Bootstrapping takes a long time

[#1]

Hello,

I am performing a bootstrap on some data created using the mixture distribution function as part of the life distribution to calculate reliability. I am using the bootstrap data to develop confidence intervals for the mixture distribution parameters. The 5 parameters estimated are sigma 1&2, mu 1&2, and portion which is what I am creating the bootstrap for. I am not designating a random seed and left as a '.', I am using fractional weights though with a generated sample size of 2500 as the standard input. The question I have is should this bootstrap normally take 1.5 hours or so? It seems to be pretty long and I was discussing with a coworker that this doesn't seem right and to get my computer checked out as the processing speed may be low. But my computer is fairly new, I have a Dell Precision 7560 with i7 intel processor 2.50GHz and 32GB of RAM, so I am curious if the long time for bootstrapping is typical or if there is something else with JMP I should look into?

Thanks for the help!

[#2]

Cannot judge this, but probably we need a benchmark script to compare,

see code encluded. The analysis needs 79 s on my computer (JMP 16.1 Pro on Win10 on Dell precision 3551 Intel core i7 2.7 MHz 32GB, probably comparable to yours).

Names Default To Here( 1 );

starttime = tick seconds();
dt = New Table( "Random",
	add rows( 1000000 ),
	New Column( "X", formula( Row() ) ),
	New Column( "Y", formula( :X + Random Normal( 0, 3 ) ) )
);

obj = Bivariate( Y( :Y ), X( :X ) );
obj << Fit Line;
(obj << Report)[Number Col Box( 1 )] << Bootstrap(
	100,
	Fractional Weights( 1 ),
	Split Selected Column( 1 )
);
endtime= tick seconds();
print(eval insert("Analysis took ^round(endtime - starttime)^ seconds"))

[#3]

166 seconds on older 4 cpu machine. You have nice hardware!

[#4]

Can't resist a benchmark :). 45 seconds on my MacBook Pro (JMP 16.2):

[#5]

Me neither: "Analysis took 21 seconds"

macOS Monterey on MacBook Pro M1 Max with JMP Pro 16.2

[#6]

That makes me feel good.  My machine is at least 12 years old (i7, 3.4GHz) and it took 142 seconds.  I keep thinking I need to upgrade my machine, but it seems wasteful and unnecessary.  These benchmark results help - yes, I can do better, but I'm not completely out of line!

[#7]

JMP Pro 14、core i5

66

[#8]

Hi @Georg,

Thanks for the benchmark! I ran it and it took me 77 seconds, so I think the issue is more in regards to the computationally intensive method of the bootstrap for estimating the statistical parameters from the mixture distribution data, as @peng_liu pointed out in a later response. I attached an example of the parameters I perform the bootstrap on which takes the 1.5 hours.

[#9]

194 seconds

Windows 10

11th Gen Intel(R) Core(TM) i7-1185G7 @ 3.00GHz 3.00 GHz

16.0 GB (15.7 GB usable)

JMP Pro 16

With many applications running concurrently in RAM. 

[#10]

Mixture distribution fitting is computationally intensive, and the success of fitting heavily depends on the quality of data assuming the model is the right choice. Try the sample data "Mixture Demo" under reliability sample data folder, and see how slow it can be comparing to other modeling types (e.g. least square). That sample data was simulated and considered very good.