def _add_weather_db_config(cls, loader):
"""
添加天气问答数据库配置
"""
_logger = loader.logger
# 检查配置是否已经增加
if (StdQuestion.select().where(
(StdQuestion.milvus_id == -1) & (StdQuestion.q_type == 'context')
& (StdQuestion.question == '天气问答')).count()) > 0:
if _logger is not None:
_logger.info('Weather config already exists!')
return
# 插入标准问题
_std_q = StdQuestion.create(q_type='context',
milvus_id=-1,
collection=WEATHER_COLLECTION,
partition=WEATHER_PARTITION,
question='天气问答')
# 插入问题答案
Answer.create(
std_question_id=_std_q.id,
a_type='ask',
type_param="['ApiToolAsk', 'weather', '%s', '%s', {}, True]" %
(WEATHER_COLLECTION, WEATHER_PARTITION),
replace_pre_def='N',
answer=WEATHER_ERROR)
# 插入NLP意图识别配置
NlpPurposConfigDict.create(
action='天气查询',
match_collection='',
match_partition='',
collection=WEATHER_COLLECTION,
partition=WEATHER_PARTITION,
std_question_id=_std_q.id,
order_num=0,
exact_match_words='[]',
exact_ignorecase='N',
match_words="['天气', '今天天气']",
ignorecase='N',
word_scale=0.0,
info=
"['InitInfo', 'get_wordclass_list', {'condition': [{'key': 'time', 'class': ['t']}, {'key': 'addr', 'class': ['ns']}]}]",
check=
"['InitCheck', 'reject_by_nest', {'next': {'天气': ['真好', '不错', '真差']}, }]"
)
def import_config(cls, qa_manager: QAManager, logger: Logger):
"""
添加标准配置(不考虑删除问题)
@param {QAManager} qa_manager - 数据管理对象
@param {Logger} logger - 日志对象
"""
# 插入标准问题
_std_q = StdQuestion.create(tag='leave_message_direct_action',
q_type='context',
milvus_id=-1,
collection=LEAVE_MESSAGE_PLUGIN_COLLECTION,
partition=LEAVE_MESSAGE_PLUGIN_PARTITION,
question='留言插件通用处理')
# 插入问题答案
Answer.create(
std_question_id=_std_q.id,
a_type='ask',
type_param="['LeaveMessagePlugin', 'save_msg', '', '', {}, True]",
replace_pre_def='N',
answer='留言插件通用处理')
if logger is not None:
logger.info(
'create leave message plugin std question config success!')
# 创建留言意图参数
NlpPurposConfigDict.create(
action='leave_message',
match_collection='',
match_partition='',
collection=LEAVE_MESSAGE_PLUGIN_COLLECTION,
partition=LEAVE_MESSAGE_PLUGIN_PARTITION,
std_question_id=_std_q.id,
order_num=LEAVE_MESSAGE_PLUGIN_NLP_CONFIG['order_num'],
exact_match_words=str(
LEAVE_MESSAGE_PLUGIN_NLP_CONFIG['exact_match_words']),
exact_ignorecase=LEAVE_MESSAGE_PLUGIN_NLP_CONFIG[
'exact_ignorecase'],
match_words=str(LEAVE_MESSAGE_PLUGIN_NLP_CONFIG['match_words']),
ignorecase=LEAVE_MESSAGE_PLUGIN_NLP_CONFIG['ignorecase'],
word_scale=LEAVE_MESSAGE_PLUGIN_NLP_CONFIG['word_scale'],
info=str(LEAVE_MESSAGE_PLUGIN_NLP_CONFIG['info']),
check=str(LEAVE_MESSAGE_PLUGIN_NLP_CONFIG['check']))
if logger is not None:
logger.info('create leave message plugin nlp config success!')
