Resampling with replacement
in [Matlab]
|
From: Mohammed Khalilia on 1 Dec 2009 09:29 "Mohammed Khalilia" <mohammedsk(a)gmail.com> wrote in message <hf2fan$kv8$1(a)fred.mathworks.com>... > Hi, > I am trying to understand resampling with replacement. Basically, I have a dataset (2000 instances x 2 features) and a weight vector W (2000x1) and trying to produce a bootstrap replicate of the original data according to the probability W for every instance by resampling with replacement from the original dataset. > > I know MATLAB has bootstrap functions, but the closest I found is randsample(), which take as arguments the vector dataset and a weight vector, but I guess it does not work in matrices. > > How can I do that? > > Thanks, Does that make sense to anyone? Or do I need to add additional details?
From: Peter Perkins on 1 Dec 2009 09:45 Mohammed Khalilia wrote: > I am trying to understand resampling with replacement. Basically, I have a dataset (2000 instances x 2 features) and a weight vector W (2000x1) and trying to produce a bootstrap replicate of the original data according to the probability W for every instance by resampling with replacement from the original dataset. > > I know MATLAB has bootstrap functions, but the closest I found is randsample(), which take as arguments the vector dataset and a weight vector, but I guess it does not work in matrices. You haven't exactly defined what you mean by, "work in matrices", but it sounds like you want to use RANDSAMPLE to generate a vector of row indices, and then use that to index into your matrix of data, with a ':' for the column index. Hope this helps.
From: Mohammed Khalilia on 1 Dec 2009 10:07 Peter Perkins <Peter.Perkins(a)MathRemoveThisWorks.com> wrote in message <hf3a6t$laf$1(a)fred.mathworks.com>... > Mohammed Khalilia wrote: > > > I am trying to understand resampling with replacement. Basically, I have a dataset (2000 instances x 2 features) and a weight vector W (2000x1) and trying to produce a bootstrap replicate of the original data according to the probability W for every instance by resampling with replacement from the original dataset. > > > > I know MATLAB has bootstrap functions, but the closest I found is randsample(), which take as arguments the vector dataset and a weight vector, but I guess it does not work in matrices. > > You haven't exactly defined what you mean by, "work in matrices", but it sounds like you want to use RANDSAMPLE to generate a vector of row indices, and then use that to index into your matrix of data, with a ':' for the column index. > > Hope this helps. That is close to what I want, here is what I have so far: a = [1 2;5 6;7 8;34 5]; w = [0.1250 0.1250; 0.1250 0.1250; 0.1250 0.1250; 0.1250 0.1250]; Since randsample takes only vectors for the data and the weight, I had to reshape a a1 = reshape(a,1,8); w1 = reshape(w,1,8); y=randsample(a1,8,true,w1); Then reshape y reshape(y,4,2); This does not really work because reshape will completely change the shape of the data after resampling.
From: Lorenzo Guerrasio on 1 Dec 2009 10:19 "Mohammed Khalilia" <mohammedsk(a)gmail.com> wrote in message <hf3beo$575$1(a)fred.mathworks.com>... > Peter Perkins <Peter.Perkins(a)MathRemoveThisWorks.com> wrote in message <hf3a6t$laf$1(a)fred.mathworks.com>... > > Mohammed Khalilia wrote: > > > > > I am trying to understand resampling with replacement. Basically, I have a dataset (2000 instances x 2 features) and a weight vector W (2000x1) and trying to produce a bootstrap replicate of the original data according to the probability W for every instance by resampling with replacement from the original dataset. > > > > > > I know MATLAB has bootstrap functions, but the closest I found is randsample(), which take as arguments the vector dataset and a weight vector, but I guess it does not work in matrices. > > > > You haven't exactly defined what you mean by, "work in matrices", but it sounds like you want to use RANDSAMPLE to generate a vector of row indices, and then use that to index into your matrix of data, with a ':' for the column index. > > > > Hope this helps. > > That is close to what I want, here is what I have so far: > a = [1 2;5 6;7 8;34 5]; > w = [0.1250 0.1250; 0.1250 0.1250; 0.1250 0.1250; 0.1250 0.1250]; > > Since randsample takes only vectors for the data and the weight, I had to reshape a > a1 = reshape(a,1,8); > w1 = reshape(w,1,8); > y=randsample(a1,8,true,w1); > > Then reshape y > reshape(y,4,2); > > This does not really work because reshape will completely change the shape of the data after resampling. a = [1 2;5 6;7 8;34 5]; w = [0.1250 0.1250; 0.1250 0.1250; 0.1250 0.1250; 0.1250 0.1250]; ind=1:lsize(a,1); n_sample=100; Rind=randsamp(ind,n_sample,true,w); new_sample=a(Rind,:);
From: Lorenzo Guerrasio on 1 Dec 2009 10:32 "Lorenzo Guerrasio" <lorenzo.guerrasio(a)email.it> wrote in message <hf3c57$l13$1(a)fred.mathworks.com>... > "Mohammed Khalilia" <mohammedsk(a)gmail.com> wrote in message <hf3beo$575$1(a)fred.mathworks.com>... > > Peter Perkins <Peter.Perkins(a)MathRemoveThisWorks.com> wrote in message <hf3a6t$laf$1(a)fred.mathworks.com>... > > > Mohammed Khalilia wrote: > > > > > > > I am trying to understand resampling with replacement. Basically, I have a dataset (2000 instances x 2 features) and a weight vector W (2000x1) and trying to produce a bootstrap replicate of the original data according to the probability W for every instance by resampling with replacement from the original dataset. > > > > > > > > I know MATLAB has bootstrap functions, but the closest I found is randsample(), which take as arguments the vector dataset and a weight vector, but I guess it does not work in matrices. > > > > > > You haven't exactly defined what you mean by, "work in matrices", but it sounds like you want to use RANDSAMPLE to generate a vector of row indices, and then use that to index into your matrix of data, with a ':' for the column index. > > > > > > Hope this helps. > > > > That is close to what I want, here is what I have so far: > > a = [1 2;5 6;7 8;34 5]; > > w = [0.1250 0.1250; 0.1250 0.1250; 0.1250 0.1250; 0.1250 0.1250]; > > > > Since randsample takes only vectors for the data and the weight, I had to reshape a > > a1 = reshape(a,1,8); > > w1 = reshape(w,1,8); > > y=randsample(a1,8,true,w1); > > > > Then reshape y > > reshape(y,4,2); > > > > This does not really work because reshape will completely change the shape of the data after resampling. > > a = [1 2;5 6;7 8;34 5]; > w = [0.1250 0.1250; 0.1250 0.1250; 0.1250 0.1250; 0.1250 0.1250]; > ind=1:lsize(a,1); > n_sample=100; > Rind=randsamp(ind,n_sample,true,w); > new_sample=a(Rind,:); sorry,I didn't see that in the example you provide w is a matrix, while in the initial problem w was a vector. The code I wrote before work if w is a vector, i.e. if you write Rind=randsamp(ind,n_sample,true,w(:,1)); If, instead, the two columns of a have different probabilities you can split matrix a in two vector a1 and a2 and matrix w in two vector w1 and w2, than resample separately.
|
Next
|
Last
Pages: 1 2 Prev: mincx to gevp Next: Converting Binary images/Grayscale images to move frames |