Scatter Matrix
Overview
The Scatter Matrix (also called a pairplot) shows pairwise scatter plots for every combination of selected parameters and objectives. It provides a compact overview of correlations and interactions across all variable pairs simultaneously.
Controls
| Item | Description |
|---|---|
| Color by | Use the "Color by" dropdown to choose which objective function is used to color the scatter points. Points are colored according to the selected objective's value using the active colormap. |
| Show Infeasible (constrained studies only) | Toggle display of infeasible trials. |
Features
- Cells are arranged in an grid.
- Lower triangle: shows a scatter plot for each variable pair.
- Upper triangle: shows the correlation coefficient for each variable pair, with the cell colored by its magnitude.
- Diagonal: shows a histogram for each variable.
- Selection linkage: Scatter Matrix does not participate in cross-widget selection. It neither highlights selections made in other widgets nor generates its own selection through interaction.
Correlation Coefficient Cells (Upper Triangle)
The correlation coefficient shown in the upper-triangle cells is the Pearson product-moment correlation of the two variables' raw values. It ranges from (negative linear correlation) through (no correlation) to (positive linear correlation); a larger absolute value means a stronger linear co-movement. The cell is shaded according to the magnitude.
Because Pearson correlation measures linear co-movement, note that a curved (non-linear) relationship can yield a small coefficient even when the relationship is strong.
How to Read
- Linear pattern: points forming a clear diagonal line indicate a strong linear correlation between the two variables.
- Curved or nonlinear pattern: a curved cloud of points indicates a nonlinear relationship — use PDP Chart 2D for a detailed view.
- Circular cloud: no systematic correlation between the two variables.
- Clusters: distinct groups of points in a cell suggest that trials fall into discrete categories for those two variables.
- Triangular layout: diagonal cells show a histogram of each variable.
- Lower-triangle cells show a scatter plot with the column variable on the X axis and the row variable on the Y axis.
- Upper-triangle cells show the correlation coefficient instead of a mirrored scatter plot, so each variable pair is plotted only once.
- Looking for important parameters: look for cells where a parameter axis shows a strong pattern with an objective axis — those parameters likely have high importance.