Machine Learning - News Articles classification with sklearn

Classify news articles into different categories using Machine Learning. The dataset consists of 6000 documents and 47 categories.

My goal is to show you how to create a predictive model(s) that will classification labels for news articles.

Full documentation: ml-classifier.readthedocs.io.

Objective

• To classify news articles
• Learn the basics of natural language processing
• Build models using sklearn and choose the best one
• Use sklearn's make_pipeline class
• Learn how to turn it into a service
• Learn how to make it composable and portable
• ...
• Profit?

Prerequisite

• Python >= v3.13
• uv
• Jupyter Notebook
• Some knowledge of Machine Learning

Python Libs

• NumPy
• Pandas
• SciPy
• Matplotlib
• Jupyter
• Scikit-learn (the library that we will use later in this post when creating the classifier model(s))

We Will

• Apply some preprocessing steps to prepare the data.
• We will perform a descriptive analysis of the data to better understand the main characteristics that they have
• We will continue by practicing how to train different machine learning models using scikit-learn. It is one of the most popular python libraries for machine learning
• We will also use a subset of the dataset for training purposes
• We will iterate and evaluate the learned models by using unseen data. Later, we will compare them until we find a good models that meets our expectations, and use a VotingClassifier soft voting for unfitted estimators.
• Once we have chosen the candidate model(s), we will use it to perform predictions and to create a simple web application that consumes this predictive model

Getting started with the machine learning tutorial

See HowTo Jupyter Notebook

Deployment

As a container:

docker run -d -p 7070:7070 docker.io/saidsef/ml-classifier:latest

As a Python application, once the model is in data/, see running the service:

uv sync

uv run classifier-ml.py

On Kubernetes, see deploying on Kubernetes:

kubectl apply -k ./deployment

The Request

The request must be a POST with a JSON payload:

{ "body": "text-goes-here" }

curl -XPOST http://localhost:7070/api/v1/news -H 'Content-Type: application/json' -d @test/test.json

The response names the category and its probability:

{
  "category": "World",
  "score": 0.126
}

Documentation

Published at ml-classifier.readthedocs.io, with the source in docs/.

• Architecture - how the notebook, the image and the cluster fit together
• Training the model - the dataset, the pipeline, and how to run the notebook
• Running the service - the container, a source checkout, and configuration
• API reference - every endpoint, with payloads and responses
• Deploying on Kubernetes - the manifests, the overlays, and what to change in each
• Operations - health checks, metrics, and what the common failures mean
• CI and releases - what CI builds and tests, the image tags, and how releases are cut