26 Jun 2020

Model Parameters vs Model Hyperparameters

Model Parameters

Configuration variables whose value can be estimated from the data:

  • saved as part of the learned model
  • not set manually

Some examples of model parameters:

  • weights in a neural network
  • support vectors in a support vector machine
  • coefficients in linear regression

Model Hyperparameters

Configuration variables that are external to the model and cannot be estimated from the data. Some examples of model hyperparameters:

  • the learning rate for training a neural network
  • C and sigma for support vector machines
  • k in k-nearest neighbours