Project Description

This is a basic, ML logistic regression training/serving http stack using flask to deliver the api, celery and redis to train and temporarily store trained model information.

This api comes with a couple of endpoints that could be useful, and here they are:

POST /models

• Endpoint description: Training a new logistic regression model. Post training data,
• Expected fields in post request:
  • Content-Type = 'text/csv'
  • data field contains training data in csv format with no header.
• Response Structure:
  • 201 Success: {"message": "You're model is being trained right now, you can check it's status by hitting the check_task endpoint", "model_id": "some model id", "task_id": "some task id"}
  • 415 Data Format wrong: Please post data in csv format, and make sure the label column is the last column in the csv.'

GET /models/:model_id

• Endpoint description: Making predictions with data provided in the get request. This endpoint supports both single predictions and batch predictions.
• Expected fields in post request:
  • Content-Type = 'text/csv'
  • data field contains training data in csv format with no header.
• Response Structure:
  • 202 Accepted: {"predictions": [some class], "prediction_probability": [[some class probabilities...]]}
  • 200 Model Still Training: "Model is Training"
  • 404 Model not Found: "The model id you supplied either doesn't exist or the model_id provided is wrong."
  • 415 Data Format wrong: Please post data in csv format, and make sure the label column is the last column in the csv.'

POST /check_model_status/:model_id

• Endpoint Description: Check the status of your model using the model id provided in the training endpoint response.
• Response Structure:
  • 200 Model Exists: Response will either be "Model is Training" or "Ready to Use", depending on the state of the model.
  • 404 Model not found: "The model id you supplied either doesn't exist or the model_id provided is wrong."

Setting up Dev Environment

If you'd like to run this locally, you can! First, create a new python virtual env using python 3.6. Then pip install requirements.txt to get all your dependencies installed. Then cd into the ml_api folder.

starting redis

redis-server /usr/local/etc/redis.conf check with redis-cli ping. If you get a "PONG" back, your redis is up and ready to go.

starting a celery worker

celery -A tasks worker --loglevel=info

starting the flask server

python app.py

gunicorn --bind 127.0.0.1:5000 --workers 1 app:app

Running with Docker

Docker makes it a little easier to run the full app environment anywhere. If you want to run the app on Docker, cd into api_ml and make sure you have docker and docker compose installed. Then run docker-compose up -d --build You'll be able to hit the endpoints at localhost:5000.

Using the API

Sending a post request with postman (just an example)

Set a post request to http://localhost:5000/models. Set content-type in headers to 'text/csv'. In the body, click on raw and set the data type to text. You can copy and paste the csv input here. Then push send and watch the magic.

Making a request for predict not with postman

Somehow, postman doesn't allow you to attach a body to a get request. It's okay, you can do it programmatically in your favorite programmatic method. If you're running locally, you can hit the endpoints with localhost:5000.

If you're running with docker, things can get a bit complicated. If you have ::1 localhost in your /etc/hosts file, then you can also curl or programmatically hit localhost:5000. If not, it's okay, you can hit [::1]:5000.

Here's an example of a python line which will send a get request to [::1]:5000.

import requests
model = 'model_20181025-132027' #enter a model id here that you know exists
response = requests.get('http://[::1]:5000/models/%s' % (model), headers = {'Content-Type':'text/csv'}, data = '6.9,3.1,5.1,2.3')
print(response.content)
print(response.status_code)

An example of training from python

import requests
import json

with open('iris.csv', 'r') as myfile:
  data = myfile.read()

response = requests.post('http://localhost:5000/models' % (model), headers = {'Content-Type':'text/csv'}, data = data)
print(response.content)
print(response.status_code)

Running Tests

If you are in the ml_api folder, you can run: python -m unittest discover