0
$\begingroup$

We are setting up an experiment for a model that is able to predict the evolution of a time series in different horizons. One of the parameters to decide is the granularity of frequency of our samples (every 1/5/10 mins). My question is: does higher frequency sampling provide better result as a general rule?

$\endgroup$
1
$\begingroup$

You are limited by the Nyquist frequency, which is a theoretical frequency. Let's suppose you want to capture a phenomenon that occurs at $60\,\text{Hz},$ such as standard electrical power from your wall output. The Nyquist frequency says that, theoretically, you should sample at $120\,\text{Hz}$, twice the phenomenon frequency, in order to recreate your signal.

However, in theory there's no difference between theory and practice, but in practice, there is. The usual rule-of-thumb in the data acquisition world is to sample at ten times the frequency of the phenomenon you want to capture. In the example above, therefore, you'd really want to sample at $600\,\text{Hz}$ to get a good picture for your signal.

Higher sampling rates will generally produce a better representation of your phenomenon. The flip side, of course, is that you produce more data, which makes analysis slower. There's definitely a law of diminishing returns going on here, and sampling the living daylights out of your system is unlikely to help you much unless you're in the RF world.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.