Journal ML Take Home exercise

In this exercise, you'll be creating a REST API service that serves an ML model that you'll create to solve the following problem:

Based on the contents of a news article, we want to be able to predict if the article was written in one of the following publications:

• NPR
• CNN
• Washington Post
• Reuters
• Guardian
• New York Times
• Atlantic
• Business Insider
• Vox
• Buzzfeed News
• Talking Points Memo

To assist with training a model to perform this, we are providing gold data in this file: train.csv. The file has the following columns:

column	description
title	The title of the article
publication	The name of the publication
date	The publication date in the following format "YYYY-MM-DD" (eg. 2016-12-06)
year	The publication year
month	The publication month
content	The article body text

The service you create should respond to the endpoint /predict_publication, where users can POST JSON data in the following form to get a prediction:

{
  "title": "FILL WITH TITLE OF ARTICLE",
  "content": "FILL WITH CONTENT OF ARTICLE"
}

and expect a response in the form:

{
  "publication": "FILL WITH PREDICTED PUBLICATION NAME"
}

An example request:

{
  "title": "How To Teach A Sea Lion Who’s Fussy About Grammar",
  "content": "When it comes to sentence structure, Rocky, a sea lion, was a 
  stickler. ”It really mattered to her, what’s going to be the direct and 
  indirect object,” says Kathy Streeter, an animal trainer. For Sierra, it 
  isn’t the grammar that interests her. It’s the vocalizations. This California 
  sea lion loves experimenting with her vocal range, and she hates being 
  interrupted. More than 1 million people visit the New England Aquarium in 
  Boston each year. Before walking through the front door, they watch Atlantic 
  harbor seals play. Inside, visitors watch sea lions cruise around the pool. 
  What these visitors may not know is that the aquarium’s 10 seals and two sea 
  lions go to school each day Streeter is one of their teachers. Rushing between 
  classes, Streeter can rattle off the disposition of each student. Cayenne, a 
  harbor seal, has a ”hard time totally, totally focusing.” Sierra is smart and 
  eager. She’s like: ”Come on! Let’s go! Let’s move fast! You’re taking too long.” 
  Zoe, a sea lion, isn’t great at transitioning between tasks. With such different 
  personalities, what’s the classroom dynamic like? ”They get along really, really 
  well.” But Streeter rarely works with the whole student body. Instead, most of her 
  lessons are taught   . Or, to put it in language that would sound familiar to just 
  about any   teacher, she has an individualized learning plan for each of her 
  students. With rubber boots and waterproof pants, Streeter starts her lesson with 
  Chucky  —   short for Chacoda  —   by asking slowly and clearly, ”How are you?” 
  In a rush of excitement, the harbor seal answers in a garbled voice, ”How are you. 
  How are you!” ”Good boy! Bubbles?” she asks. Much to the delight of the tourists 
  watching, Chucky dives underwater and starts blowing."
}

and the corresponding response:

{
  "publication": "NPR"
}

Expected results

Model

Feel free to use any language of your choosing to train your model, and feel to use any external libraries to help you solve this problem. We're more interested in the following:

• how do you evaluate performance of your model/system
• get a working, baseline model with acceptable performance
• how/why did you choose the components of your prediction system
• what else would you try if you had more time

Please include a written document explaining the above.

Service

Please package up your service with Docker to make it easier to run. This can just be a Dockerfile included in your submission that would contain the service when built. An example submission might include the following files:

Dockerfile
app.py: The service implementation
model.pkl: A binary file containing the prediction model
README.md: A document describing the solution

Submission

Please don't spend more than 2-3 hours on this problem. As mentioned previously, we're more interested in your ability to get a working solution to the problem and your thought process around what can be tried next to potentially improve it.