每週相同時間必會發生的情景。週四凌晨，早上10點的微積分作業死線就快到了，這時寫作業(尤其是積分)的時候算到卡住，想破頭也想不到怎麼積，卻看著隔壁的室友神來一筆代換就算出答案，然後就去睡了。這時心裡必然相當苦惱，有點想找個人聊聊，如果這個人能幫我完成微積分作業那就更完美了。於是LINE的聊天機器人 --- Calculuseal(微積海豹)現身來解決問題了。
用LINE傳訊息給Calculuseal
- 傳圖片：用Mathpix解析圖片中數學式後，回傳計算結果(圖片)
- 如果失敗，叫用戶再傳一次
- 輸入簡單數學式子：微積海豹會用文字回傳答案，可能新增額外功能：
- 畫函數圖形
- 天氣
- 單位換算
- 匯率換算
- 問候：內建罐頭回應
- 其它：回傳功能說明
分為三部分：
- LINE Bot API
- Mathpix API
- Wolfram Alpha (web service)
- 申請LINE開發者帳號：https://developers.line.biz/en/
- 用虛擬主機建立webhook，將Python code丟上去當LINE BOT的後臺
- free virtual host
- Azure: https://azure.microsoft.com/zh-tw/free/
- Amazon: https://aws.amazon.com/tw/cloud9/?origin=c9io
- Heroku: https://www.heroku.com/
- DigitalOcean: https://www.digitalocean.com/
- 選擇最方便使用的heroku
- 用git push程式碼就可以運作
- 支援HTTPS
- ngrok: generate temperate public URL
- free virtual host
- 罐頭回應：在 https://admin-official.line.me 做設定
- API (Python)
MathPix提供API，可以讀取上傳的圖片並解析其中的數學式成JSON格式，支援轉換成wolfram格式。
- https://cn.docs.mathpix.com
- https://docs.mathpix.com
- https://github.com/Moe-Net/Mathpix
- https://www.topbestalternatives.com/mathpix/
輸入符合wolfram格式的數學式，計算並回傳結果(圖片)
- https://www.wolframalpha.com/
- 用BeautifulSoup把結果抓下來
gantt
title Project Calculuseal
axisFormat %m-%d
section Main Task
setup environment :a1, 2018-12-17, 7d
Complete LINE webhook and backend code :a2, after a1, 7d
Complete Mathpix part :a3, after a2, 7d
Finish and intergrate all functions: a4, after a3, 7d
12/24：架好環境 12/31：完成LINE的webhook和後臺程式碼 1/7：完成Mathpix部分 1/17前：完成並整合所有功能
Tutorial: https://ithelp.ithome.com.tw/users/20107309/ironman/1253 Notice: https://engineering.linecorp.com/zh-hant/blog/line-device-10/ ChatBot: https://www.appcoda.com.tw/line-chatbot-sdk/ linebot: https://www.oxxostudio.tw/articles/201701/line-bot.html
- LINE is now able to correctly forward message to and from this server
- hosted by Heroku
- using Django framework to server request
- Mathpix(TM) for OCR, finding mathematical equations in a picture
- WolframAlpha(TM), for calculating given mathematical equations and producing result
- both are called directly with web request (API), and results are in JSON format, which is friendly for Python to parse
- nowhere to put BeautifulSoup in use
- some bugs
- only replying the first result of the equations
- Heroku goes sleeping after a while without requests
- images are stored on server as database entry, which could be a loss on performance with heavy loading
- for improvement, use Cloud Service (AWS S3 for example) and store image as statics files
- more types of replies
- stickers
- audio
- video
- answers to simple questions
- joke
- weather
- mood
- simple calculation (as text message)
- notify developer (me) whenever new member follows
- only replying first image of results, or replying the same image repeatedly
- timestamp not precise enough
- add 3 more digit (milliseconds) to conquer
- break 5 reply messages limit
- push messages instead of reply message
- remove LINE secure token for safety
- notify developer (my LINE account) whenever new member follows
- record new friends and backup text messages into database
- we build a backend web management interface: https://calculuseal.herokuapp.com/admin/
- some information is concealed in case of privacy issues
- search Yahoo!(R) dictionary when user sends a English word or phrase
- using BeautifulSoup to fetch the result
- simple calculation when user sends equations
- e.g. 8 / 3 + 9**2
- must be in python syntax
- eval()
only showing essential files
+ 0710029/
+ calculuseal/ # virtualenv dir
+ requirements.txt
+ ngrok
+ Procfile
+ bin/
+ pip
+ activate
+ include/
+ lib/
+ (necessary dependencies)
+ .git/
+ calculuseal/ # django project
+ manage.py
+ calculuseal/
+ settings.py
+ urls.py
+ webhook/ # django app
+ views.py
+ admins.py
+ models.py
+ apis/
+ mathpix.py
+ wolfram.py
+ yahoodict.py
+ config/
+ line.py # LINE secure tokens
we support two ways to launch the app:
- with ngrok on local side, for easy debugging
- upload to Heroku server, real webhook of LINE BOT
before running each of them, run the following command to enter python virtual environment (all dependencies is installed already):
on Unix-based OS: source bin/activate
on Windows: bin\activate.bat
However, in case of security issues, we will not give you login information of either our LINE developer account or our Heroku account. But that won't be significant hindrance, since method 1 is sufficient to test the functionality of our app. If you should want to test the app with method 1, follow instructions in the presentation "linebot.pptx".
