Model selection

From GusWiki

Jump to: navigation, search

The problem of model selection is the problem of inferring good models from sets if data as well as principles, for the purposes of inferring the "true model". Being as hard as the problem of induction (?), model selection is an unsolvable formal problem (?), and one can only hope to do well in practice. In the case of stationary processes, future prediction tasks are the way of evaluating model selection.

The idea is to fit observed data while avoiding overfitting.

selecting among cognitive models

See also

Empirical science of science

Machine learning

Occam's razor

Overfitting

Prior selection

Personal tools
Navigation