help="Filter size of each layer, \
e.g. --fs 2 2 --fs 2 2 2 3 3 --fs 3")
parser.add_argument('--vis_embds',
type=int,
default=0,
choices=[0, 1],
help="Whether to visualize embeddings")
tf.reset_default_graph()
args = parser.parse_args()
total_vocab, vocabs, data, labels, max_len = load_datasets()
padded_data = padding(data, max_len)
vocab, words_in_word2vec, embds_in_word2vec, train_unknown_words, \
dev_and_test_unknown_words = word_embeddings(total_vocab,
vocabs)
vocab_lookup, embds = build_embeddings(vocab, words_in_word2vec,
embds_in_word2vec,
train_unknown_words,
dev_and_test_unknown_words)
if args.vis_embds != 0:
filename = "./images/embds.png"
vis_embds(embds, words_in_word2vec, train_unknown_words,
dev_and_test_unknown_words, filename)
trainer = Trainer()
output = trainer.build(max_len, vocab_lookup, embds, args.n_classes,
args.n_layers, args.fs, args.n_filters)
trainer.train(output, padded_data, labels, args.n_epochs, args.bs, args.lr)
if args.vis_embds != 0:
filename = "./images/embds_new.png"
vis_embds(embds, words_in_word2vec, train_unknown_words,
dev_and_test_unknown_words, filename)
