4

This is my first post on Cross-Validated. I read a lot of question related to my problem, but no one was completely satisfying.

I have two time series that are sampled at very different time intervals, e.g. one is sampled every 10 seconds, while another every day. I'd like to evaluate the correlation between the two series (perhaps something like cross-correlation).

Is it right to align the one series to the other (maybe averaging the value of the one with the finest sampling rate)? Is there any common practise in treating this problem? Is there any Spark-based library that can be used to easily deal with such problems (with Java support, as sparkts)?

geekoverdose
  • 3,691
  • 2
  • 14
  • 27
McKracken
  • 51
  • 3
  • Which relation are you trying to discover? Sec. vs day does not seem intuitive (think of sec. vs day on a e.g. acceleration, temperature, sales, etc. -basis). Should there be similar pattern in short and long term? – geekoverdose Jun 06 '16 at 17:51
  • 2
    The nature of the sample--what time interval it represents and how it represents it--is essential to know. For instance, if both time series give instantaneous readings of something, that leads to one solution; but if (say) the daily time series gave averages within 24-hour windows, that would lead to a different solution. (This is the one-dimensional "change of support" problem.) What can you tell us about your data in this regard and what can you say about *why* you want to compute the cross-correlation? – whuber Jun 06 '16 at 18:37
  • I have data from various sensors. Every sensor gives one value, but the survey is done at a different time interval: they can be one every 20 seconds, while another one every minute (someone every day). I might have bus delays at every stop (given every bunch of minutes, let's say 5) and amount of rain during the day (given about every 30 seconds - I live in a really rainy country). So I might want to find how much the two events are correlated. Moreover, time intervals are not exact (if it's about 30 seconds, it can be 30 like 32, sometimes). – McKracken Jun 07 '16 at 11:04
  • Any suggestion? – McKracken Jun 13 '16 at 09:10

0 Answers0