def nmt_construction():
config_file = "/home/user_data/baoy/projects/seq2seq_parser/configs/data_configs/nmt.yaml"
args_dict = yaml_load_dict(config_file)
args = dict_to_args(args_dict)
data_dir = args.origin_tgts
data_dict = {
"train": "train.s2b",
"dev": "dev.s2b",
"test": "test.s2b",
}
make_dataset(
data_dir=data_dir,
data_dict=data_dict,
tgt_dir=data_dir,
max_src_vocab=args.max_src_vocab,
max_tgt_vocab=args.max_tgt_vocab,
vocab_freq_cutoff=args.cut_off,
max_src_length=-1,
max_tgt_length=-1,
)
make_dataset(
data_dir=data_dir,
data_dict=data_dict,
tgt_dir=args.sample_tgts,
max_src_vocab=args.max_src_vocab,
max_tgt_vocab=args.max_tgt_vocab,
vocab_freq_cutoff=args.cut_off,
max_src_length=args.max_src_length,
max_tgt_length=args.max_tgt_length,
)
def prepare_con():
config_file = "/home/user_data/baoy/projects/seq2seq_parser/configs/data_configs/quora.yaml"
args_dict = yaml_load_dict(config_file)
args = dict_to_args(args_dict)
pair_file = os.path.join(args.origin_tgts, args.pair_file)
unpair_file = os.path.join(args.origin_tgts, args.unpair_file)
pair_tree = load_tree_file(pair_file)
unpair_tree = load_tree_file(unpair_file)
all_size = args.train_size + args.valid_size + args.test_size
all_tree = pair_tree + unpair_tree
all_idx = random.sample(range(len(all_tree)), all_size)
train_idx = all_idx[:args.train_size]
valid_idx = all_idx[args.train_size:args.train_size + args.valid_size]
test_idx = all_idx[-args.test_size:]
train_tree = [all_tree[idx] for idx in train_idx]
valid_tree = [all_tree[idx] for idx in valid_idx]
test_tree = [all_tree[idx] for idx in test_idx]
if not os.path.exists(args.data_tgts):
os.makedirs(args.data_tgts)
write_docs(docs=train_tree,
fname=os.path.join(args.data_tgts, "train.con"))
write_docs(docs=valid_tree, fname=os.path.join(args.data_tgts, "dev.con"))
write_docs(docs=test_tree, fname=os.path.join(args.data_tgts, "test.con"))
def prepare_s2b(
config_file="/home/user_data/baoy/projects/seq2seq_parser/configs/data_configs/quora.yaml"
):
args_dict = yaml_load_dict(config_file)
args = dict_to_args(args_dict)
make_s2b_dataset(train_file=os.path.join(args.data_tgts, "train.con"),
dev_file=os.path.join(args.data_tgts, "dev.con"),
test_file=os.path.join(args.data_tgts, "test.con"),
tgt_dir=args.data_tgts)
def webnlg_preprocess():
config_file = "/home/user_data/baoy/projects/seq2seq_parser/configs/data_configs/webnlg.yaml"
args_dict = yaml_load_dict(config_file)
args = dict_to_args(args_dict)
make_nmt_simple_dataset(
train_file=args.train_file,
dev_file=args.dev_file,
test_file=args.test_file,
tgt_dir=args.origin_tgts,
)
def prepare_raw_paraphrase():
config_file = "/home/user_data/baoy/projects/seq2seq_parser/configs/data_configs/quora.yaml"
args_dict = yaml_load_dict(config_file)
args = dict_to_args(args_dict)
with open(os.path.join(args.origin_tgts, "para.raw.token"), "r") as f:
pair_raws = [line for line in f.readlines()]
select_idx = random.sample(range(len(pair_raws)), args.test_size)
test_paraphrase = []
for idx in select_idx:
test_paraphrase.append(pair_raws[idx].strip())
write_docs(fname=os.path.join(args.origin_tgts, "para.raw.text"),
docs=test_paraphrase)
def prepare_paraphrase():
config_file = "/home/user_data/baoy/projects/seq2seq_parser/configs/data_configs/quora.yaml"
args_dict = yaml_load_dict(config_file)
args = dict_to_args(args_dict)
pair_file = os.path.join(args.origin_tgts, args.pair_file)
pair_trees = load_to_pair_tree(pair_file)
select_idx = random.sample(range(len(pair_trees)), args.test_size)
test_paraphrase = []
for idx in select_idx:
pair_tree = pair_trees[idx]
test_paraphrase.append("\t".join(
[pair_tree[0].words, pair_tree[1].words]))
write_docs(fname=os.path.join(args.origin_tgts, "para.text"),
docs=test_paraphrase)
def parsed_to_pair(args=None):
if args is None:
config_file = "/home/user_data/baoy/projects/seq2seq_parser/configs/data_configs/quora.yaml"
args_dict = yaml_load_dict(config_file)
args = dict_to_args(args_dict)
origin_file = os.path.join(args.origin_tgts, args.origin_file)
label_file = os.path.join(args.origin_tgts, args.label_file)
pair_file = os.path.join(args.origin_tgts, args.pair_file)
unpair_file = os.path.join(args.origin_tgts, args.unpair_file)
with open(origin_file, "r") as f:
