On Wed, Dec 22, 2010 at 3:48 PM, new_user
<address@hidden> wrote:
> Take a look here.
>
http://en.wikipedia.org/wiki/Cross-correlation
> What autocorrelation is doing is sliding (lag) 1 step at a time, a copy of
> the original past the original, and then testing to see if they are the
> same(correlated).
> So if you jump by 10 you could completely miss the important point.
Thanks for cross-correlation explanation. It looks interesting, I haven't met this technique yet. I'll continue with this adventure tomorrow
> From your picture you sent the center point is when they are lined up, and
> yes you can throw away all to the left of this.
> The center point is about 30000 and there is a peek at 40000 so the period
> might be 10000 samples long, But if you just did ever 10'th data point then
> you might have missed the important points.
> I hope this is helpful
> if you have too many data points, would it be valid to sample less often? If
> you use ever 10'th point then it is like sampling at 1/10 of the rate.
> Doug
Picture I posted is perfect symmetrical graph with center in data point number center, which is 1/10 of all data, as I used "every 10". I don't know if I should write about cross-correlation as I just said I didn't know about it, but shouldn't missing data be of minor relevance because we are talking about correlation? I mean are exact data values must for this method?
No exact data values are not needed, but if they are then you can get perfect correlation.
cross-correlation is when you take a smaller set of data and look to see where it (or something close to it) shows up in the bigger data set.