all_texts.append(texts)
enc = EncodeLoad(args)
out_dir = os.path.dirname(args.output)
if not os.path.exists(out_dir):
print(' - creating directory {}'.format(out_dir))
os.mkdir(out_dir)
all_data = []
all_index = []
for l in args.lang:
Token(os.path.join(args.base_dir, args.data + '.' + l),
os.path.join(args.base_dir, args.output + '.tok.' + l),
lang=l,
romanize=True if l == 'el' else False,
lower_case=True,
verbose=args.verbose,
over_write=False)
BPEfastApply(os.path.join(args.base_dir, args.output + '.tok.' + l),
os.path.join(args.base_dir, args.output + '.bpe.' + l),
args.bpe_codes,
verbose=args.verbose,
over_write=False)
EncodeFile(enc,
os.path.join(args.base_dir, args.output + '.bpe.' + l),
os.path.join(args.base_dir, args.output + '.enc.' + l),
verbose=args.verbose,
over_write=False)
d, idx = IndexCreate(os.path.join(args.base_dir,
args.output + '.enc.' + l),
############################################################
# load and preprocess input file
BPE_CODES = LASER + '/models/93langs.fcodes'
LANGUAGE_CODE = 'en'
VERBOSE = True
input_file = 'data/test_sentences.txt'
tokenized_f = 'data/test_tokenized.txt'
bpe_f = 'data/test_bpe.txt'
# tokenize
Token(
input_file,
tokenized_f,
lang=LANGUAGE_CODE,
romanize=False, #kept static for simplicity
lower_case=True,
gzip=False,
verbose=VERBOSE,
over_write=False)
# BPE
BPEfastApply(tokenized_f, bpe_f, BPE_CODES, verbose=VERBOSE, over_write=False)
############################################################
# Load + infer model
############################################################
model_path = LASER + '/models/bilstm.93langs.2018-12-26.pt'
model = LASEREmbedderIV(model_path, LASERHiddenExtractor, 300, 100, 10)
bpe_to_idx = torch.load(model_path)['dictionary']
