Skip to content
BENIBIX Learn

Hyperparameter

Also known as Hyper-parameter, Training Setting

A setting you pick before training. It shapes how the model learns.

Think of it like

Think of it like the oven heat you pick before you bake. The cake does not choose it.

Example

Lucy trains the same model three times, with steps of size 0.1, 0.01 and 0.001. The middle one gives the best results, so she keeps it.

Why it matters

The same data and method can give a good or a bad model, just from these settings. Tuning them is a normal part of the job.

Definition

A hyperparameter is a setting you choose before training starts. It shapes how the model learns, but the model does not learn it. Examples are how big each learning step is, how deep a tree may grow, or how many layers a network has. People try a few values and keep the one that works best. That search is called tuning.
The dials outside the box are hyperparameters. You set them by hand. You do it before training starts.

Common mistake

Tuning the settings by checking the test set. Use the validation set for that. If not, your final score will look too good.

Key takeaways

  1. 1
    Chosen by you. Set before training starts.
  2. 2
    Shapes learning. Not learned by the model.
  3. 3
    Found by trying. Test a few and keep the best.

Compare

Learn it properly