Additional features on Gaussian ProcessWhat inspired this Wish List Request?
I recently had a meeting with NIPPON SHOKUBAI CO LTD. The customer request two features on Gaussian Process (correlation structure = Gaussian).
The first request is the feature to set up a common theta parameter. The current JMP only allows for different theta parameter for each predictor. The customer said that different theta model tends to be overfitting.
The second request is feature to create models with user-specified parameter values. They said that they would be happy to specify their own parameters and compare models.
Please describe the new or improved feature or solution you would like to see.
Additional features on Gaussian Process (correlation structure = Gaussian).
- Feature to set up a common theta parameter.
- Feature to create models with user-specified parameter values.
What would the value be to you and/or other users if this new feature is provided?
Users can consider various Gaussian process models.
Is this an external customer request?
Yes. NIPPON SHOKUBAI CO LTD requested this feature.
Is there a competitive software package that has this feature or solution that we should be aware of?
Python.
https://scikit-learn.org/stable/modules/generated/sklearn.gaussian_process.GaussianProcessRegressor.html#sklearn.gaussian_process.GaussianProcessRegressor
Will this impact a renewal, expansion, or adoption at a customer, and does this put an account at risk if we do not address this?
There is potential for expansion.
Advanced Statistical ModelingRe: Additional features on Gaussian Process - Status changed to: AcknowledgedRe: Additional features on Gaussian Process - Status changed to: InvestigatingHi @Takumi_Saru , thank you for these suggestions, and for providing the details behind the request. We are researching for possible inclusion in a future release.
Re: Additional features on Gaussian ProcessHi @Takumi_Saru , does the customer have any data that would help us test and show the usefulness of forcing the same parameter value?
The fixing parameter values for exploration is also a good idea. In the meantime they could edit the formula and change the values themselves to compare. It's a little tedious to do this way though, so it makes sense to try and make this easier.
Re: Additional features on Gaussian Process@mia_stephens -san, @RyanParker -san, thank you for reply.
I checked with the customer and received the following response.
--
We cannot provide specific sample data.
Since it is difficult to imagine a case in which the degree of nonlinearity differs greatly for each explanatory variable, a common θ is often used when dealing with experimental prescriptions.
If we optimize θ for each explanatory variable in a situation where the common θ is appropriate, there is no problem if it fits the common θ.
However, in our experience, it is often the case that each θ becomes a very small value (too nonlinear) due to a strong fit to the data.
We are sorry that we cannot provide specific sample data, but we can create toy data for you.
If you would like toy data, could you let us know the deadline?
--
Re: Additional features on Gaussian Process@Takumi_Saru a toy data set would be very helpful, and no deadline!
Re: Additional features on Gaussian Process@RyanParker -san, I will ask the customer to create the toy data.
I will contact you upon receipt of the toy data.
Re: Additional features on Gaussian Process - Status changed to: Acknowledged
