constants.py

"""
constants.py

此模块定义了各种常量，这些常量在整个项目中被广泛使用。通过集中管理这些常量，可以提高代码的可维护性和可读性，避免硬编码的值散布在代码的各个部分。
"""

import logging
import re

from torch._C import device

DEFAULT_CHARSET = 'UTF-8'

DEFAULT_RUN_MODE = 0

DEFAULT_SCORE_MIN = 0.8

DEFAULT_TRAIN_STAGE = 0

DEFAULT_RANDOM_SEED = 256

DEFAULT_GROUP_SIZE = 10

DEFAULT_THREAD_POOL_QUEUE_SIZE = 2048

DEFAULT_THREAD_POOL_SIZE = 8

DEFAULT_ONLINE_THREAD_POOL_SIZE = 1

DEFAULT_SENSITIVE_RATE = 0.9

DEFAULT_EXPECT_LENGTH = 8

DEFAULT_PROBLEM_TYPE = False  # True

DEFAULT_DATA_MODE = True

DEFAULT_OUT_MODE = True  # False

DEFAULT_LOG_MODE = False

DEFAULT_DATA_DIR = 'data'

DEFAULT_META_DATA = 'all_seq_2026-05-08_meta'

DEFAULT_FINAL_TRAIN_DATA = 'all_seq_2026-05-08_train_final'

DEFAULT_FINAL_DEV_DATA = 'all_seq_2026-05-08_dev_final'

DEFAULT_FINAL_TEST_DATA = 'all_seq_2026-05-08_test_final'

DEFAULT_RESULT_DATA = 'all_seq_2026-05-08_result'

DEFAULT_MODEL_DIR = 'model'

DEFAULT_DEV_MODEL_BEST = 'all_seq_2026-05-08_dev'

DEFAULT_TEST_MODEL_BEST = 'all_seq_2026-05-08_test'

DEFAULT_ONLINE_MODEL = 'all_seq_2026-05-08_online'

DEFAULT_TRAIN_MODE = True

DEFAULT_MODEL_MODE = 2

DEFAULT_FEATURE_SIZE = 52  # 72 50 76

DEFAULT_FIELD_SIZE = 36  # 56 34 59

DEFAULT_SEQ_SIZE = 150

DEFAULT_CLASS_SIZE = 3

DEFAULT_USE_DEEP = True

DEFAULT_EMBEDDING_SIZE = 32

DEFAULT_FM_FIRST_NORM_EPS = 0.00001

DEFAULT_FM_FIRST_NORM_ELEMENTWISE_AFFINE = True

DEFAULT_FM_FIRST_DROPOUT = 0.5

DEFAULT_FM_SECOND_NORM_EPS = 0.00001

DEFAULT_FM_SECOND_NORM_ELEMENTWISE_AFFINE = True

DEFAULT_FM_SECOND_DROPOUT = 0.5

DEFAULT_DEEP_SIZES = [128, 128]

DEFAULT_DEEP_NORM_EPSES = [0.00001, 0.00001]

DEFAULT_DEEP_NORM_ELEMENTWISE_AFFINES = [True, True]

DEFAULT_DEEP_DROPOUTS = [0.5, 0.5]

DEFAULT_COMBINATION_SIZE = 128

DEFAULT_COMBINATION_NORM_EPS = 0.00001

DEFAULT_COMBINATION_NORM_ELEMENTWISE_AFFINE = True

DEFAULT_COMBINATION_DROPOUT = 0.35

DEFAULT_LSTM_RECURRENT_DROPOUT = 0.2

DEFAULT_LSTM_SIZE = 128

DEFAULT_LSTM_NUM_LAYERS = 2

DEFAULT_LSTM_NORM_EPS = 0.00001

DEFAULT_LSTM_NORM_ELEMENTWISE_AFFINE = True

DEFAULT_LSTM_DROPOUT = 0.35

DEFAULT_ATTENTION_SIZE = 512

DEFAULT_ATTENTION_NUM_HEADS = 8

DEFAULT_ATTENTION_QUERY_NORM_EPS = 0.00001

DEFAULT_ATTENTION_QUERY_NORM_ELEMENTWISE_AFFINE = True

DEFAULT_ATTENTION_QUERY_DROPOUT = 0.2

DEFAULT_ATTENTION_KEY_NORM_EPS = 0.00001

DEFAULT_ATTENTION_KEY_NORM_ELEMENTWISE_AFFINE = True

DEFAULT_ATTENTION_KEY_DROPOUT = 0.2

DEFAULT_ATTENTION_VALUE_NORM_EPS = 0.00001

DEFAULT_ATTENTION_VALUE_NORM_ELEMENTWISE_AFFINE = True

DEFAULT_ATTENTION_VALUE_DROPOUT = 0.2

DEFAULT_ATTENTION_NORM_EPS = 0.00001

DEFAULT_ATTENTION_NORM_ELEMENTWISE_AFFINE = True

DEFAULT_ATTENTION_DROPOUT = 0.35

DEFAULT_TRANSFORMER_NUM_LAYERS = 8

DEFAULT_MULTIPLE_OF = 256

DEFAULT_FFN_DIM_MULTIPLIER = None

DEFAULT_NORM_EPS = 0.000001

DEFAULT_EPOCH_NUM = 128

DEFAULT_BUFFER_SIZE = 4096

DEFAULT_BATCH_SIZE = 64

DEFAULT_NUM_WORKERS = 0

DEFAULT_LEARN_RATE = 0.001

DEFAULT_EPS = 0.00000001

DEFAULT_BETA_1 = 0.9

DEFAULT_BETA_2 = 0.999

DEFAULT_WEIGHT_DECAY = 0.003

DEFAULT_GRAD_NORM_MAX = 5.0

DEFAULT_TIMESTAMP_FORMAT = '%Y-%m-%d'

DEVICE_CPU = device('cpu')

DEVICE_GPU = device('cuda')

SCORE_PRED_WEIGHT = 0.9

SCORE_CONFIDENCE_WEIGHT = 1.0 - SCORE_PRED_WEIGHT

SCORE_0_2_WEIGHT = 0.4

SCORE_0_3_WEIGHT = 0.3

SCORE_0_4_WEIGHT = 0.2

SCORE_0_5_WEIGHT = 0.1

SCORE_0_2_CONFIDENCE = 0.9126

SCORE_0_3_CONFIDENCE = 0.9254

SCORE_0_4_CONFIDENCE = 0.9747

SCORE_0_5_CONFIDENCE = 0.9886

SCORE_0_2_LOW = -0.1539  # -0.1448

SCORE_0_3_LOW = -0.2026  # -0.1928

SCORE_0_4_LOW = -0.1539  # -0.2331

SCORE_0_5_LOW = -0.2026  # -0.2724

SCORE_0_2_HIGH = 0.2

SCORE_0_3_HIGH = 0.3

SCORE_0_4_HIGH = 0.4

SCORE_0_5_HIGH = 0.5

LOG_LEVEL = logging.INFO

LOG_FORMAT = '%(asctime)s - %(filename)s[line:%(lineno)d] - %(levelname)s: %(message)s'

COMMA_REGEX = re.compile('\s*,+\s*', re.M)

NAME_RUBBISH_REGEX = re.compile('^\s*(?:(?:(?:[SP]\s*)?\*\s*)?(?:[SP]\s*)(?:T\s*)?|退\s*市\s*).{2,}|.{2,}\s*退\s*$',
                                re.I | re.M)