Spaces:
Sleeping
Sleeping
File size: 5,148 Bytes
6fc683c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 |
import argparse
import json
import random
from transformers import XLMRobertaTokenizer
from transformers import xglue_processors as processors
if __name__ == "__main__":
parser = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
"--data_dir",
default=None,
type=str,
required=True,
help="The input data dir. Should contain the .tsv files (or other data files) for the task.",
)
parser.add_argument(
"--nbest_size", type=int, default=-1, help="nbest size in sampling subword sequence"
)
parser.add_argument(
"--alpha", type=float, default=0.2, help="alpha"
)
parser.add_argument(
"--train_language", default=None, type=str, help="Train language if is different of the evaluation language."
)
parser.add_argument(
"--language",
default=None,
type=str,
required=True,
help="Evaluation language. Also train language if `train_language` is set to None.",
)
parser.add_argument(
"--do_lower_case", action="store_true", help="Set this flag if you are using an uncased model."
)
parser.add_argument(
"--model_name_or_path",
default=None,
type=str,
required=True,
help="Path to pre-trained model",
)
parser.add_argument(
"--sample_rounds",
default=1,
type=int,
required=True,
help="Path to pre-trained model",
)
args = parser.parse_args()
tokenizer = XLMRobertaTokenizer.from_pretrained(
args.model_name_or_path,
do_lower_case=args.do_lower_case,
cache_dir=None,
)
task = "xnli"
processor = processors[task](language=args.train_language, train_language=args.train_language)
examples = processor.get_train_examples(args.data_dir)
train_word_cnt_origin = {}
for example in examples:
tokens_a = tokenizer.tokenize(example.text_a, add_special_tokens=True)
tokens_b = tokenizer.tokenize(example.text_b, add_special_tokens=True)
for token in tokens_a + tokens_b:
if token not in train_word_cnt_origin:
train_word_cnt_origin[token] = 0
train_word_cnt_origin[token] += 1
all_examples = []
for i in range(args.sample_rounds):
all_examples += examples
examples = all_examples
sent_len = []
example_len = []
train_word_cnt = {}
for example in examples:
tokens_a = tokenizer.tokenize(example.text_a, add_special_tokens=True, nbest_size=args.nbest_size,
alpha=args.alpha)
tokens_b = tokenizer.tokenize(example.text_b, add_special_tokens=True, nbest_size=args.nbest_size,
alpha=args.alpha)
for token in tokens_a + tokens_b:
if token not in train_word_cnt:
train_word_cnt[token] = 0
train_word_cnt[token] += 1
sent_len += [len(tokens_a), len(tokens_b)]
example_len += [len(tokens_a) + len(tokens_b)]
print(sum(sent_len) / len(sent_len))
print(sum(example_len) / len(example_len))
total = 0
n_oov = 0
for token in train_word_cnt:
if token not in train_word_cnt_origin:
n_oov += train_word_cnt[token]
# n_oov += 1
total += train_word_cnt[token]
# total += 1
print("{} oov rate: {}".format("extra", n_oov / total))
total = 0
n_oov = 0
for token in train_word_cnt_origin:
if token not in train_word_cnt:
n_oov += train_word_cnt_origin[token]
# n_oov += 1
total += train_word_cnt_origin[token]
# total += 1
print("{} oov rate: {}".format("origin", n_oov / total))
# exit(0)
eval_datasets = []
eval_langs = args.language.split(',')
for split in ["valid"]:
for lang in eval_langs:
eval_datasets.append((split, lang))
for split, lang in eval_datasets:
processor = processors[task](language=lang, train_language=lang)
examples = processor.get_valid_examples(args.data_dir)
sent_len = []
example_len = []
valid_word_cnt = {}
for example in examples:
tokens_a = tokenizer.tokenize(example.text_a, add_special_tokens=True)
tokens_b = tokenizer.tokenize(example.text_b, add_special_tokens=True)
for token in tokens_a + tokens_b:
if token not in valid_word_cnt:
valid_word_cnt[token] = 0
valid_word_cnt[token] += 1
sent_len += [len(tokens_a), len(tokens_b)]
example_len += [len(tokens_a) + len(tokens_b)]
print(sum(sent_len) / len(sent_len))
print(sum(example_len) / len(example_len))
total = 0
n_oov = 0
for token in valid_word_cnt:
if token not in train_word_cnt:
n_oov += valid_word_cnt[token]
# n_oov += 1
total += valid_word_cnt[token]
# total += 1
print("{} oov rate: {}".format(lang, n_oov / total))
|