Multi-Class Classification — Save, Load, and Run an AI Model

Training a multi-class classifier is only half the job — you also need to persist it and run it reliably in production. This guide covers the save/load/serve loop.

Train, then persist

Serialize both the model and any preprocessing (scalers, encoders, vocab) so inference reproduces training exactly. Mismatched preprocessing is the most common cause of "it worked in the notebook" bugs.

Load and predict

import joblib

model = joblib.load("model.pkl")
probs = model.predict_proba(X_new)   # one column per class
label = probs.argmax(axis=1)

Serving notes

• Version the artifact alongside the code that produced it.
• Keep preprocessing and model in the same loadable bundle.
• Return class probabilities, not just the top label, when downstream logic needs confidence.

Takeaway

A model is only useful once it runs outside the notebook — treat saving, loading, and serving as first-class parts of the workflow.