Building the backend - Part II (Live tweet sentiment analysis)
Amal Shaji
Posted on August 14, 2020
In the previous article, I explained how to build a sentiment classifier. Read it here. I this, we'll create the backend to serve predictions via an API.
Building the API
make sure you're in the backend folder
API
# server.py
import nltk
import uvicorn
import utils as utils
from fastapi import FastAPI
from pydantic import BaseModel
from classify import remove_noise
from nltk.tokenize import word_tokenize
app = FastAPI()
classifier = utils.load_model()
class Tweet(BaseModel):
tweet: str
@app.get("/")
def read_root():
return {"message": "Welcome to sentiment classifier API"}
@app.post("/api")
def analyse_tweet(tweet: Tweet):
custom_tokens = remove_noise(word_tokenize(tweet.tweet))
result = classifier.classify(dict([token, True] for token in custom_tokens))
return {"sentiment": result}
if __name__ == "__main__":
uvicorn.run("server:app", host="0.0.0.0", log_level="info")
This is similar to the other articles where I showed how to build APIs using FastAPI. Only change is that, In the previous ones, we launched the server using uvicorn main: app, here we're doing it programmatically. This way we can run the server using python server.py
Dockerize the API
FROM python:3.8-slim
WORKDIR /app
COPY requirements.txt .
RUN pip install -r requirements.txt
COPY . .
RUN python3 -m nltk.downloader punkt
RUN python3 -m nltk.downloader wordnet
RUN python3 -m nltk.downloader stopwords
RUN python3 -m nltk.downloader averaged_perceptron_tagger
EXPOSE 8000
CMD ["python3", "server.py"]
Build and run the image
docker built -t sentwitter-backend .
docker run -d -p 8000:8000 setwitter-backend
Test the API
❯ docker run -d -p 8000:8000 sentwitter_backend
ccd9fa0740cfd0f497308ef570e16c950233c29378029aac61c1479cc94163bc
❯ curl http://localhost:8000
{"message":"Welcome to sentiment classifier API"}
❯ curl -X POST "http://localhost:8000/api" -H "accept: application/json" -H "Content-Type: application/json" -d "{\"tweet\":\"I hate you\"}"
{"sentiment": "Negative"}
We have successfully built the backend. In the next article, we'll build a frontend that retrieves tweets and sent to the backend for prediction.
References
- Article Cover by Taylor Vick
💖 💪 🙅 🚩
Amal Shaji
Posted on August 14, 2020
Join Our Newsletter. No Spam, Only the good stuff.
Sign up to receive the latest update from our blog.
Related
