twinlab.Emulator.summarise

Emulator.summarise(detailed=False, verbose=False)

Get a summary of a trained emulator on the twinLab cloud.

This summary returns transformer diagnostics, with details about the input/output decomposition. It also returns the estimator diagnostics, detailing properties of the trained emulator. The estimator diagnostics includes information about the kernel, likelihood, mean modules and some additional properties of the emulator. This information can help inform a user about the makeup of an emulator – for example, what kind of kernel was used. The summary contains the information about the following components of an emulator.

Kernel	Details about the structure of the kernel used, kernel lengthscale priors, kernel lengthscale bounds, etc.
Likelihood	Details about the noise model used (could be another GP in the case of Heteroskedastic GP), raw noise variance, etc.
Mean	Details about the mean function used and the raw mean constant.
Properties	Additional properties of the emulator like noise variance, input transform parameters, output transform parameters, properties of the variational distribution (only in the case of Variational GPs), etc.

By default, the summary dictionary is not detailed and contains only a high-level overview of the emulator which includes information about the learned hyperparameters, kernel function used and the mean function used. To view a detailed summary innvolving information about all the parameters of the emulator, set the detailed parameter to True.

Parameters:

verbose (bool, optional) – Display information about the operation while running.

Return type:

dict

Example

emulator = tl.Emulator("quickstart")
emulator.summarise()

{
    'kernel': ...,
    'likelihood': ...,
    'mean': ...,
    'properties': ...,
}

emulator = tl.Emulator("quickstart")
emulator_summary = emulator.summarise()
kernel_info = emulator_summary.get("kernel")
print("Kernel function used: ", kernel_info["kernel_function_used"])
print("Lengthscale:", kernel_info["lengthscale"])

Kernel function used: ScaleKernel((base_kernel): MaternKernel((lengthscale_prior): GammaPrior()
(raw_lengthscale_constraint): Positive())
(outputscale_prior): GammaPrior()  (raw_outputscale_constraint): Positive())

Lengthscale: [[0.4234508763532827]]