คาดการณ์จากข้อความใหม่เอี่ยม

Training is done. Now comes inference: feeding brand-new, unseen text to your saved model and reading back its prediction. 🚀

Start every prediction script by loading the saved pipeline. From here it behaves exactly like the model you trained.

The predict method takes a list of texts, not a single string. Even one example must be wrapped in a list.

Pass a whole list to score many texts in one go. Batch prediction is far faster than looping one item at a time.

Want confidence, not just a label? Call predict_proba to see how sure the model is about each class.

New text must go through the same cleaning as your training data. The pipeline handles this, which is exactly why you saved it whole.

Words the model never saw are simply ignored by the vectorizer. Your vocabulary is fixed at training time, so prediction stays stable.

Models often return numbers like 0 and 1. Turn them into a readable label name so people can understand the output.

Wrap loading and predicting in one helper function. Now any part of your app can classify text with a single clean call.

Real text changes over time. When accuracy slips, that is data drift, a signal it is time to retrain on fresh examples.

Your classify function can sit behind a web API or a script. The same saved model now powers real predictions in production. 🎯

Think about the input format predict requires.

You loaded the model, predicted on new text in batches, read probabilities, mapped labels, wrapped it in a function, and watched for drift. 🏁