ML Lab

Interactive machine learning sandbox for experimenting with regression, clustering, and anomaly detection. Load a dataset, preprocess it, train a model, and explore the results — no code required.

Live: https://ml-lab.maxharrison.xyz

Stack

• Frontend — React, Vite, Material UI, Recharts
• Backend — FastAPI, scikit-learn, pandas
• Models — Linear Regression, Decision Tree, Random Forest, DBSCAN, Isolation Forest

---

Features

Data tab — load datasets via example, URL, Kaggle slug, or CSV upload. Column browser shows type, null counts, and per-column stats. Correlation matrix on demand.

Prepare tab — missing value handling (drop or impute), one-hot encoding for categoricals, feature scaling, IQR outlier removal. Detects issues automatically.

Train tab — pick task (regression / clustering / anomaly detection), select features, tune hyperparameters, train and view results. Metrics, scatter, and line plots for regression; PCA 2D plots for clustering and anomaly detection.

---

Project structure

ML-Lab/
  app/                  # original model code (scikit-learn)
    models/
    example_datasets/
    utils.py
  backend/
    main.py             # FastAPI — wraps the models, serves the frontend in prod
    requirements.txt
  frontend/
    src/
      api.js
      hooks/            # useDataset, useTrain
      components/       # DataTab, PrepareTab, TrainTab, charts, ErrorBoundary

---

Running locally

git clone https://github.com/minvoker/ML-Lab.git
cd ML-Lab

Install dependencies first:

cd backend && python -m venv venv && source venv/bin/activate && pip install -r requirements.txt
cd ../frontend && npm install

Then from the root:

./start.sh

App runs at http://localhost:5173, API at http://localhost:8000.