# 创建文件上传参数
UploadFileConfig.create(
upload_type=LEAVE_MESSAGE_PLUGIN_UPLOAD_FILE_CONFIG['upload_type'],
exts=LEAVE_MESSAGE_PLUGIN_UPLOAD_FILE_CONFIG['exts'],
size=LEAVE_MESSAGE_PLUGIN_UPLOAD_FILE_CONFIG['size'],
save_path=LEAVE_MESSAGE_PLUGIN_UPLOAD_FILE_CONFIG['save_path'],
url=LEAVE_MESSAGE_PLUGIN_UPLOAD_FILE_CONFIG['url'],
rename=LEAVE_MESSAGE_PLUGIN_UPLOAD_FILE_CONFIG['rename'],
after=LEAVE_MESSAGE_PLUGIN_UPLOAD_FILE_CONFIG['after'],
remark=LEAVE_MESSAGE_PLUGIN_UPLOAD_FILE_CONFIG['remark'])
if logger is not None:
logger.info(
'create leave message plugin upload file config success!')
def import_config(cls, qa_manager: QAManager, logger: Logger):
"""
添加标准配置(不考虑删除问题)
@param {QAManager} qa_manager - 数据管理对象
@param {Logger} logger - 日志对象
"""
FORM_PLUGIN_CONFIG = RunTool.get_global_var('FORM_PLUGIN_CONFIG')
if FORM_PLUGIN_CONFIG is None:
FORM_PLUGIN_CONFIG = dict()
RunTool.set_global_var('FORM_PLUGIN_CONFIG', FORM_PLUGIN_CONFIG)
FORM_PLUGIN_SELF_TABLE = RunTool.get_global_var(
'FORM_PLUGIN_SELF_TABLE')
if FORM_PLUGIN_SELF_TABLE is None:
FORM_PLUGIN_SELF_TABLE = dict()
RunTool.set_global_var('FORM_PLUGIN_SELF_TABLE',
FORM_PLUGIN_SELF_TABLE)
# 插入标准问题
_std_q = StdQuestion.create(tag='form_direct_action',
q_type='context',
milvus_id=-1,
collection=FORM_PLUGIN_COLLECTION,
partition=FORM_PLUGIN_PARTITION,
question='表单插件通用处理')
# 插入问题答案
Answer.create(std_question_id=_std_q.id,
a_type='job',
type_param="['FormPlugin', 'operate', {}]",
replace_pre_def='N',
answer='表单插件通用处理')
if logger is not None:
logger.info('create form plugin std question config success!')
# 处理扩展插件
if FORM_PLUGIN_SEARCH_PATH is not None:
_path = os.path.join(os.path.dirname(__file__),
FORM_PLUGIN_SEARCH_PATH)
_file_list = FileTool.get_filelist(path=_path,
regex_str=r'.*\.py$',
is_fullname=False)
for _file in _file_list:
if _file == '__init__.py':
continue
# 执行加载
_module = ImportTool.import_module(_file[0:-3],
extend_path=_path,
is_force=True)
_clsmembers = inspect.getmembers(_module, inspect.isclass)
for (_class_name, _class) in _clsmembers:
if _module.__name__ != _class.__module__:
# 不是当前模块定义的函数
continue
# 判断类型
_get_form_type = getattr(_class, 'get_form_type', None)
if _get_form_type is None or not callable(_get_form_type):
# 不是标准的插件类
continue
_form_type = _get_form_type()
_get_form_config = getattr(_class, 'get_form_config', None)
# 加入配置
FORM_PLUGIN_CONFIG[_form_type] = _get_form_config()
# 循环插件实例进行处理
for _form_type in FORM_PLUGIN_CONFIG.keys():
_config = FORM_PLUGIN_CONFIG[_form_type]
# 创建表单类型意图参数
NlpPurposConfigDict.create(
action=_form_type,
match_collection='',
match_partition='',
collection=FORM_PLUGIN_COLLECTION,
partition=FORM_PLUGIN_PARTITION,
std_question_id=_std_q.id,
order_num=_config['order_num'],
exact_match_words=str(_config['exact_match_words']),
exact_ignorecase=_config['exact_ignorecase'],
match_words=str(_config['match_words']),
ignorecase=_config['ignorecase'],
word_scale=_config['word_scale'],
info=str(_config['info']),
check=str(_config['check']))
if logger is not None:
logger.info('create form plugin [%s] success!' % _form_type)
def _import_answers_by_xls(self, excel_io, milvus: mv.Milvus, bert: BertClient,
std_question_id_mapping: dict):
"""
导入Answers
@param {object} excel_io - pd.io.excel.ExcelFile的IO文件
@param {Milvus} milvus - Milvus连接对象
@param {BertClient} bert - bert服务连接对象
