config.py

import os

BASE_DIR = '/home/rhodochrosited/'

PREPROCESS = {
    'vocab_file': os.path.join(BASE_DIR, 'data/vocab-9171.txt'),
    'use_morphs': True,
    'max_seq_length-search': 384,
    'max_query_length-search': 64,
    'max_query_length-similarity': 25,  # = seq_length
    'keywords_tags': ['NNG', 'NNP', 'NNB', 'NNBC', 'MAG', 'VV', 'VA', 'VCP', 'VCN', 'SL', 'SN'],
    'clean_tags': ['JK', 'JX', 'JC']
}

BERT = {  # 새로운 TENSOR SERVING 모델을 만들 때 사용
    'model_path-search': os.path.join(BASE_DIR, 'hdd2/FINAL_SQUAD/model.ckpt-12408'),
    'model_path-similarity': os.path.join(BASE_DIR, 'hdd2/FINAL_PRETRAIN/model.ckpt-990000'),
    'bert_json': os.path.join(BASE_DIR, 'squad_train_model/bert_config.json'),
    'similarity_layer': -1,
    # ELMO LIKE FEATURE VECTOR LAYERS 여러레이어를 더하거나, -2, -3... 하위 레이어 만을 사용 해보는 방법들 시도해보기
    'version-similarity': 5,
    # 1: 128 seq length 75000 step
    # 2: 512 seq length 990000 step
    # 3: 2 + similarity layer -2
    # 4: -12 layer(단어임베딩)
    'version-search': 2,
    # 1: F1 score 71
    # 2: F1 score 83.6 + train+dev -> (92)
    'version-sentiment': 3,
    # 2: predict - 1.0/ 0.0
    # 3: predict - 0.0~1.0 (중립추가 하기위해)
    'max_seq_length-search': 384,
    'max_seq_length-similarity': 25,
    'MODEL_DIR': os.path.join(BASE_DIR, 'hdd2/tensor_serving_models')
}

TENSOR_SERVING = {
    'url-search': 'http://10.140.0.8:8501/v1/models/search:predict',
    'url-similarity': 'http://10.140.0.8:8502/v1/models/similarity:predict',
    'url-sentiment': 'http://10.140.0.8:8503/v1/models/sentiment:predict',
    'url-search-v': 'http://10.140.0.8:8501/v1/models/search',
    'url-similarity-v': 'http://10.140.0.8:8502/v1/models/similarity',
    'url-sentiment-v': 'http://10.140.0.8:8503/v1/models/sentiment'
}

HANDLER = {
    'DUMMY': None
}

QUESTION = {
    'categories': ['shuttle_bus', 'food', 'talk', 'search', 'book', 'prepared', 'test'],
    'tfidf_token_pattern': r'(?u)\b[가-힣]+\b'
}

SEARCH = {
    'sublinear_tf': True,
    'max_context_num': 3  # search에서 tf idf 로 최대한 찾을 문단의 수
}

ANALYSIS = {
    # T-SNE
    'perplexity': 30.0,  # 5~50 권장
    'learning_rate': 200.0,  # 10~1000
    'n_iter': 1000,  # 최소 250
    'metric': 'cosine',  # distance metric ( x_i <-> x_j )
    'method': 'barnes_hut',  # 속도개선
    'n_components': 2,  # y 차원
    'categories': QUESTION['categories']
}

QUERY = {
    'distance': 'cosine',
    'jaccard_threshold': 0.7,
    'search_threshold': 15,
    'idf_weight': 0.1,
    'cosine_threshold': 0.87,
}

MONGODB = {
    'ip': 'localhost',
    'port': 27017,
    'db_name': 'chatbot',
    'col_questions': 'questions',
    'col_queries': 'queries',
    'col_contexts': 'contexts',
    'col_julimmal': 'julimmal',
    'col_stopword': 'stopword',
    'username': "mhlee",
    'password': "mhlee"
}

FLASK = {
    'host': '0.0.0.0',
    'port': 8000,
    'desc': None,
    'version': '0.3',
    'title': 'Willson-Hanyang_Chatbot',
    'debug': False
}

if __name__ == '__main__':
    print(BASE_DIR)
    print(PREPROCESS)