TweeSent Backend
Tweesent backend. This repo uses fastAPI as the web framework. Tweepy to request from Twitter all the needed data and ONNX to do the inference.
fastAPI
Basically, the backend works with two types of requests. With the HTTP request, the user can ask for a certain amount of tweets inputing a text of an username (starting by @).
It also supports WebSockets connections, so the user can ask for a certain topic and have a live stream of tweets related with its predictions.
ONNX
Models are trained using pytorch-lightning and results are exported to ONNX so they can be easily handled.
At startup, the backend will load the pertinent weights and will do the prediction in the tweets.
Tweepy
Even the Twitter API v2 supports a maximum of 100 tweets per request, the backend can continue where it left using a token. So the first time the user asks for some tweets the backend will answer with the pertinent toker to continue the search later on.
Docker
Best way to start the repo is using… Docker!! There is an awesome Dockerfile (pending to be improved) to start easily the backend.
Pre commits hooks
Once downloaded the repo, and installed the needed packages in requirements.txt… it is desirable to install the git hooks for this repo!
pre-commit install
This will install git hooks such as black, flake8 or mypy.
To-do
- Improve Dockerfile so it supports multi-stage builds.
- Support some way to fetch a model in case there is no one in the weights folder at startup.
