Matrix of Contribution Correlation Plots

The first type of (interactive) similarity plot that we will view in matrix form is the contribution correlation plot discussed in Contribution Correlation Plots. This shows the contribution correlation plot between all pairs of applications and experiments in a (symmetric) matrix. First we consider the case of a square matrix where the set of applications and experiments are the same.

Interactive Matrix Plots

Interactive matrix plots wrap several Plotly plots in an html grid that’s served by a Dash app that allows users to interact with each of the individual plots. This is useful for comparing multiple applications and experiments in a single figure. Specifically, it is useful for viewing all combinations of a given set of SDF files in a single figure.

Note that by default the the correlation plots in the off-diagonal cells are tsunami_ip_utils.viz.scatter_plot.InteractiveScatterLegend plots, which are fully interactive, and dynamically update the calculated summary statistics and regression when points are excluded with the legend. However, this interactivity requires clientiside python to run, and so is not represented in the documentation. The off-diagonal plot types can be changed, and doing so is documented in Changing Plot Defaults.

from tsunami_ip_utils.viz.viz import matrix_plot
from tsunami_ip_utils.viz.plot_utils import generate_plot_objects_array_from_contributions
from tsunami_ip_utils.integral_indices import get_uncertainty_contributions
from paths import EXAMPLES

application_files = [ EXAMPLES / 'data' / 'example_sdfs' / 'MCT' / f'MIX-COMP-THERM-001-00{i}.sdf' for i in range(1,4) ]
experiment_files = application_files

Now we need to get the uncertainty contributions for each experiment and each application using tsunami_ip_utils.integral_indices.get_uncertainty_contributions() (described in Uncertainty/Variance Contributions).

contributions_nuclide, contributions_nuclide_reaction = get_uncertainty_contributions(
    application_files,
    experiment_files,
    variance=True
)

Nuclide-Wise Contributions

Now, the tsunami_ip_utils.viz.viz.matrix_plot() function needs a numpy array of plot objects to generate the matrix plot. We can generate this array using the tsunami_ip_utils.viz.plot_utils.generate_plot_objects_array_from_contributions() function. This function takes a dictionary of contributions and the name of the integral index, and returns a numpy array of plot objects that can be passed to the tsunami_ip_utils.viz.viz.matrix_plot() function.

plot_objects_array_nuclide = generate_plot_objects_array_from_contributions(contributions_nuclide, '%(Δk/k)^2')

Now the matrix plot can easily be generated

fig = matrix_plot(plot_objects_array_nuclide, plot_type='interactive')
fig.show()

- Zoom Out + Zoom In

This plot is fully interactive. You can hold shift and scroll using the mousewheel to scroll horizontally, or just use the mousewheel alone to scroll vertically. Individual subfigures can also be saved by clicking the camera icon in the top right, or the entire plot can be saved by calling the to_image method on the figure object.

fig = matrix_plot(plot_objects_array_nuclide, plot_type='interactive')
fig.to_image( EXAMPLES / '_static' / 'matrix_plot.png' )

# sphinx_gallery_thumbnail_path = '../../examples/_static/matrix_plot.png'

- Zoom Out + Zoom In

Note if the default, nondescriptive labels aren’t preferred, an arbitrary dictionary of labels for the applications and experiments may be passed instead. Perhaps the most useful is the SDF file names.

labels = {
    'applications': [ application_file.name for application_file in application_files ],
    'experiments': [ experiment_file.name for experiment_file in experiment_files ],
}
fig = matrix_plot(plot_objects_array_nuclide, plot_type='interactive', labels=labels)
fig.show()

- Zoom Out + Zoom In

Nuclide-Reaction-Wise Contributions

The same can be done for the nuclide-reaction-wise contributions

plot_objects_array_nuclide_reaction = generate_plot_objects_array_from_contributions(contributions_nuclide_reaction, '%(Δk/k)^2')
fig = matrix_plot(plot_objects_array_nuclide_reaction, plot_type='interactive', labels=labels)
fig.show()

- Zoom Out + Zoom In

Static Matrix Plots

Static matrix plots are used for creating an analogous figure using static matplotlib plots. This implementation would have no need for the html of the interactive matrix plot, but as it stands, the interactive plot can contain all static plots as well (and a mixture of static and interactive plots), and so this feature has not yet been implemented.