ConversationAgent 0.0.7

ConversationAgent

Using

Example:build a conversion system

1. build stages
2. Add stages to Agent
3. (option) using mock_client to up the agent

Setup Stage Class

class YourStage(Stage):
    def __init__(self):
        super(YourStage, self).__init__()
        self.stage_type = __YOUR_STAGE__
        self.sys_reply_q1 = "哈囉你好，現在能幫忙訂車票喔"
        self.sys_reply_q2 = "如果想訂票，請說 「線上訂票」，或是「我要訂票」"
        self.sys_reply_complete = f"好的，您將從 {self.get_default_var_label(__SAVE_LABEL__)} 出發"


    @staticmethod
    def is_fit_needs_n_gen_entity(kwargs) -> (bool, dict):
        if fit_re(".*訂票.*", kwargs.get(__USER_TEXT__, "")):
            kwargs = self.set_default_var(kwargs, __SAVE_LABEL__, __SAVE_CONTENT__)
            return True, kwargs
        else:
            return False, kwargs

- init Var

• stage_type: str, unique label for identify stage
• sys_reply_q1: str, reply for the user access the stage at the first time.
• sys_reply_q2: str, reply for the user when user's reply not fit to stage's needs.
• sys_reply_complete: str, reply for the user when user's reply fit to stage's needs.
  • self.get_default_var_label(__SAVE_LABEL__): would replace text by __SAVE_LABEL__

- setup stage's needs.

@staticmethod
def is_fit_needs_n_gen_entity(kwargs) -> (bool, dict):
    if fit_re(__RE_RULE__, kwargs.get(__USER_TEXT__, "")):
        kwargs = self.set_default_var(kwargs, __SAVE_LABEL__, __SAVE_CONTENT__)
        return True, kwargs
    else:
        return False, kwargs

• __RE_RULE__: str, re rule for catch entities, or you could build other function to identified entities
• kwargs.get(__USER_TEXT__, ""): str, user's source text
• __SAVE_LABEL__: str, a label for the var
• __SAVE_CONTENT__: str, content for the var

- (optional) disable first time reply (sys_reply_q1)

@staticmethod
def is_first_access(data, stage_id):
    return False

test your stage Class

a = YourStage()

# first time access (q1)
res = {}
res = a.run(**res)
assert res.get(__SYS_STAGE__, None) == StageStatus.FIRST

# refuse access (q2)
res[__USER_TEXT__] = "{something text not fit to your RE rule}"
res.pop(__SYS_REPLY__)
res = a.run(**res)
assert res.get(__SYS_STAGE__, None) == StageStatus.REFUSE

# accecpt access (complete)-1
res[__USER_TEXT__] = "{something text fit to your RE rule}"
res.pop(__SYS_REPLY__)
res = a.run(**res)
assert res.get(__SYS_STAGE__, None) == StageStatus.COMPLETE

# accecpt access (complete)-2
res.pop(__SYS_REPLY__)
res = a.run(**res)
assert res.get(__SYS_STAGE__, None) == StageStatus.COMPLETE

run your stage Class on mock client

from StageAgent import Agent, mock_client_human
from StageAgent.SentenceStage import HelloStage, OrderHelloStage,OrderStartStage,OrderEndStage,OrderConfirmStage,CleanStage

stages = [YourStage()]
mock_agent = Agent(stages)
mock_client_human(mock_agent)

- mock client structure

Example:build a conversion system by Lib-rule

python ./example/lib_order_ticket.py

請輸入：?
系統: ['請問是哪一站上車呢？']
請輸入：?
系統: ['請說站名，例如『板橋火車站』']
請輸入：?
系統: ['請說站名，例如『板橋火車站』']
請輸入：?
系統: ['請說站名，例如『板橋火車站』']
請輸入：板ㄑㄠˊ
系統: ['請說站名，例如『板橋火車站』']
請輸入：板橋
系統: ['好的，您將從板橋出發', '請問是哪一站下車呢？']
請輸入：新竹
系統: ['好的，您將在新竹下車，\n感謝您的使用，退出請輸入exit\n', 'thanks for using.']
請輸入：qwe
系統: ['thanks for using.']
請輸入：qwe
系統: ['thanks for using.']
請輸入：exit

Process finished with exit code 0

LIB-Stage Class

An easy way to build chat-flow by Json file

- init a LibStage Class （Only support LIB-Stage Class）

test_stage1 = {
    "stage_type": __RE_STAGE__,
    "question": {
        "sys_reply_q1": "請問是哪一站上車呢？",
        "sys_reply_q2": "請說站名，例如『板橋火車站』",
        "sys_reply_complete": f"好的，您將從 %%on_board_station%% 出發",
    },
    "is_fits": [
        ("[板橋|南港|新竹]+[火車|捷運|車]?[站]?$", "on_board_station"),
        ("[板橋|南港|新竹]+[公車站|轉運站]+$", "on_board_station"),
    ]

}
mock_client_human(gen_agent([test_stage1, test_stage2]))

• gen_agent: support REStage class, return Agent class

- init a LibSwitchStage Class

test_switch_stage = {
    "stage_type": __SWITCH_STAGE__,
    "stages_filter": [
        ("set_level", "商務", "s_1"),
        ("set_level", "商務座位", "s_1"),
        ("set_level", "商務座", "s_1"),
        ("set_level", "普通", "s_2"),
        ("set_level", "普通座", "s_2"),
        ("set_level", "普通座位", "s_2"),
    ]

}

mock_client_human(gen_multi_agent({
    MultiAgent.__MAIN_STAGES__: [__MOCK_STAGE_1__, test_switch_stage],
    "s_1": [__MOCK_STAGE_2__],
    "s_2": [__MOCK_STAGE_3__],
}))

• gen_multi_agent: input use dict, return MultiAgent class
  • MultiAgent.__MAIN_STAGES__ key is must to have