Tunny Icon
TunnyDocs

The next-gen Grasshopper optimization tool.

Sensitivity Heatmap

The Sensitivity Heatmap shows one method's scores for parameters against all objectives in a colored grid. The Importance Chart offers a ranked bar view for one objective or grouped bars for ALL objectives on a shared axis. See Sensitivity Analysis for each method's meaning and limitations.

Method selection and execution

  1. Select the method from the family-grouped dropdown: Correlation / Linear (Spearman, Ridge), Tree-based (RF-ANOVA, MDI, SHAP, Permutation), or Global Sensitivity (Sobol first-order and total-effect).
  2. Spearman/Ridge compute automatically when selected. Heavy methods wait for Run to avoid accidental long computations.
  3. In constrained studies, Feasible only restricts fitting to feasible trials.

Numerical-only Spearman and Ridge

Only numerical parameters receive score cells. Nominal categorical parameters, including numeric-looking Optuna choices, are listed separately as Unsupported (categorical), not painted as zero or given an ordinal score. All-categorical input shows No supported numerical parameters, with no score grid. Optuna Int/Float parameters are supported. Spearman excludes non-finite parameter/objective pairs; Ridge excludes rows with any non-finite included numerical feature or objective before sampling and splitting. Numerical NaN is not categorical unsupported.

CSV includes categorical rows with Unsupported (categorical) for each objective instead of numbers. Ridge fits numerical features only; its R² is shown in the Importance Chart as numerical-only, and there is no Ridge R² for all-categorical input.

Reading the grid

  • Rows are supported parameters and columns are objectives. Read down a column to compare parameters for one objective, across a row to compare objectives.
  • Each cell shows its numeric score to two decimals.
  • Signed Spearman/Ridge values use a diverging color scale over [-1, 1]; values outside that range are clamped for color display.
  • Non-negative tree/Sobol values use a sequential scale normalized by each column's maximum. Colors are comparable within a column, not across columns.

Use this overview to choose an objective, then use the Importance Chart for its exact ranking, or select ALL there to compare objectives with numeric bar labels on a shared axis. Unlike the heatmap's per-column color normalization for tree/Sobol methods, the bars keep the method's scores unchanged. Unsupported categories cannot be identified as irrelevant from missing bars or cells.

Caveats

  • Scores are association/predictive contribution, not causality. Confirm with another method.
  • Model quality affects model-based scores. Other methods' categorical encoding imposes arbitrary order or distance.