' if torch.cuda.is_available() else 'cpu')
args_dict[
'word_status'] = 'frozen' if args.freeze_word_embeddings else 'fine tune'
args_dict['char_status'] = 'enabled' if args.use_char else 'disabled'
args_dict['pos_status'] = 'enabled' if args.use_pos else 'disabled'
logger.info("Saving arguments to file")
save_args(args, args_dict['full_model_name'])
logger.info("Creating Alphabets")
alphabet_dict = creating_alphabets(args_dict['alphabet_path'],
args_dict['alphabet_parser_path'],
args_dict['alphabet_data_paths'])
args_dict = {**args_dict, **alphabet_dict}
ARGS = namedtuple('ARGS', args_dict.keys())
my_args = ARGS(**args_dict)
return my_args