Time Series

Hi,

a simple use case for lagging is to find or verify periodic patterns in the data. E. g. if you have daily data, lag by 1-10 days, and at 7 days you see a large correlation, then you have a weekly period. Similarly with monthly data and -12 months etc.

Lagging in smaller amounts gives you a base level where you can expect new values to be. A good model would then find that e. g. lag - 1, lag - 2 and lag - 7 are the best predictors in a scenario. (Or different ones.)

With lagging you can also express other things, like a relative change from day to day (in percent) in addition to the absolute one.

Regards,

Balázs

Balazs,
from the following where is it used: tf-idf word vector, time series feature generation, SVMs and sentiment analysis scoring.

my thinking was it is used on Time series feature generation, but from your explanation you did not mention anything about feature generation.

Hi @fungayism ,

lagging creates new attributes (=features), so it is a form of time series feature generation.

Regards,
Balázs
 

Again on on same subject time series. What is periodogram kernel estimator plus svm AND windowing plus regression learner used for?

Hi @fungayism, It is as what you thinkg: Time series feature generation.
In fact it generates the new feature for time series by using the "lag" time.
New Feature = Old Feature+5 (days), for example