20  Kriging with Varying Correlation-p

This chapter illustrates the difference between Kriging models with varying p. The difference is illustrated with the help of the spotpython package.

20.1 Example: Spot Surrogate and the 2-dim Sphere Function

import numpy as np
from math import inf
from spotpython.fun.objectivefunctions import Analytical
from spotpython.spot import Spot
from spotpython.utils.init import fun_control_init, surrogate_control_init
PREFIX="015"

20.1.1 The Objective Function: 2-dim Sphere

  • The spotpython package provides several classes of objective functions.
  • We will use an analytical objective function, i.e., a function that can be described by a (closed) formula: \[f(x, y) = x^2 + y^2\]
  • The size of the lower bound vector determines the problem dimension.
  • Here we will use np.array([-1, -1]), i.e., a two-dim function.
fun = Analytical().fun_sphere
fun_control = fun_control_init(PREFIX=PREFIX,
                               lower = np.array([-1, -1]),
                               upper = np.array([1, 1]))
  • Although the default spot surrogate model is an isotropic Kriging model, we will explicitly set the theta parameter to a value of 1 for both dimensions. This is done to illustrate the difference between isotropic and anisotropic Kriging models.
surrogate_control=surrogate_control_init(n_p=1,
                                         p_val=2.0,)
spot_2 = Spot(fun=fun,
                   fun_control=fun_control,
                   surrogate_control=surrogate_control)

spot_2.run()
spotpython tuning: 7.295426096197904e-06 [#######---] 73.33% 
spotpython tuning: 7.295426096197904e-06 [########--] 80.00% 
spotpython tuning: 7.295426096197904e-06 [#########-] 86.67% 
spotpython tuning: 7.295426096197904e-06 [#########-] 93.33% 
spotpython tuning: 7.295426096197904e-06 [##########] 100.00% Done...

Experiment saved to 015_res.pkl
<spotpython.spot.spot.Spot at 0x121ef76e0>

20.1.2 Results

spot_2.print_results()
min y: 7.295426096197904e-06
x0: 0.0005343129105654898
x1: 0.002647628336795204
[['x0', np.float64(0.0005343129105654898)],
 ['x1', np.float64(0.002647628336795204)]]
spot_2.plot_progress(log_y=True)

spot_2.surrogate.plot()

20.2 Example With Modified p

  • We can use set p_val to a value other than 2 to obtain a different Kriging model.
surrogate_control = surrogate_control_init(n_p=1,
                                           p_val=1.0)
spot_2_p1= Spot(fun=fun,
                    fun_control=fun_control,
                    surrogate_control=surrogate_control)
spot_2_p1.run()
spotpython tuning: 0.008151340444524795 [#######---] 73.33% 
spotpython tuning: 0.008151338321477783 [########--] 80.00% 
spotpython tuning: 0.008151338321477783 [#########-] 86.67% 
spotpython tuning: 0.008151329471958513 [#########-] 93.33% 
spotpython tuning: 0.008151328637051841 [##########] 100.00% Done...

Experiment saved to 015_res.pkl
<spotpython.spot.spot.Spot at 0x1552b8560>
  • The search progress of the optimization with the anisotropic model can be visualized:
spot_2_p1.plot_progress(log_y=True)

spot_2_p1.print_results()
min y: 0.008151328637051841
x0: 0.051706566088337895
x1: 0.07401188864232683
[['x0', np.float64(0.051706566088337895)],
 ['x1', np.float64(0.07401188864232683)]]
spot_2_p1.surrogate.plot()

20.2.1 Taking a Look at the p_val Values

20.2.1.1 p_val Values from the spot Model

  • We can check, which p_val values the spot model has used:
  • The p_val values from the surrogate can be printed as follows:
spot_2_p1.surrogate.p_val
1.0
  • Since the surrogate from the isotropic setting was stored as spot_2, we can also take a look at the theta value from this model:
spot_2.surrogate.p_val
2.0

