figure_type

Back to L7 | Browse all axes | Browse all options

Axis figure_type on sub-layer L7_B_output_shape_export (layer l7).

Sub-layer

L7_B_output_shape_export

Axis metadata

  • Default: 'auto'

  • Sweepable: False

  • Status: operational

Operational status summary

  • Operational: 6 option(s)

  • Future: 0 option(s)

Options

auto – operational

Pick the figure type matching the importance op’s default mapping.

Each L7.A op declares its canonical figure type (shap_* → bar/beeswarm; partial_dependence → lineplot; shap_interaction → heatmap; rolling_recompute → heatmap; etc.). Setting figure_type = auto honours that default.

When to use

Default; lets each L7.A op choose the canonical figure for its output.

References

  • macroforecast design Part 3, L7: ‘every importance op produces (table, figure) pairs; the L7.B sub-layer governs export shape.’

Related options: bar, boxplot, heatmap, lineplot, scatter

Last reviewed 2026-05-05 by macroforecast author.

bar – operational

Horizontal bar chart – one bar per feature, length = importance score.

The standard global-importance visualisation. Renders features sorted by mean-|importance| so the most important variables surface at the top of the chart. Pair with output_table_format = wide for direct table-figure cross-reference.

When to use

Global importance rankings (linear coefficients, permutation, mean SHAP).

References

  • macroforecast design Part 3, L7: ‘every importance op produces (table, figure) pairs; the L7.B sub-layer governs export shape.’

Related options: auto, boxplot, heatmap, lineplot, scatter

Last reviewed 2026-05-05 by macroforecast author.

boxplot – operational

Boxplot of per-fold / per-bootstrap importance distributions.

Renders each feature as a box capturing the distribution of its importance score across folds (cross-validation, bootstrap, rolling windows). Reveals stability information that a single bar cannot convey.

When to use

Stability-of-importance audits; bootstrap_jackknife / rolling_recompute outputs.

References

  • macroforecast design Part 3, L7: ‘every importance op produces (table, figure) pairs; the L7.B sub-layer governs export shape.’

Related options: auto, bar, heatmap, lineplot, scatter

Last reviewed 2026-05-05 by macroforecast author.

heatmap – operational

Two-axis heatmap (feature × time / model / state).

Visualises importance across an additional dimension. Used for time-varying importance (rolling_recompute), per-state aggregation (group_aggregate over FRED-SD blocks), and pairwise interaction strength (shap_interaction). The us_state_choropleth figure is a specialised heatmap on the US state grid.

When to use

Time-varying importance, FRED-SD state choropleth, group-aggregate matrices.

References

  • macroforecast design Part 3, L7: ‘every importance op produces (table, figure) pairs; the L7.B sub-layer governs export shape.’

Related options: auto, bar, boxplot, lineplot, scatter

Last reviewed 2026-05-05 by macroforecast author.

lineplot – operational

Line plot of importance over time / origin.

Tracks importance evolution across walk-forward origins. Pair with rolling_recompute to surface trends in which features matter as new data arrives.

When to use

Tracking importance evolution across walk-forward origins.

References

  • macroforecast design Part 3, L7: ‘every importance op produces (table, figure) pairs; the L7.B sub-layer governs export shape.’

Related options: auto, bar, boxplot, heatmap, scatter

Last reviewed 2026-05-05 by macroforecast author.

scatter – operational

Scatter plot (e.g. SHAP value vs feature value).

PDP / ALE / SHAP dependence-plot family. Each point is a single observation; the x-axis is the feature value, the y-axis is the importance contribution. Reveals non-linearity in the model’s response.

When to use

PDP / ALE / SHAP dependence-plot family.

References

  • macroforecast design Part 3, L7: ‘every importance op produces (table, figure) pairs; the L7.B sub-layer governs export shape.’

Related options: auto, bar, boxplot, heatmap, lineplot

Last reviewed 2026-05-05 by macroforecast author.