how to surrogate model with multiple variable? about pysot HOT 5 CLOSED

Your data points should have size 20 x 3 since you have 20 points in 3 dimensions.

A simple example where you build a cubic RBF from 100 data points in 3 dimensions may look like this:

import numpy as np
from pySOT.surrogate import RBFInterpolant, CubicKernel, LinearTail

np.random.seed(0)
X = np.random.rand(100, 3)
y = np.sin(np.sum(X, 1))
rbf = RBFInterpolant(dim=3, kernel=CubicKernel(), tail=LinearTail(dim=3), eta=1e-6)
rbf.add_points(X, y)
print(rbf.predict(np.zeros((1, 3))))  # Should be close to 0

Output:

[[0.01681138]]

The functional form of the surrogate model depends on what model you're using. RBFs aren't easily interpretable from the model weights, but other models like polynomial regression are.

from pysot.

Hello,

My query is related to the above example.

Using your example above, I would like to build a cubic RBF over 20 points in 3*3 dimensions.

Does pySOT support this feature?

Thank you!

from pysot.

Hi Zosezhuo,
Can you clarify what you mean by 3*3 dimensions?

from pysot.

Hi

Specifically, I would like to do the following in 3 by 3 dimensions.

np.random.seed(0)
X = np.random.uniform(low=0.0, high=1.0, size=[3,3])
y= fun(X)
y2= fun2(X)

print(X)
print(y)
print(y2)

[[0.5488135  0.71518937 0.60276338]
 [0.54488318 0.4236548  0.64589411]
 [0.43758721 0.891773   0.96366276]]
134408.75071136828
3.144276104044118

When I tried to add points, I received the following output

rbf_cubic = RBFInterpolant(dim=[3,3], kernel=CubicKernel(), tail=LinearTail(dim=[3,3]), eta=1e-6)
rbf_cubic.add_points(X, y)
rbf_cubic.add_points(X, y2)

TypeError: unsupported operand type(s) for +: 'int' and 'list

Appreciate your time and help very much

from pysot.

An RBF interpolant is a function from R^d -> R. You need to pass in a matrix X with shape (n, d) and y with shape (n,).

from pysot.