ngrok provides a public domain that forwards all HTTP/HTTPS requests to a local port (such as 127.0.0.1:8000)
first run the server locally:
calculuseal/calculuseal/manage.py runserver 127.0.0.1:8000
then register a ngrok account, where authentication token will be given:
https://ngrok.com
calculuseal/ngrok authtoken <authentication_token_of_your_account>
calculuseal/ngrok http 8000
finally, register a LINE developer account (choose developer trial plan): https://developers.line.biz and set LINE channel webhook URL to: https://<some_random_token>.ngrok.io/webhook/ the URL will be shown after ngrok launches
git remote set-url heroku https://git.heroku.com/calculuseal.git
git push heroku master
and set LINE channel webhook URL to: https://calculuseal.herokuapp.com/webhook/
to watch logs:
heroku logs -t
- if a message cannot find its sender in Friends model, add that friend into Friends model
- HTTPS redirection
- broadcast message from admin site
- fix problem that admin site doesn't serve static file
django&heroku: http://djangogirlstaipei.herokuapp.com/tutorials/deploy-to-heroku/?os=windows
django-admin startproject calculuseal
./manage.py startapp webhook
python3 -m virtualenv calculuseal
cd calculuseal
source bin/activate
pip3 install django dj-database-url gunicorn dj-static psycopg2
pip3 freeze > requirements.txt
heroku login --interactive
heroku create <myapp> --buildpack heroku/python
heroku buildpacks:set heroku/python # https://devcenter.heroku.com/articles/buildpacks
heroku config:set DJANGO_SETTINGS_MODULE=calculuseal.settings
# in production stage:
# heroku config:set DJANGO_SETTINGS_MODULE=calculuseal.production_settings
echo 'web: gunicorn --pythonpath calculuseal calculuseal.wsgi' > Procfile
heroku ps:scale web=1
heroku run python calculuseal/manage.py migrate
heroku run python calculuseal/manage.py createsuperuser
heroku logs -t
git filter-branch --force --index-filter 'git rm --cached --ignore-unmatch PATH-TO-YOUR-FILE-WITH-SENSITIVE-DATA' --prune-empty --tag-name-filter cat -- --all
heroku git:remote -a calculuseal
git remote -v
output:
> heroku https://git.heroku.com/calculuseal.git (fetch)
> heroku https://git.heroku.com/calculuseal.git (push)
> origin https://github.com/soyccan/calculuseal.git (fetch)
> origin https://github.com/soyccan/calculuseal.git (push)
git push heroku master
if such error is encountered:
> Permission denied (publickey).
> fatal: The remote end hung up unexpectedly
then:
heroku keys:add
sudo -u postgres createuser test
createdb test
psql -U test test
- https://devcenter.heroku.com/articles/heroku-postgresql
- https://docs.postgresql.tw/server-administration
heroku addons:create heroku-postgresql:hobby-dev
heroku pg:info
watch heroku pg:info
> === DATABASE_URL
> Plan: Hobby-dev
> Status: Available
> Connections: 0/20
> PG Version: 10.6
> Created: 2018-12-24 10:28 UTC
> Data Size: 7.6 MB
> Tables: 0
> Rows: 0/10000 (In compliance)
> Fork/Follow: Unsupported
> Rollback: Unsupported
> Continuous Protection: Off
> Add-on: postgresql-rugged-56733
heroku pg:credentials
export DATABASE_URL=postgres://$(whoami)
# pull database
heroku pg:pull HEROKU_POSTGRESQL_MAGENTA mylocaldb --app sushi
# or with username and password
PGUSER=postgres PGPASSWORD=password heroku pg:pull HEROKU_POSTGRESQL_MAGENTA mylocaldb --app sushi
# access local database
heroku pg:psql postgresql-rugged-56733 --app calculuseal
# push database
heroku pg:push mylocaldb HEROKU_POSTGRESQL_MAGENTA --app sushi
# for troubleshooting
heroku pg:ps
heroku pg:kill
heroku pg:killall