NLP Engineering Interview

File Description

app.py: Simple Flask server and API that exposes a single end-point /sentiment [POST] that uses the model to predict the sentiment of text passed to it.
rest.js: Node app and API, please set LOCAL_FLASK_SERVER variable before running. For example: const LOCAL_FLASK_SERVER = 'http://10.0.0.233:5000/sentiment';. This server exposes multiple end-points:
- /: Just a sanity-check to quickly check server is running
- /getsentiment_nlpjs: Do sentiment analysis using built-in NLP.js module
- /getsentiment: Do sentiment analysis by talking to Flask server
sentiment_clf.pkl: Pickled log-reg trained classifier/model.
sentiment_vec.pkl: Pickled TF-IDF vectorizer used to train above model.
Kea.ai.ipynb: Jupyter notebook with analysis.

Install all dependencies (npm or pip) - don't forget the ones in the first cell in the Jupyter notebooks.
Start Flask server: flask run --host=0.0.0.0
Start Node server: node rest.js
Use Postman or cURL to HTTP POST to x.x.x.x:8080/getsentiment with JSON payload containing one field/key called "text". For example {"text":"I really hate this place, terrible food and terrible service!"}
Get back JSON response with sentiment. Example: `{"sentiment": "neg"}

I wanted to show off a bit of NodeJS skills so added some extra functionality to "show off". I know it's not complex but a little more signal for you to consider.
All pickled files use protocol level 4 so should be usable by Python 3.4+
You can download the final dataset I used here: https://s3-us-west-2.amazonaws.com/ml-data.avital.ca/all_useful_reviews_tokenized_w_sentiment.pkl.gz
I have a Google Doc file with my assumptions and reasoning that I used during the presentation. I'd be happy to share it if you'd like.