@param {dict} std_question_id_mapping - 标准问题id映射字典
"""
try:
# 读取标题行
_df_header = pd.read_excel(
excel_io, sheet_name='Answers', nrows=0, engine=self.excel_engine
)
except:
_df_header = None # 没有获取到指定的页
if _df_header is not None:
# 定义替换变量函数
def replace_var_fun(m):
_match_str = m.group(0)
if _match_str.startswith('{$id='):
# 替换为映射id
_id: str = _match_str[5: -2]
if _id.isdigit():
# 是数字
_new_id = std_question_id_mapping.get(
int(_id), _id
)
else:
# 是字符串
_new_id = std_question_id_mapping.get(
_id, _id
)
return str(_new_id)
# 没有匹配到
return _match_str
_skiprows = 1 # 跳过的记录数
_columns = {i: col for i, col in enumerate(_df_header.columns.tolist())}
while True:
# 循环处理
_df = pd.read_excel(
excel_io, sheet_name='Answers', nrows=self.excel_batch_num,
header=None, skiprows=_skiprows, engine=self.excel_engine
)
_skiprows += self.excel_batch_num
if not _df.shape[0]:
# 获取不到数据
break
# 变更标题
_df.rename(columns=_columns, inplace=True)
for _index, _row in _df.iterrows():
# 逐行添加标准问题答案, _index为行，_row为数据集
try:
_std_question_id = std_question_id_mapping.get(
_row['std_question_id'], _row['std_question_id']
)
_type_param = re.sub(
r'\{\$.+?\$\}', replace_var_fun, str(_row['type_param']), re.M
)
Answer.create(
std_question_id=_std_question_id, a_type=_row['a_type'],
type_param=_type_param, replace_pre_def=_row['replace_pre_def'],
answer=_row['answer']
)
except:
self._log_error('imported answer [id: %s] [%s] error: %s' % (
str(_row['std_question_id']), _row['answer'], traceback.format_exc()
))
self._log_debug('imported answers[%d]: %s' % (_skiprows, str(_df)))
def add_std_question(self, question: str, collection: str = 'chat', q_type: str = 'ask',
partition: str = None, answer: str = None, a_type: str = 'text',
replace_pre_def: str = 'N', a_type_param: str = '') -> int:
"""
添加标准问题
@param {str} question - 标准问题
@param {str} collection='chat' - 问题分类，默认为'chat', 可以自定义分类
@param {str} q_type='ask' - 问题类型
ask-问答类（问题对应答案）
context-场景类（问题对应上下文场景）
@param {str} partition=None - 问题所属场景, q_type为context时使用
@param {str} answer=None - 标准问题对应的答案
@param {string} a_type='text' - 答案类型
text-文字答案
@param {str} replace_pre_def='N' - 是否替换答案的预定义字符
@param {string} a_type_param='' - 答案类型扩展参数
@returns {int} - 返回问题记录对应的id
"""
# 简单校验
if q_type == 'ask' and answer is None:
raise AttributeError('parameter answer should be not None!')
# 获取问题的向量值
with self.get_bert_client() as _bert, self.get_milvus() as _milvus:
_vectors = _bert.encode([question, ])
_question_vectors = self.normaliz_vec(_vectors.tolist())
self._log_debug('get question vectors: %s' % str(len(_question_vectors)))
# 存入Milvus服务, 先创建分类
self._add_collection(collection, _milvus)
# 创建场景
if partition is not None and partition != '':
self._add_partition(collection, partition, _milvus)
else:
partition = None
_milvus_id = self._add_milvus_question(
_question_vectors[0], collection, partition, _milvus
)
# 存入AnswerDb，通过事务处理
with self.database.atomic() as _txn:
# 插入标准问题
_std_q = StdQuestion.create(
q_type=q_type, milvus_id=_milvus_id, collection=collection,
partition=('' if partition is None else partition),
question=question
)
# 插入对应的答案
if answer is not None:
Answer.create(
std_question_id=_std_q.id, a_type=a_type, replace_pre_def=replace_pre_def,
type_param=a_type_param, answer=answer
)
# 提交事务
_txn.commit()
# 返回结果
self._log_debug('insert question: %s' % str(_std_q))
return _std_q.id