Problem while adding custom kernel on gaussian processes

There’s an issue with the ability of allowing ppl to easily add custom kernels for gaussian processes. I have custom method but keep getting error when adding it directly to the GP constructor it complains about the get_params method.

Steps/Code to Reproduce

Example:

from sklearn.gaussian_process import GaussianProcessRegressor
X = np.random.randn(100, 4)
y = np.random.randn(100)

def custom_kernel(x=None):
       x = np.random.randn(10, 3)
       return x.dot(x.T)
       
gp = GaussianProcessRegressor(kernel=custom_kernel)
gp.fit(X, y)

Expected Results

*** TypeError: Cannot clone object 'array([0.99946371, 0.99859908, 0.99818002, 0.9994673 , 0.99954341,
       0.99969851, 0.99905772, 0.99932515, 0.99906479, 0.99983097])' (type <class 'numpy.ndarray'>): it does not seem to be a scikit-learn estimator as it does not implement a 'get_params' methods.

Versions

System

python: 3.6.7 |Anaconda, Inc.| (default, Oct 23 2018, 14:01:38)  [GCC 4.2.1 Compatible Clang 4.0.1 (tags/RELEASE_401/final)]

executable: /Users/jm/miniconda3/envs/tf/bin/python machine: Darwin-18.2.0-x86_64-i386-64bit

BLAS

macros: SCIPY_MKL_H=None, HAVE_CBLAS=None

lib_dirs: /Users/kirk/miniconda3/envs/tf/lib cblas_libs: mkl_rt, pthread

Python deps

   pip: 18.1

setuptools: 40.5.0 sklearn: 0.20.0 numpy: 1.15.4 scipy: 1.1.0 Cython: None pandas: 0.23.4

[amueller]

And https://scikit-learn.org/dev/modules/generated/sklearn.gaussian_process.kernels.PairwiseKernel.html#sklearn.gaussian_process.kernels.PairwiseKernel

Maybe adding an example to the user guide would be nice?

[kirk86]

@eamanu Is there any documentation anywhere to be found on how to create a custom Kernel? Does scikit docs include this information? It’s easy to close an issue and say that’s not a Kernel. Sure. Give me an example please! If I google about how to construct custom kernel in scikit, this is the first result that comes. And all the example about constructing custom kernels are functions related to the svm api. As you @jnothman have already understood this is confusing to the end user since the interface is not consistent. And correct me if I’m wrong but still cannot find any documentation on creating custom kernels in GPs. Apparently that’s class and should implement the get_params method, one has to look into the actual api code to figure things out, but even then things are not completely clear. Thanks!