Recognize the Task Type

Different questions need different outputs

Classification predicts a category, such as fraud/not-fraud or one of several ticket queues. Regression predicts a numeric quantity, such as delivery minutes or power consumption. Clustering groups similar examples without known answer labels. Ranking orders candidates, as a search engine orders documents. Generation produces a new sequence such as text. These task types are not interchangeable because their training signals and evaluation metrics differ.

question                                  task
Will this job fail?                       classification
How many seconds will it run?             regression
Which machines behave alike?              clustering
Which runbook should appear first?         ranking
Write a concise incident summary           generation

A supervised task learns from examples carrying known labels; classification and regression usually fit here. An unsupervised task looks for structure without an answer column; clustering is the standard example. Self-supervised learning creates training targets from the data itself - hiding a word and asking a language model to predict it is one form. Reinforcement learning learns behavior from rewards produced after actions, a different feedback setup from ordinary labeled rows.

Warning: Turning every problem into generation can hide simpler, safer solutions. If the required output is one of four fixed categories, a classifier or even validated rules may be cheaper, faster, and easier to test than free-form text.

Scenario: A team describes ticket routing as text generation because the input is text. The required output is actually one of five queues, so supervised classification gives a constrained answer that is easier to score and integrate.