Fix to the parameter transformation functionality

[yonatank93]

List of changes:

• Set kliff.model.Models().model_params as an attribute that store the parameter values in the original parameterization.
• Move the transformation and inverse transform to kliff.model.OptimizingParameters(), e.g., the transform function in get_opt_params and the inverse transform in update_opt_params methods. This also requires the parameter transfomation class to be passed in to OptimizingParameters.
• Add a line in the echo methods saying that the printed values are in the original parameterization.

[mjwen]

@yonatank93 thank for the PR! We may not want to touch the OptimizingParameters class. Can you please enable Allow edits from maintainers for me to make some adjustments?

[yonatank93]

@mjwen I enabled Allow edits from maintainers. Let me know if this is not working.

[mjwen]

The functionality to deal with parameter transformation has all been moved to the Model class, leaving OptimizingParams free. Expect the below behaviors:

• model.echo_opt_params() will always show the transformed params

• A new parameter params_space added model.echo_model_params(params_space=...). A value of original will show params in the untransformed space (this is the default behavior), and a value of transformed will show it in the transformed space.

@yonatank93 Can you test this out and let me know whether this solves your problem?

[yonatank93]

In OptimizingParams().echo_opt_params(), would it be useful to add a line in the heading "Parameters in transformed space.", like in model.echo_model_params()? Or to the least, we should mention in the documentation for this method that the printed values are in the transformed space.

[mjwen]

Agreed! It's useful to let the user know that it's in the transformed space. Added.

Also, it would be helpful to add a tutorial example to demonstrate this. Can you convert your Jupyter notebook to an example as those listed here? Probably take a look at example_kim_SW_Si.py. The docstrings in the examples follow the sphinx-gallery syntax.

[yonatank93]

Sure, I will do it later today. In the example, we want to show:

• How to use, say, log transformation, which will also include some motivation why we want to transform the parameter space.
• Do we also want to show how to construct different parameter transformation class?
• Update the parameter values in the transformed space.
• Do we want to show the difference in the result when we use the original vs transformed space?

[mjwen]

Thanks!

• We may not want to show how to construct different parameter transformation in this tutorial. But can say that a user can do it easily by subclassing the ParameterTransform class.

• It might be difficult/lengthy to compare the results. So, probably not do it at this stage. We may need longer docs/tutorials once all the UQ stuff is fleshed out. Then, we can do it in a more thorough way.

I will merge and close this. Feel free to open a new PR for the tutorial.