20.3 Optimization of the p_val Values

surrogate_control = surrogate_control_init(n_p=1,
                                           optim_p=True)
spot_2_pm= Spot(fun=fun,
                    fun_control=fun_control,
                    surrogate_control=surrogate_control)
spot_2_pm.run()
spotpython tuning: 7.476853710610035e-06 [#######---] 73.33% 
spotpython tuning: 7.476853710610035e-06 [########--] 80.00% 
spotpython tuning: 7.476853710610035e-06 [#########-] 86.67% 
spotpython tuning: 6.058492032872421e-06 [#########-] 93.33% 
spotpython tuning: 6.058492032872421e-06 [##########] 100.00% Done...

Experiment saved to 015_res.pkl
<spotpython.spot.spot.Spot at 0x156f69070>
spot_2_pm.plot_progress(log_y=True)

spot_2_pm.print_results()
min y: 6.058492032872421e-06
x0: 0.0008753496178213194
x1: 0.0023004901824290602
[['x0', np.float64(0.0008753496178213194)],
 ['x1', np.float64(0.0023004901824290602)]]
spot_2_pm.surrogate.plot()

spot_2_pm.surrogate.p_val
array([1.99943293])

20.4 Optimization of Multiple p_val Values

surrogate_control = surrogate_control_init(n_p=2,
                                           optim_p=True)
spot_2_pmo= Spot(fun=fun,
                    fun_control=fun_control,
                    surrogate_control=surrogate_control)
spot_2_pmo.run()
spotpython tuning: 1.0859280773037223e-05 [#######---] 73.33% 
spotpython tuning: 1.0859280773037223e-05 [########--] 80.00% 
spotpython tuning: 1.0859280773037223e-05 [#########-] 86.67% 
spotpython tuning: 1.0859280773037223e-05 [#########-] 93.33% 
spotpython tuning: 1.0859280773037223e-05 [##########] 100.00% Done...

Experiment saved to 015_res.pkl
<spotpython.spot.spot.Spot at 0x156ea6cc0>
spot_2_pmo.plot_progress(log_y=True)

spot_2_pmo.print_results()
min y: 1.0859280773037223e-05
x0: 0.0007730756995786265
x1: 0.0032033786438318904
[['x0', np.float64(0.0007730756995786265)],
 ['x1', np.float64(0.0032033786438318904)]]
spot_2_pmo.surrogate.plot()

spot_2_pmo.surrogate.p_val
array([2.        , 1.99967559])

20.5 Exercises

20.5.1 fun_branin

  • Describe the function.
    • The input dimension is 2. The search range is \(-5 \leq x_1 \leq 10\) and \(0 \leq x_2 \leq 15\).
  • Compare the results from spotpython runs with different options for p_val.
  • Modify the termination criterion: instead of the number of evaluations (which is specified via fun_evals), the time should be used as the termination criterion. This can be done as follows (max_time=1 specifies a run time of one minute):
fun_evals=inf,
max_time=1,

20.5.2 fun_sin_cos

  • Describe the function.
    • The input dimension is 2. The search range is \(-2\pi \leq x_1 \leq 2\pi\) and \(-2\pi \leq x_2 \leq 2\pi\).
  • Compare the results from spotpython run a) with isotropic and b) anisotropic surrogate models.
  • Modify the termination criterion (max_time instead of fun_evals) as described for fun_branin.

20.5.3 fun_runge

  • Describe the function.
    • The input dimension is 2. The search range is \(-5 \leq x_1 \leq 5\) and \(-5 \leq x_2 \leq 5\).
  • Compare the results from spotpython runs with different options for p_val.
  • Modify the termination criterion (max_time instead of fun_evals) as described for fun_branin.

20.5.4 fun_wingwt

  • Describe the function.
    • The input dimension is 10. The search ranges are between 0 and 1 (values are mapped internally to their natural bounds).
  • Compare the results from spotpython runs with different options for p_val.
  • Modify the termination criterion (max_time instead of fun_evals) as described for fun_branin.

20.6 Jupyter Notebook

Note