Prev: neural network
Next: object recognition
From: matlaberboy on 28 Apr 2010 11:24
I am having problems getting synchronize.m do what I want it to do. Have any users here got any experince with it? I outline the problem I am trying to solve below, but I do not think synchronize.m is able to help with it.

http://www.mathworks.com/access/helpdesk/help/techdoc/ref/synchronize.html

I have two timeseries, with datestamps dd:mm:yy HH:MM:SS. The days do not necessarily match, the start and end points of the timeseries do not necessarily match and when the days do match, the times (HHMMSS) do not necessarily match. The frequency of the observations is not necessairly the same either (eg one could have a daily timestamp and the other could have a obs-timestamp every 30 mins)

What I would like to be able to do is fit one ts to the other;
ts1=myFit(ts1, ts2)
where ts1 is fitted to the timestamps of ts2. Hence both ts1 and ts2 end up with the same timestamps. both ts1 and ts2 end up the same length.

Lets say ts1 is a predictive signal of ts2 (a random variable). The key point being that the timestamps of ts1 can never be moved back in time, (ie can not go 7pm->6pm) but can be moved forward in time, so they can be aligned with the timestamp of ts2. Hence if an obervation in ts1 is at 6pm and the timestamp for that day in ts2 is at 6.30pm, the timestamp in ts1 is shifted to 6.30pm.

The other key point being if you only have daily data (say) and want to fit it to min30 data (say) you will be short of datapoints! Hence you have the option of either filling in with NaNs or replicating the last data observation.

I think this procedure is a really common think to want to do; has anyone tried it before/ can offer any advice?

Many Thanks
 | 
Pages: 1
Prev: neural network
Next: object recognition