AI dev assignment - News classification

Create a model for text classification based on given data. Make your code available on a public Git repo.

Data

The data is a subset of News Category Dataset available on Kaggle. It is stored in JSON-per-line (JSONL) format. Important fields:

• category - this is the value that should be predicted by the model
• headline - headline of the article
• short_description - short snippet of text

The data is divided into three files:

• train.jsonl - training data
• dev.jsonl - validation data; hyperparameter tuning is optional for this assignment, manually coming with 'good enough' values is OK.
• test.jsonl - test data for final evaluation

Deliverables

• train.py - training script
  • inputs: train data, (optional) hyperparameter values
  • outputs: serialized model file
• eval.py - evaluation script
  • inputs: serialized model file, evaluation data (in the JSONL format defined above)
  • outputs: accuracy, optionally confusion matrix and precision/recall numbers for each category
• classify.py - classification script
  • inputs: serialized model file, data to classify (the JSONL without category field)
  • outputs: the input JSONL with category field set to model's predicted value
• requirements.txt - list of packages which your scripts depend on in Pip requirements file format
• report.md - your comments about design decisions/tradeoffs you made, observations about the model and data and possible improvements. Bullet points are fine.
• other files at your discretion

Notes

• you can use any model type and any publicly available Python library you consider appropriate
• the scripts should be able to run on a regular laptop
• the scripts should not make API calls through network