fix deprecation warning

[ArnoStrouwen]

@nograd is being deprecated.
However, seems strange, this needs to be non-differentiable in any case?

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (b66096e) 78.00% compared to head (d1c7321) 78.12%.
Report is 4 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #462      +/-   ##
==========================================
+ Coverage   78.00%   78.12%   +0.12%     
==========================================
  Files          23       23              
  Lines        3155     3155              
==========================================
+ Hits         2461     2465       +4     
+ Misses        694      690       -4     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[ArnoStrouwen]

@oxinabox do you know if these method ambiguities are something to worry about? And how to best fix them?
https://github.com/SciML/Surrogates.jl/actions/runs/7435983790/job/20231942821?pr=462#step:6:833

[oxinabox]

Looks like a bug in ChainRulesCore. I don't understand why they are only showing up recently.
There is nothing for you to do about them here.
I will open a PR presently to fix them in ChainRulesCore

[ArnoStrouwen]

@sathvikbhagavan I don't see this error locally. Is this another stochastic result?
https://github.com/SciML/Surrogates.jl/actions/runs/7498946122/job/20414889506?pr=462#step:6:1897

[sathvikbhagavan]

@sathvikbhagavan I don't see this error locally. Is this another stochastic result?
SciML/Surrogates.jl/actions/runs/7498946122/job/20414889506?pr=462#step:6:1897

Yes, it is indeed stochastic. It happens because one of the test cluster which is sampled contains no points.

[sathvikbhagavan]

I was thinking of fixing it such that if any test cluster is empty, it uses training points for obtaining the best model for that cluster. I will make a PR for it shortly.

[ArnoStrouwen]

@ChrisRackauckas, I think this is ready. The remaining stochastic error should be looked into separately.