Re: [Gretl-users] Loop recursion: Speed comparison with Matlab

On Tue, 24 Oct 2017, Allin Cottrell wrote: [...] I took a different approach, which avoids loops completely. If I understand the logic correctly, basically resampling is done by putting together a series of "runs" of random length (b on average) which start at random points between 1 and n. So I re-wrote the function as follows: <hansl> function matrix SB2 (matrix x "Data to be resampled", int b[0::] "Block size") scalar n = rows(x) if n == 0 funcerr "data input error, check data" endif if b==0 # if required set mean block size to rule-of-thumb value b = round(1.75*(n^(1/3))) endif p = 1/b # probability of a new run s = seq(1,n)' # sequence u = 1 | (muniform(n-1,1) .< p) # run starts r = cum(u) # id of each run nr = r[n] # how many runs ? sub = selifr(r ~ s, u) # starting row for each run sub[,2] -= mrandgen(i, 1, n, nr, 1) # adjust starting points # --- create mini-trends ---------------- s -= replace(r, sub[,1], sub[,2]) # roll over if necessary s = ((s-1)%n) + 1 return x[s,] end function </hansl> This appears to be quite fast. ------------------------------------------------------- Riccardo (Jack) Lucchetti Dipartimento di Scienze Economiche e Sociali (DiSES) Università Politecnica delle Marche (formerly known as Università di Ancona) r.lucchetti(a)univpm.it http://www2.econ.univpm.it/servizi/hpp/lucchetti -------------------------------------------------------