Subscribe Bookmark RSS Feed

Is there randomness in fitting a Gaussian Process with JMP? If so, how can I set the random seed?

jazzy09

Community Trekker

Joined:

Aug 18, 2015

I am trying to compare JMP's Gaussian Process model fitting with other software packages. Most other packages have a random aspect to it: the initial values, how long to run the optimization routine for. Does JMP work the same way? Or is the same fit given every time given the same data? And if there is randomness, how can I change the random seed to see how it affects the results?

Below I have pasted the code I've used to test this. It creates a 1-D data set then fits a GP to it. The random reset function is called before fitting the model. From what I've done so far (with this and other data sets) it seems that the same fit (parameter estimates) are given every time. I would love to know definitively what exactly is going on with the GP function call.

New Table( "fdt",

  Add Rows( 10 ),

  New Column( "x",

  Numeric,

  Continuous,

  Format( "Best", 11 ),

  Set Values(

  [0.778935623, 0.50233312, 0.947723007, 0.173231374, 0.369273156,

  0.647761962, 0.286120948, 0.443809711, 0.024479728, 0.807067905]

  )

  ),

  New Column( "y",

  Numeric,

  Continuous,

  Format( "Best", 11 ),

  Set Values(

  [13.73502137, 0.006748857, 3.933447467, 6.585368515, 14.74814471,

  16.95462647, 20.04919594, 3.668553052, 7.662042986, 9.481372346]

  )

  )

);

Random reset(400);

Gaussian Process(

  Y( :y ),

  X( :x ),

  Estimate Nugget( 0 ),

  Set Correlation Function( "Gaussian" )

);