mathtext-fastapi / mathtext_fastapi /conversation_manager.py
Greg Thompson
Update conversation manager to modularize fsm call
bf90676
raw
history blame
9.29 kB
import base64
import dill
import os
import json
import jsonpickle
import pickle
import random
import requests
from dotenv import load_dotenv
from mathtext_fastapi.nlu import evaluate_message_with_nlu
from mathtext_fastapi.math_quiz_fsm import MathQuizFSM
from supabase import create_client
from transitions import Machine
load_dotenv()
SUPA = create_client(
os.environ.get('SUPABASE_URL'),
os.environ.get('SUPABASE_KEY')
)
def create_text_message(message_text, whatsapp_id):
""" Fills a template with input values to send a text message to Whatsapp
Inputs
- message_text: str - the content that the message should display
- whatsapp_id: str - the message recipient's phone number
Outputs
- message_data: dict - a preformatted template filled with inputs
"""
message_data = {
"preview_url": False,
"recipient_type": "individual",
"to": whatsapp_id,
"type": "text",
"text": {
"body": message_text
}
}
return message_data
def create_button_objects(button_options):
""" Creates a list of button objects using the input values
Input
- button_options: list - a list of text to be displayed in buttons
Output
- button_arr: list - preformatted button objects filled with the inputs
NOTE: Not fully implemented and tested
"""
button_arr = []
for option in button_options:
button_choice = {
"type": "reply",
"reply": {
"id": "inquiry-yes",
"title": option['text']
}
}
button_arr.append(button_choice)
return button_arr
def create_interactive_message(message_text, button_options, whatsapp_id):
""" Fills a template to create a button message for Whatsapp
* NOTE: Not fully implemented and tested
* NOTE/TODO: It is possible to create other kinds of messages
with the 'interactive message' template
* Documentation:
https://whatsapp.turn.io/docs/api/messages#interactive-messages
Inputs
- message_text: str - the content that the message should display
- button_options: list - what each button option should display
- whatsapp_id: str - the message recipient's phone number
"""
button_arr = create_button_objects(button_options)
data = {
"to": whatsapp_id,
"type": "interactive",
"interactive": {
"type": "button",
# "header": { },
"body": {
"text": message_text
},
# "footer": { },
"action": {
"buttons": button_arr
}
}
}
return data
def pickle_and_encode_state_machine(state_machine):
dump = pickle.dumps(state_machine)
dump_encoded = base64.b64encode(dump).decode('utf-8')
return dump_encoded
def manage_math_quiz_fsm(user_message, contact_uuid):
fsm_check = SUPA.table('state_machines').select("*").eq(
"contact_uuid",
contact_uuid
).execute()
if fsm_check.data == []:
math_quiz_state_machine = MathQuizFSM()
messages = [math_quiz_state_machine.response_text]
dump_encoded = pickle_and_encode_state_machine(math_quiz_state_machine)
SUPA.table('state_machines').insert({
'contact_uuid': contact_uuid,
'addition3': dump_encoded
}).execute()
else:
undump_encoded = base64.b64decode(
fsm_check.data[0]['addition3'].encode('utf-8')
)
math_quiz_state_machine = pickle.loads(undump_encoded)
math_quiz_state_machine.student_answer = user_message
math_quiz_state_machine.correct_answer = str(math_quiz_state_machine.correct_answer)
messages = math_quiz_state_machine.validate_answer()
dump_encoded = pickle_and_encode_state_machine(math_quiz_state_machine)
SUPA.table('state_machines').update({
'addition3': dump_encoded
}).eq(
"contact_uuid", contact_uuid
).execute()
return messages
def return_next_conversational_state(context_data, user_message, contact_uuid):
""" Evaluates the conversation's current state to determine the next state
Input
- context_data: dict - data about the conversation's current state
- user_message: str - the message the user sent in response to the state
Output
- message_package: dict - a series of messages and prompt to send
"""
if context_data['user_message'] == '' and \
context_data['state'] == 'start-conversation':
message_package = {
'messages': [],
'input_prompt': "Welcome to our math practice. What would you like to try? Type add or subtract.",
'state': "welcome-sequence"
}
elif context_data['state'] == 'addition-question-sequence' or \
user_message == 'add':
messages = manage_math_quiz_fsm(user_message, contact_uuid)
if user_message == 'exit':
state_label = 'exit'
else:
state_label = 'addition-question-sequence'
input_prompt = messages.pop()
message_package = {
'messages': messages,
'input_prompt': input_prompt,
'state': state_label
}
elif user_message == 'subtract':
message_package = {
'messages': [
"Time for some subtraction!",
"Type your response as a number. For example, for '1 - 1', you'd write 0."
],
'input_prompt': "Here's the first one... What's 3-1?",
'state': "subtract-question-sequence"
}
elif context_data['state'] == 'exit' or user_message == 'exit':
message_package = {
'messages': [
"Great, thanks for practicing math today. Come back any time."
],
'input_prompt': "",
'state': "exit"
}
else:
message_package = {
'messages': [
"Hmmm...sorry friend. I'm not really sure what to do."
],
'input_prompt': "Please type add or subtract to start a math activity.",
'state': "reprompt-menu-options"
}
return message_package
def manage_conversation_response(data_json):
""" Calls functions necessary to determine message and context data to send
Input
- data_json: dict - message data from Turn.io/Whatsapp
Output
- context: dict - a record of the state at a given point a conversation
TODOs
- implement logging of message
- test interactive messages
- review context object and re-work to use a standardized format
- review ways for more robust error handling
- need to make util functions that apply to both /nlu and /conversation_manager
"""
message_data = data_json.get('message_data', '')
context_data = data_json.get('context_data', '')
whatsapp_id = message_data['author_id']
user_message = message_data['message_body']
contact_uuid = message_data['contact_uuid']
# TODO: Need to incorporate nlu_response into wormhole by checking answers against database (spreadsheet?)
nlu_response = evaluate_message_with_nlu(message_data)
message_package = return_next_conversational_state(
context_data,
user_message,
contact_uuid
)
print("MESSAGE PACKAGE")
print(message_package)
headers = {
'Authorization': f"Bearer {os.environ.get('TURN_AUTHENTICATION_TOKEN')}",
'Content-Type': 'application/json'
}
# Send all messages for the current state before a user input prompt (text/button input request)
for message in message_package['messages']:
data = create_text_message(message, whatsapp_id)
print("data")
print(data)
r = requests.post(
f'https://whatsapp.turn.io/v1/messages',
data=json.dumps(data),
headers=headers
)
# Update the context object with the new state of the conversation
context = {
"context":{
"user": whatsapp_id,
"state": message_package['state'],
"bot_message": message_package['input_prompt'],
"user_message": user_message,
"type": 'ask'
}
}
return context
# data = {
# "to": whatsapp_id,
# "type": "interactive",
# "interactive": {
# "type": "button",
# # "header": { },
# "body": {
# "text": "Did I answer your question?"
# },
# # "footer": { },
# "action": {
# "buttons": [
# {
# "type": "reply",
# "reply": {
# "id": "inquiry-yes",
# "title": "Yes"
# }
# },
# {
# "type": "reply",
# "reply": {
# "id": "inquiry-no",
# "title": "No"
# }
# }
# ]
# }
# }
# }