Gradient of ConstantKernel is Incorrect #11581
Description
Description
Given the function f(x,y) = x * y. The partial derivative of f(x,y) with respect to x is y.
When using a kernel combination of the kind:
ConstantKernel * RBF
The gradient of ConstantKernel should be the same as the evaluation of RBF.
Steps/Code to Reproduce
The following:
import numpy as np
from sklearn.gaussian_process.kernels import ConstantKernel, RBF
data = np.array([[1., 3.], [5., 6.]])
kernel = ConstantKernel(0.5) * RBF(1.0)
kernel(data, eval_gradient=True)[1][:, :, 0]
>>> array([[5.00000000e-01, 1.86332659e-06],
[1.86332659e-06, 5.00000000e-01]])Expected Results
Should match:
RBF(1.0)(data)
>>> array([[1.00000000e+00, 3.72665317e-06],
[3.72665317e-06, 1.00000000e+00]])Actual Results
Instead we are getting the value of evaluating kernel:
kernel(data)
>>> array([[5.00000000e-01, 1.86332659e-06],
[1.86332659e-06, 5.00000000e-01]])Versions
NumPy 1.14.5
SciPy 1.1.0
Scikit-Learn 0.19.2