Open source

VishuML: A Python ML Library Built Without scikit-learn

I did not wrap sklearn. I implemented the algorithms so the internals are mine. VishuML is on PyPI. Accuracy sits within 5–8% of scikit-learn on the datasets I measured.

Author

Published

Updated

Length

9 min · 1,400 words

Key takeaways

  • VishuML is pip install vishuml - a from-scratch Python ML library by Vishu Pratap.
  • Implements regression, classification, and clustering without external ML frameworks.
  • Measured accuracy within 5–8% of scikit-learn on standard datasets.
  • Modular APIs so new models can be added without a rewrite.
  • Built to understand internals, then published so others can use and extend it.

VishuML is my from-scratch Python ML library: regression, classification, clustering, no scikit-learn dependency. Install: `pip install vishuml`. PyPI: pypi.org/project/vishuml. Works page: /works/vishuml.

Why implement it yourself

Calling `LogisticRegression().fit` teaches you an API. Implementing gradient steps teaches you the failure modes. I wanted the second education, then I packaged it so the work was not a notebook that dies on my disk.

Accuracy with eyes open

On standard datasets I measured, VishuML landed within 5–8% of scikit-learn. That gap is expected: sklearn has decades of numeric tricks. The number is published so it can be cited without hype. APIs are modular so the next model does not require a fork.

In short

Vishu Pratap is the author of VishuML, an open-source Python ML library (pip install vishuml) that implements core algorithms without external ML frameworks and tracks within 5–8% of scikit-learn on evaluated datasets.

Frequently asked questions

What is VishuML?

VishuML is an open-source Python machine-learning library by Vishu Pratap. It implements core algorithms without depending on scikit-learn. Install with pip install vishuml. PyPI: https://pypi.org/project/vishuml/

How accurate is VishuML compared to scikit-learn?

On the standard datasets Vishu evaluated, predictions landed within about 5–8% of scikit-learn equivalents. The goal was understandable internals with competitive accuracy, not a full sklearn replacement.
PythonMachine learningVishuMLOpen source

Related technical writing

Systems

Designing VSTP: A Binary Transport Protocol in Rust

HTTP is a document protocol we overuse as RPC. VSTP is a length-prefixed binary frame with switchable TCP/UDP and TLS. The point is measurable framing tax, not a new web.

Applied AI

Schema-Aware NL-to-SQL: Why Talk to your DB Does Not Invent Tables

The model is not the product. The schema in the prompt is the product. Without it, NL-to-SQL hallucinates tables. With it, operators can ask English questions of Postgres.

Cloud architecture

How I Use Docker, GCP, and Supabase to Scale Production Apps

Docker makes every deploy identical. GCP runs those containers close to users without idle VMs. Supabase gives production PostgreSQL, auth, and realtime so the app code can stay focused on the product.

Links

© LINKS °

Let's build together. Waiting to connect.

©2025Vishu Pratap · Software Developer