Plotting comparisons

The bnmetrics.viz functions are gated behind the optional [viz] extra (graphviz, plotly, ipython); calling them without the extra installed raises an ImportError with an actionable hint.

Side-by-side comparison

bnmetrics.plot_side_by_side(g1, g2, name1=..., name2=...) renders two graphs as paired graphviz diagrams. Edges that match between the two panels (same skeleton, same orientation) are highlighted in pastel red so true positives stand out from errors:

import numpy as np
import bnmetrics

true_cpdag = bnmetrics.to_graphlike(
    np.array([
        [0, 1, 1, 0],
        [1, 0, 0, 2],
        [1, 0, 0, 2],
        [0, 1, 1, 0],
    ], dtype=np.int8),
    var_names=("A", "B", "C", "D"),
)
recovered = bnmetrics.to_graphlike(
    np.array([
        [0, 1, 1, 0],
        [2, 0, 0, 2],
        [2, 0, 0, 2],
        [0, 1, 1, 0],
    ], dtype=np.int8),
    var_names=("A", "B", "C", "D"),
)

bnmetrics.plot_side_by_side(
    true_cpdag, recovered,
    name1="truth", name2="recovered",
    direction="LR",
    save="comparison.svg",
)

truth	recovered

save accepts either a single path (the renderer derives two sibling filenames from name1 and name2) or an explicit (path_left, path_right) tuple. The direction argument controls graphviz layout — "LR" (left-to-right) or "TB" (top-to-bottom).

The matched edges (B → D and C → D in this fixture) appear in pastel red in both panels; the reversed upper edges fall back to the default stroke.

Single-graph rendering with highlights

bnmetrics.plot_graph(g, highlight=[...]) renders a single graph with selected nodes painted in a highlight colour. This is the building block underneath both plot_side_by_side and analyse_mb:

bnmetrics.plot_graph(true_cpdag, highlight=["D"], direction="LR",
               save="true_with_D_highlighted.svg")

SID incorrect-edge heatmap

bnmetrics.plot_sid_matrix renders the \((i, j)\) indicator matrix of intervention pairs on which the recovered graph predicts a different distribution than the reference. bnmetrics.sid requires the reference (g1) to be a pure DAG; the recovered graph may be a DAG or a CPDAG (the latter yields separate lower/upper bounds).

true_dag = bnmetrics.to_graphlike(
    np.array([
        [0, 2, 2, 0],   # A → B, A → C
        [1, 0, 0, 2],   # B → D
        [1, 0, 0, 2],   # C → D
        [0, 1, 1, 0],
    ], dtype=np.int8),
    var_names=("A", "B", "C", "D"),
)
recovered_dag = bnmetrics.to_graphlike(
    np.array([
        [0, 1, 2, 0],   # A → C, but B → A (reversed)
        [2, 0, 0, 2],
        [1, 0, 0, 2],
        [0, 1, 1, 0],
    ], dtype=np.int8),
    var_names=("A", "B", "C", "D"),
)

sid_result = bnmetrics.sid(true_dag, recovered_dag)
sid_result.sid              # 5
bnmetrics.plot_sid_matrix(sid_result, save="sid_matrix.html")

The save format is inferred from the file extension. .html is the default; static-image formats (.png, .svg, .pdf, .jpg, .jpeg, .webp) additionally require kaleido (pip install kaleido).