def main(unused):
parser = argparse.ArgumentParser()
parser.add_argument("embedding_path")
parser.add_argument("vocabulary_dir")
parser.add_argument("--training_dir")
parser.add_argument("--validation_dir")
parser.add_argument("--decode_dir")
parser.add_argument("--decode_out_dir")
parser.add_argument("--mode",
choices=["train", "validate", "decode"],
default="train")
parser.add_argument("--logdir")
parser.add_argument("--batch_size", type=int, default=30)
parser.add_argument("--validation_interval", type=int, default=20000)
parser.add_argument("--beam_width", type=int, default=5)
parser.add_argument("--max_output_length", type=int, default=32)
parser.add_argument("--target_vocabulary_size", type=int, default=20000)
parser.add_argument("--synthetic", action="store_true")
parser.add_argument("--allow_gpu_growth", action="store_true")
parser.add_argument("--collect_run_metadata", action="store_true")
parser.add_argument("--log_weight_images", action="store_true")
options = parser.parse_args()
if options.mode == "decode":
# Batching not supported in decoding
options.batch_size = 1
embedding_words, word_dict, word_embedding_dim = load_word_embeddings(
options.embedding_path)
vocabulary = Vocabulary()
summary_vocabulary_path = path.join(options.vocabulary_dir,
"summary_vocabulary.txt")
vocabulary.add_from_file(
summary_vocabulary_path,
options.target_vocabulary_size - len(vocabulary.words))
document_vocabulary_path = path.join(options.vocabulary_dir,
"document_vocabulary.txt")
# Add the most common words from vocabulary
vocabulary.add_from_file(document_vocabulary_path, 150000)
# Add additional common words from loaded embeddings
# note that embedding_words contains 2D numpy arrays
vocabulary.add_words(embedding_words[:100000])
run(options, word_dict, word_embedding_dim, vocabulary)
def main(unused):
parser = argparse.ArgumentParser()
parser.add_argument('embedding_path')
parser.add_argument('vocabulary_dir')
parser.add_argument('--training_dir')
parser.add_argument('--validation_dir')
parser.add_argument('--decode_dir')
parser.add_argument('--decode_out_dir')
parser.add_argument('--mode',
choices=['train', 'validate', 'decode'],
default='train')
parser.add_argument('--logdir')
parser.add_argument('--batch_size', type=int, default=30)
parser.add_argument('--validation_interval', type=int, default=20000)
parser.add_argument('--beam_width', type=int, default=5)
parser.add_argument('--max_output_length', type=int, default=32)
parser.add_argument('--target_vocabulary_size', type=int, default=20000)
parser.add_argument('--synthetic', action='store_true')
parser.add_argument('--allow_gpu_growth', action='store_true')
parser.add_argument('--collect_run_metadata', action='store_true')
parser.add_argument('--log_weight_images', action='store_true')
options = parser.parse_args()
if options.mode == 'decode':
# Batching not supported in decoding
options.batch_size = 1
embedding_words, word_dict, word_embedding_dim = load_word_embeddings(
options.embedding_path)
vocabulary = Vocabulary()
summary_vocabulary_path = path.join(options.vocabulary_dir,
'summary_vocabulary.txt')
vocabulary.add_from_file(
summary_vocabulary_path,
options.target_vocabulary_size - len(vocabulary.words))
document_vocabulary_path = path.join(options.vocabulary_dir,
'document_vocabulary.txt')
# Add the most common words from vocabulary
vocabulary.add_from_file(document_vocabulary_path, 150000)
# Add additional common words from loaded embeddings
vocabulary.add_words(embedding_words[:100000])
run(options, word_dict, word_embedding_dim, vocabulary)
