Plotting and Export

The shared plotting functions accept either an existing Matplotlib axis or create a new figure. This supports both interactive use and multi-panel publication figures.

Visual system

IMVpy uses the same visual language as the IMV research figures:

• A navy-to-steel-blue-to-cream-to-red continuous colormap registered as imv and imv_r.
• Helvetica typography when installed, followed by Nimbus Sans and DejaVu Sans as portable fallbacks.
• White backgrounds, thin dark axes and mark outlines, dashed light-gray grids, bold left-aligned titles, and compact 5.4 x 4.6 inch panel geometry.
• Three-decimal bar and default heatmap annotations. The lower-level heatmap_style switches to two decimals for more than five classes.

Package plotting functions apply these choices within a scoped Matplotlib context, so they do not overwrite unrelated plotting defaults.

import matplotlib.pyplot as plt

from imvpy.utils import plotting_context, spectral_colors

with plotting_context():
    figure, axis = plt.subplots()
    axis.plot([0, 1, 2], [0.1, 0.4, 0.3], color=spectral_colors(1)[0])

For a notebook dedicated entirely to IMV figures, opt into process-wide styling:

from imvpy.utils import configure_plotting

configure_plotting()

configure_plotting intentionally modifies Matplotlib's global rcParams. Prefer plotting_context in reusable modules or applications. Use rc_params when another context manager or plotting framework needs the settings mapping.

Heatmaps

from imvpy.utils import plot_ablation_matrix, plot_imv_heatmap

figure, axis = plot_imv_heatmap(
    matrix,
    title="Pairwise class IMV",
    fmt=".3f",
)

figure_ablation, axis_ablation = plot_ablation_matrix(ablation_matrix)

Matrices must be non-empty square two-dimensional arrays. A pandas DataFrame supplies labels from its columns automatically; otherwise provide labels. plot_ablation_matrix uses an ablation-specific title and places zero at the neutral cream midpoint. plot_imv_heatmap accepts cmap, center, and colorbar_label when another explicit encoding is required.

The lower-level heatmap_style, style_heatmap_frame, and style_heatmap_axes helpers support custom seaborn heatmaps. For a multi-panel layout, add_heatmap_colorbar creates a full-height vector colorbar without resizing sibling panels.

One-vs-rest distributions

from imvpy.utils import plot_ova_boxplot

figure, axis = plot_ova_boxplot(
    fold_scores,
    labels=["class A", "class B", "class C"],
    title="One-vs-rest stability",
)

fold_scores is a non-empty two-dimensional folds-by-classes matrix. Box width or fold spread describes sensitivity to the fold partition, not a confidence interval.

Publication bars

Use plot_bars for the black-edged, palette-colored, three-decimal bar style used by the research figures. Error bars and negative values are annotated on the correct side of their caps. Call set_bar_limits after all panels in a shared-y row have been drawn.

import matplotlib.pyplot as plt

from imvpy.utils import plot_bars, set_bar_limits

figure, axis = plt.subplots()
plot_bars(axis, ["Full", "Ablated"], [0.32, 0.18], yerr=[0.02, 0.03])
set_bar_limits(axis)

bar_style and annotate_bars expose the same choices for custom vertical bar charts. categorical_colors, sequential_cmap, and spectral_colors expose the palette independently of a particular plot.

Panel composition

figure_size(2, 3) computes dimensions from the canonical panel geometry and label_panels(axes) applies bold a., b., ... labels in row-major order. Pass only data axes to avoid labelling colorbars. style_axis applies the Cartesian frame and accepts grid_axis="x", "y", "both", or None. apply_tight_layout reserves room for inset colorbars and figure-level legends.

Existing axes

import matplotlib.pyplot as plt

figure, axes = plt.subplots(1, 2, figsize=(12, 5))
returned = plot_imv_heatmap(matrix_a, ax=axes[0], title="Estimator A")
plot_imv_heatmap(matrix_b, ax=axes[1], title="Estimator B")

assert returned is axes[0]

When ax=None, plotting functions return (figure, axis). When ax is provided, they return that axis.

Publication export

from imvpy.utils import FIGURE_DPI, FIGURE_FORMATS, save_figure

paths = save_figure(figure, "artifacts/result.png")

assert FIGURE_DPI == 800
assert FIGURE_FORMATS == ("png", "pdf", "svg")
print(paths)

The destination may have no suffix or any supported suffix. save_figure removes that suffix and writes three siblings:

artifacts/result.png
artifacts/result.pdf
artifacts/result.svg

PNG is written at 800 DPI. The DPI is also passed to PDF and SVG so rasterized artists inside vector figures use the same resolution. Parent directories are created automatically, and the returned dictionary maps format name to Path. Exports use a white background and 0.04-inch padding.

The caller controls the output directory. Keep generated figures outside the package source tree unless they are intentional documentation assets.