tree_list = [line.strip() for line in f.readlines()]
with open(label_file, "r") as f:
label_list = [line.strip().split(" ") for line in f.readlines()]
pair_list = []
unpair_list = []
for label in label_list:
try:
num_i1 = int(label[0]) - 1
t1 = tree_list[num_i1]
num_i2 = int(label[1]) - 1
t2 = tree_list[num_i2]
num_l = int(label[2])
if len(t1.strip()) > 0 and len(t2.strip()) > 0:
item = [t1, t2]
item_str = "\t".join(item)
if num_l == 0:
unpair_list.append(item_str)
elif num_l == 1:
pair_list.append(item_str)
else:
print(t1)
print(t2)
except:
pass
with open(pair_file, "w") as f:
for line in pair_list:
f.write(line.strip())
f.write("\n")
with open(unpair_file, "w") as f:
for line in unpair_list:
f.write(line.strip())
f.write("\n")
def snli_sample_construction(is_write=True):
config_file = "/home/user_data/baoy/projects/seq2seq_parser/configs/data_configs/snli-sample.yaml"
args_dict = yaml_load_dict(config_file)
args = dict_to_args(args_dict)
data_dir = args.origin_tgts
data_dict = {
"train": "train.s2b",
"dev": "dev.s2b",
"test": "test.s2b",
}
make_dataset(
data_dir=data_dir,
data_dict=data_dict,
tgt_dir=args.target_tgts,
max_src_vocab=args.max_src_vocab,
max_tgt_vocab=args.max_tgt_vocab,
vocab_freq_cutoff=args.cut_off,
max_src_length=-1,
max_tgt_length=-1,
write_down=is_write,
)
def process_args():
opt_parser = argparse.ArgumentParser()
opt_parser.add_argument('--config_files', type=str, help='config_files')
opt_parser.add_argument('--exp_name', type=str, help='config_files')
opt_parser.add_argument('--load_src_lm', type=str, default=None)
opt_parser.add_argument('--mode', type=str, default=None)
opt = opt_parser.parse_args()
configs = yaml_load_dict(opt.config_files)
base_args = dict_to_args(configs['base_configs']) if 'base_configs' in configs else None
baseline_args = dict_to_args(configs['baseline_configs']) if 'baseline_configs' in configs else None
prior_args = dict_to_args(configs['prior_configs']) if 'prior_configs' in configs else None
encoder_args = dict_to_args(configs['encoder_configs']) if 'encoder_configs' in configs else None
decoder_args = dict_to_args(configs['decoder_configs']) if 'decoder_configs' in configs else None
vae_args = dict_to_args(configs['vae_configs']) if 'vae_configs' in configs else None
ae_args = dict_to_args(configs["ae_configs"]) if 'ae_configs' in configs else None
nae_args = dict_to_args(configs["nag_configs"]) if 'nag_configs' in configs else None
if base_args is not None:
if opt.mode is not None:
base_args.mode = opt.mode
if opt.exp_name is not None:
base_args.exp_name = opt.exp_name
if opt.load_src_lm is not None:
base_args.load_src_lm = opt.load_src_lm
return {
'base': base_args,
"baseline": baseline_args,
'prior': prior_args,
'encoder': encoder_args,
"decoder": decoder_args,
"vae": vae_args,
"ae": ae_args,
"nae": nae_args
}
def quora_construction(
config_file="/home/user_data/baoy/projects/seq2seq_parser/configs/data_configs/quora-50k.yaml",
is_write=True):
# config_file = "/home/user_data/baoy/projects/seq2seq_parser/configs/data_configs/quora-50k.yaml"
args_dict = yaml_load_dict(config_file)
args = dict_to_args(args_dict)
data_dir = args.origin_tgts
data_dict = {
"train": "train.s2b",
"dev": "dev.s2b",
"test": "test.s2b",
}
make_dataset(data_dir=data_dir,
data_dict=data_dict,
tgt_dir=args.data_tgts,
max_src_vocab=args.max_src_vocab,
max_tgt_vocab=args.max_tgt_vocab,
vocab_freq_cutoff=args.cut_off,
max_src_length=args.max_src_length,
max_tgt_length=args.max_tgt_length,
train_size=args.train_size,
write_down=is_write)
print("Filter")
data_details(train, dev, test)
test, dev, train = load_ptb_to_s2b(
data_dir=args.data_dirs,
data_dict=data_dicts,
same_filter=False,
)
print("Origin")
data_details(train, dev, test)
"""
if __name__ == '__main__':
config_file = "/home/user_data/baoy/projects/seq2seq_parser/configs/snli_data.yaml"
args_dict = yaml_load_dict(config_file)
args = dict_to_args(args_dict)
data_dicts = {
'train': args.train_file,
'dev': args.dev_file,
'test': args.test_file,
}
t_dict = {
'train': args.process_train,
'dev': args.process_dev,
'test': args.process_test,
}
prepare_ptb_to_s2b(
data_dir=args.data_dirs,
data_dict=data_dicts,
target_dir=args.sample_dirs,
