def jittering -- for model stability

rasbt / pattern_classification

A collection of tutorials and examples for solving and understanding machine learning and pattern classification tasks

GNU General Public License v3.0

4.15k stars 1.29k forks source link

def jittering -- for model stability #39

Closed vmirly closed 10 years ago

rasbt commented 10 years ago

Oh, nice one, thanks! Never heard of this before! So the purpose is to measure how "certain" your original model is? Basically if you add certain noise to different models, and the model before and after the noise is almost "similar" then it is a good model, otherwise it is overfitted?

vmirly commented 10 years ago

yes! Overfitting is one source of instability! Sometimes stability is data dependant!

As I am reading the literature, there are several methods to measure stability of a statistical model! You can also use bootstrapping, subsetting, and adding outliers to see if your model is stable or not!