You are using an out of date browser. It may not display this or other websites correctly.
You should upgrade or use an alternative browser.
You should upgrade or use an alternative browser.
local extrema detection
- Thread starter ssrrkk
- Start date
Quote from ssrrkk:
if you looked at every single bar that is above it's two neighbors, that would create too many peaks.
Why stop with just the adjacent bars?
If a given high/low is above/below the adjacent bars it is, by definition, a local extrema.
If you look at just the immediately adjacent bars the "strength" of your extrema is 1. By extension, an extreme point that's greater than 2 bars on either would have a strength of 2 and so on.
That logic is the core for pretty much any zig-zag implementation. I'm sure there's at least some pseudo code on the 'Net somewhere.
Quote from jprad:
Why stop with just the adjacent bars?
If a given high/low is above/below the adjacent bars it is, by definition, a local extrema.
If you look at just the immediately adjacent bars the "strength" of your extrema is 1. By extension, an extreme point that's greater than 2 bars on either would have a strength of 2 and so on.
That logic is the core for pretty much any zig-zag implementation. I'm sure there's at least some pseudo code on the 'Net somewhere.
Okay thanks -- this is precisely what is described in the second paragraph of my original post (N-th ordered centered differences). But assigning a "strength" is a neat idea. I will do that.
Yeah, strength is a good idea. I used it when creating horizontal support/resistance lines. For resistance, i needed peaks. To detect peaks i gave it back strength and forward strength. Back strength is how many bars are lower than the high of that bar before it and forward strength is how many bars are lower than the high after that bar. The forward strength is also the length of the resistance. Well, for detecting highs it's a good idea to give them backward and forward strengths and then apply some sort of exponential weight increase to a cumulative "strength" score. This means if it only has backward strength of 1 (meaning there's a higher high only 2 bars before this one), it doesn't add much to the cumulative, if at all. A backward strength of 20 however adds quite a lot. Pick your curve for weighing them, it's subjective after all.
Quote from ssrrkk:
Okay thanks -- this is precisely what is described in the second paragraph of my original post (N-th ordered centered differences). But assigning a "strength" is a neat idea. I will do that.
You're right, your second paragraph had already touched on this. I stopped reading once you mentioned smoothing in your first, my apologies.
Quote from jprad:
You're right, your second paragraph had already touched on this. I stopped reading once you mentioned smoothing in your first, my apologies.
No apologies necessary -- you made me realize how to use those multiple differences: by assigning strengths. That's a very useful idea. Thanks.
I implemented the 5 minute centered difference aka zig zag method -- attached as a png. I also used strength to distinguish between different peaks -- the arrows show the sma crossings following a peak, color-coded by different peak categories. This is the kind of thing I was after. Even though this is using a centered difference, it can still be used as a real time signal but with a 3 minute delay. Anyway, thanks for your suggestions!!