Visualization

visualization

Visualization of surrogate modelling.

This module provides the Visualizer class, which is used to visualize the results of a surrogate from saved checkpoints.

surrogate.visualization.VisualizationSettings dataclass

Configuration for the Visualizer object.

Attributes:

Name	Type	Description
offline_checkpoint_file	Path	Checkpoint file after pretraining
online_checkpoint_filestub	Path	) Checkpoint file stub for online training, indices will be appended to the file names automatically
visualization_file	Path	File to save visualizations to
visualization_bounds	list[list[float, float]]	Bounds for plotting in each dimension
rng_seed	int	Seed for RNG used during marginalization

surrogate.visualization.Visualizer

Main object for visualizing surrogate model run results.

The visualizer is used to show the functionality of a surrogate model, trained from saved checkpoints. It plots results for pretraining, as well as online checkpoints. Depending on the dimensionality of the parameter space, the visualization is done in different ways. In any case, the plots show the predicted mean and standard deviation.

Methods:

Name	Description
run	Executes visualizations for all provided checkpoints

__init__

__init__(
    visualization_settings: VisualizationSettings,
    test_surrogate: surrogate_model.BaseSurrogateModel,
) -> None

Constructor.

Takes configuration and a surrogate model. The surrogate should be of the same type as the one used for generating the checkpoints.

Parameters:

Name	Type	Description	Default
visualization_settings	VisualizationSettings	description	required
test_surrogate	surrogate_model.BaseSurrogateModel	description	required

run

run() -> None

Executes visualization.

Creates a pdf document with visualizations of the postetior mean, std and training data for each checkpoint. The exact visualization procedure depend on the dimensionality of the parameter space. - 1D: Visualize mean and std in one plot, equipped with training data - 2D: Visualize mean and std in separate plots, training data is shown in mean plot - AND: Visualize 1D and 2D marginals, with mean and std, training data is not shown

_visualize_checkpoint

_visualize_checkpoint(pdf_file: PdfPages, name: str) -> None

Execute visualization of single checkpoint.

Execution depend on the dimensionality of the parameter space

Parameters:

Name	Type	Description	Default
pdf_file	PdfPages	PDF file to add visualization to	required
name	str	Plot title (checkpoint name)	required

_visualize_checkpoint_1D

_visualize_checkpoint_1D(pdf: PdfPages, name: str) -> None

Visualize a checkpoint for 1D parameter space.

Simple shows the posterior mean, the std as cpnfidence interval, and the training data.

Parameters:

Name	Type	Description	Default
pdf	PdfPages	PDF file to add visualization to	required
name	str	Plot title (checkpoint name)	required

_visualize_checkpoint_2D

_visualize_checkpoint_2D(pdf: PdfPages, name: str) -> None

Visualize a checkpoint for 2D parameter space.

Visualizes mean and std in separate plots, training data is shown in mean plot.

Parameters:

Name	Type	Description	Default
pdf	PdfPages	PDF file to add visualization to	required
name	str	Plot title (checkpoint name)	required

_visualize_checkpoint_AND

_visualize_checkpoint_AND(pdf: PdfPages, name: str) -> None

Visualize a checkpoint for AND parameter space.

This method is invoked for cases with D > 2 dimensional parameter space. It generals plots for all 1D marginals, as well as 2D marginals for all possible parameter tuples. Marginals are evaluated through Monte Carlo sampling in the latent dimensions. Note that this might take some time, and that the approximations are rather coarse to reduce the computational load.

Parameters:

Name	Type	Description	Default
pdf	PdfPages	PDF file to add visualization to	required
name	str	Plot title (checkpoint name)	required

_visualize_1D

_visualize_1D(
    ax: plt.axis,
    ind: int,
    param_values: np.ndarray,
    mean: np.ndarray,
    std: np.ndarray,
    training_data: tuple[np.ndarray, np.ndarray] | None = None,
) -> None

Simple 1D visualization.

Plots the posterior mean and the std as confidence interval. If training data is provided, it is shown as points.

Parameters:

Name	Type	Description	Default
ax	plt.axis	MPL axis object to insert plot into	required
ind	int	Index of parameter dimension (used for axis label)	required
param_values	np.ndarray	x-values	required
mean	np.ndarray	y-values	required
std	np.ndarray	confidence interval	required
training_data	tuple[np.ndarray, np.ndarray] | None	Input and output data for the surrogate training. Defaults to None.	None

_visualize_2D

_visualize_2D(
    ax: plt.axis,
    ind_comb: tuple[int, int],
    param_values: tuple[np.ndarray, np.ndarray],
    solution_values: np.ndarray,
    training_data: tuple[np.ndarray, np.ndarray] | None = None,
) -> None

Simple 2D visualization.

Parameters:

Name	Type	Description	Default
ax	plt.axis	MPL axis object to insert plot into	required
ind_comb	tuple[int, int]	Index combination for the tow parameters under consideration	required
param_values	tuple[np.ndarray, np.ndarray]	x-values	required
solution_values	np.ndarray	y-values (mean or std)	required
training_data	tuple[np.ndarray, np.ndarray] | None	Input and output data for the surrogate training. Defaults to None.	None

_evaluate_1D_marginal

_evaluate_1D_marginal(param_ind: int) -> tuple[np.ndarray, np.ndarray, np.ndarray]

Approximate 1D marginal distribution through Monte Calor sampling.

For every value of the active parameter, compute 100 LHS samples in the latent space and approximate mean and std from it.

Parameters:

Name	Type	Description	Default
param_ind	int	Index of parameter dimension	required

Returns:

Type	Description
tuple[np.ndarray, np.ndarray, np.ndarray]	tuple[np.ndarray, np.ndarray, np.ndarray]: parameter values, marginal mean and std

_evaluate_2D_marginal

_evaluate_2D_marginal(
    param_ind_1: int, param_ind_2: int
) -> tuple[tuple[np.ndarray, np.ndarray], np.ndarray, np.ndarray]

Approximate 2D marginal distribution through Monte Carlo sampling.

For every value combination of the two active parameter, compute 100 LHS samples in the latent space and approximate mean and std from it.

Parameters:

Name	Type	Description	Default
param_ind_1	int	Index of first parameter dimension	required
param_ind_2	int	Index of second parameter dimension	required

Returns:

Type	Description
tuple[tuple[np.ndarray, np.ndarray], np.ndarray, np.ndarray]	tuple[tuple[np.ndarray, np.ndarray], np.ndarray, np.ndarray]: meshgrid of the 2D active parameter space, marginal mean and std