def load_datasets(load_existing_dump=False):
model_config = ModelConfig()
data_reader = DataReader()
train_lines = open(
os.path.join(DataConfig.data_dir_path, DataConfig.train_path),
"r").readlines()
valid_lines = open(
os.path.join(DataConfig.data_dir_path, DataConfig.valid_path),
"r").readlines()
test_lines = open(
os.path.join(DataConfig.data_dir_path, DataConfig.test_path),
"r").readlines()
# Load data
train_data = data_reader.read_data(train_lines)
print("Loaded Train data")
valid_data = data_reader.read_data(valid_lines)
print("Loaded Dev data")
test_data = data_reader.read_data(test_lines)
print("Loaded Test data")
feature_extractor = FeatureExtractor(model_config)
dataset = Dataset(model_config, train_data, valid_data, test_data,
feature_extractor)
# Vocab processing
if load_existing_dump:
dataset.word2idx = get_pickle(
os.path.join(DataConfig.dump_dir, DataConfig.word_vocab_file))
dataset.idx2word = {
idx: word
for (word, idx) in dataset.word2idx.items()
}
dataset.pos2idx = get_pickle(
os.path.join(DataConfig.dump_dir, DataConfig.pos_vocab_file))
dataset.idx2pos = {idx: pos for (pos, idx) in dataset.pos2idx.items()}
dataset.dep2idx = get_pickle(
os.path.join(DataConfig.dump_dir, DataConfig.dep_vocab_file))
dataset.idx2dep = {idx: dep for (dep, idx) in dataset.dep2idx.items()}
dataset.model_config.load_existing_vocab = True
print "loaded existing Vocab!"
dataset.word_embedding_matrix = get_pickle(
os.path.join(DataConfig.dump_dir, DataConfig.word_emb_file))
dataset.pos_embedding_matrix = get_pickle(
os.path.join(DataConfig.dump_dir, DataConfig.pos_emb_file))
dataset.dep_embedding_matrix = get_pickle(
os.path.join(DataConfig.dump_dir, DataConfig.dep_emb_file))
print "loaded existing embedding matrix!"
else:
dataset.build_vocab()
dump_pickle(
dataset.word2idx,
os.path.join(DataConfig.dump_dir, DataConfig.word_vocab_file))
dump_pickle(
dataset.pos2idx,
os.path.join(DataConfig.dump_dir, DataConfig.pos_vocab_file))
dump_pickle(
dataset.dep2idx,
os.path.join(DataConfig.dump_dir, DataConfig.dep_vocab_file))
dataset.model_config.load_existing_vocab = True
print "Vocab Build Done!"
dataset.build_embedding_matrix()
print "embedding matrix Build Done"
dump_pickle(
dataset.word_embedding_matrix,
os.path.join(DataConfig.dump_dir, DataConfig.word_emb_file))
dump_pickle(dataset.pos_embedding_matrix,
os.path.join(DataConfig.dump_dir, DataConfig.pos_emb_file))
dump_pickle(dataset.dep_embedding_matrix,
os.path.join(DataConfig.dump_dir, DataConfig.dep_emb_file))
print "converting data into ids.."
dataset.convert_data_to_ids()
print "Done!"
dataset.model_config.word_features_types = len(dataset.train_inputs[0][0])
dataset.model_config.pos_features_types = len(dataset.train_inputs[1][0])
dataset.model_config.dep_features_types = len(dataset.train_inputs[2][0])
dataset.model_config.num_features_types = dataset.model_config.word_features_types + \
dataset.model_config.pos_features_types + dataset.model_config.dep_features_types
dataset.model_config.num_classes = len(dataset.train_targets[0])
return dataset
def load_datasets(load_existing_dump=False):
model_config = ModelConfig()
data_reader = DataReader()
train_lines = open(
os.path.join(DataConfig.data_dir_path, DataConfig.train_path),
"r").readlines()
valid_lines = open(
os.path.join(DataConfig.data_dir_path, DataConfig.valid_path),
"r").readlines()
test_lines = open(
os.path.join(DataConfig.data_dir_path, DataConfig.test_path),
"r").readlines()
# Load data
train_data_obj = data_reader.read_data(train_lines)
print("Loaded Train data")
valid_data_obj = data_reader.read_data(valid_lines)
print("Loaded Dev data")
test_data_obj = data_reader.read_data(test_lines)
print("Loaded Test data")
feature_extractor = FeatureExtractor(model_config)
dataset = Dataset(model_config, train_data_obj, valid_data_obj,
test_data_obj, feature_extractor)
dataset.model_config.max_seq_len = dataset.get_max_seq_len()
dataset.model_config.max_word_len = dataset.get_max_word_len()
# Vocab processing
if load_existing_dump:
dataset.word2idx = get_pickle(
os.path.join(DataConfig.dump_dir, DataConfig.word_vocab_file))
dataset.idx2word = {
idx: word
for (word, idx) in dataset.word2idx.items()
}
dataset.char2idx = get_pickle(
os.path.join(DataConfig.dump_dir, DataConfig.char_vocab_file))
dataset.idx2char = {
idx: char
for (char, idx) in dataset.char2idx.items()
}
dataset.label2idx = get_pickle(
os.path.join(DataConfig.dump_dir, DataConfig.label_vocab_file))
dataset.idx2label = {
idx: label
for (label, idx) in dataset.label2idx.items()
}
dataset.model_config.load_existing_vocab = True
print "loaded existing Vocab!"
dataset.word_embedding_matrix = get_pickle(
os.path.join(DataConfig.dump_dir, DataConfig.word_emb_file))
dataset.char_embedding_matrix = get_pickle(
os.path.join(DataConfig.dump_dir, DataConfig.char_emb_file))
print "loaded existing embedding matrix!"
else:
dataset.build_vocab()
dump_pickle(
dataset.word2idx,
os.path.join(DataConfig.dump_dir, DataConfig.word_vocab_file))
dump_pickle(
dataset.char2idx,
os.path.join(DataConfig.dump_dir, DataConfig.char_vocab_file))
dump_pickle(
dataset.label2idx,
os.path.join(DataConfig.dump_dir, DataConfig.label_vocab_file))
dataset.model_config.load_existing_vocab = True
print "Vocab Build Done!"
dataset.build_embedding_matrix()
print "embedding matrix Build Done"
dump_pickle(
dataset.word_embedding_matrix,
os.path.join(DataConfig.dump_dir, DataConfig.word_emb_file))
dump_pickle(
dataset.char_embedding_matrix,
os.path.join(DataConfig.dump_dir, DataConfig.char_emb_file))
dataset.model_config.num_classes = len(dataset.label2idx)
print "converting data into ids.."
dataset.convert_data_to_ids()
print "Done!"
return dataset
