Uptain Data Science Coding Challenge

It seems like you're trying out for a position at Uptain or you've found this and would like to apply. We're excited to see your creativity and skills in action — we 💚 those things at Uptain!

Your goal is to build an ML model that can detect the age of a person based on their email address. Once you've completed the challenge, please create a Pull Request and we will get in touch. 🤙

Fork this repo and get started 🥷

Brief

This repository contains a file emails.txt, which has an unsorted list of emails. Each email has a possible association with an age based on different attributes. Your task is to find these attributes in the emails and build a model that can predict the age of a person based on their email address.

Technology Selection

It is up to you to select your stack. Feel free to choose the one that enables you to complete the challenge.

• You can use any libraries, task runners, or frameworks you like; however, we expect the solution to be written in Python.

Requirements

• The output of the model must produce a single JSON line like:

  • { "age": "{age_class}", "score": {score_value} }

  For example:

  1. { "age": "young", "score": 1 }
  2. { "age": "medium", "score": 0.5 }
  3. { "age": "old", "score": 0.75 }
  4. { "age": "unsure", "score": 0 }

  Where age can be one of four options:

  • young - a person is relatively young (18-30)
  • medium - a person is middle-aged (30-50)
  • old - a person is old (50+)
  • unsure - the age can't be determined

  The score should be a float value between 0 and 1, where 1 is the most confident prediction and 0 is the least confident prediction.

• Please provide a description of your solution and the decisions you made in the README.md file.

  • This must include the method of finding the attributes in the emails and the model training process you used to predict the age.
  • And a guide of how to start the ML model from terminal, correctly provide input and receive an output.
  • You can also include any additional information you think is relevant, possible the minimal RAM and CPU requirements, etc.

GitHub

• How to fork a repository
• How to create a pull request from fork

Disclaimer

This repository contains a list of generated test emails. Any real match with existing emails is purely coincidental and unintentional. All the emails here were generated for testing purposes only.

License

This project is licensed under the MIT License - see the LICENSE file for details.

---

---

---

Solution

I like to emphasize that this is not a supervised learning problem. Altough some emails have birth year wihtin them that can easily be extracted as label, we do not have labels for all the emails. It is a grave mistake to train a model with those labels (This leads to overfitting and data leakage) and use this model for the rest of the emails or real world scenarios.

For a comprehensive explanation of the methodology and implementation details, please refer to the notebook named "solution" in the "solution" directory.

Requirements

This solution is pretty lightweight and does not require specific CPU or RAM constraint.

Installation

First navigate to the solution folder.

: cd solution 

Then make an virtual environment.

: python -m venv .venv 

Depending on the OS, switch to the environment. Here is for Windows:

: .venv\Scripts\activate

Install requirements

(.venv): pip install -r requirements.txt

Inference

The main entry point for inference is main.py. The module requires one of two argumants: --file or --email. The file format must be a text file just like what was intially provided.

-f FILE, --file FILE  Path to a text file with emails.
-e EMAIL, --email EMAIL Email address for inference. i.e. '[email protected]'

Example for an email

(.venv): python main.py -e [email protected]
{'email': '[email protected]', 'age': 'unsure', 'score': 0.68}

Example for a text file

We can use the provided emails.txt file located in root directory of this repository. Since we are in solution directory, we must supply the file as follow:

(.venv): python main.py -f ../emails.txt 
Started prediction ...
Prediction done.

Depending on your hardware, the inferance may take between 1 to 3 minutes. After completion you can check result.json file.