Systematic Resampling Algorithm

So here I’m showing you that actual
wagon-wheel algorithm that we just talked about,
this systematic resampling. You start off with no new samples, and
you build your, the c’s are going to represent the outer ring,
the cumulative distribution function. And, essentially, each one represents the one before
it plus the extra new weight. Right?
So, you start off at the first weight, and then c2 is the sum of w1 and w2. C3 is the sum of w1, w2 and w3. That’s that outer ring, okay? You can think of it,
that if it goes from zero to one, because everything sums to one,
then the c’s are sort of how far, what percentage around
the clock have you gone? The next thing you do is you have
to pick, remember I told you, you, you have to sort of randomly throw
down the wagon-wheel to start with. Well if the wagon wheel
has eight spokes, the first spoke has to go somewhere
between zero and one-eighth. That’s what this is right here, right? N to the minus 1, 1,
remember you could write it as 1 over n. That’s the offset. And then what you do is you go around
the outer ring seeing, wait, did I pass my offset yet, as soon as I’ve passed
it, that’s the cdf part that I’m in. So just drawing that on here,
if this my first one, right? And I, and, and this is where it
starts on the outer ring I say, did I pass it yet? I say, nope. ‘kay, did I pass it yet? Yes, I did.
So this first spoke landed in w2. Then I just go to the next spoke, and
I say, did my last marker pass it? No, I gotta go to the next one. Did this pass it? Yes, so my next one is from w3. So that’s all I do, is I, I initialize
it and then I just count up for however many samples that I need,
I go and I grab the next wagon-wheel chunk and
I see where on the cdf that it lands. And then I just return that set. It, it looks more
complicated than it is and it’s a way of doing that sampling in,
in linear time.

Tags: , ,

There are no comments yet

Why not be the first

Leave a Reply

Your email address will not be published. Required fields are marked *