Scatter Matrix
Overview
The Scatter Matrix (also called a pairplot) provides a compact overview of correlations and interactions across variable pairs. It includes all parameters followed by all objectives from the current study, in an grid. There is no axis-selection, display-mode, or sorting control within this widget.
Unlike Correlation Matrix, which summarizes each pair with a coefficient, Scatter Matrix also shows the point clouds, making curved relationships and clusters easier to spot.
Controls
| Item | Description |
|---|---|
| Color by | Choose the objective whose raw value colors the scatter points using the active colormap. Defaults to the first objective. |
| Show Infeasible (constrained studies only) | Show or hide confirmed infeasible scatter points. Enabled by default. Does not filter histograms or correlation coefficients. |
In constrained studies, feasible points use the objective colormap, infeasible points use a fixed color, and points labeled Feasibility unverified use a distinct fixed color. Unverified points remain visible when Show Infeasible is off. The objective color range is based on feasible trials; in an unconstrained study, all trials are treated as feasible.
Features
- Cells are arranged in an grid.
- Lower triangle: a scatter plot with the column variable on the X axis and the row variable on the Y axis.
- Upper triangle: the Pearson correlation coefficient, displayed to two decimal places. Negative values shade toward red, zero is white, and positive values shade toward blue.
- Diagonal: a histogram with 10 equal-width bins for each variable. Bin ranges and counts use only finite values, excluding missing values (NaN) and positive or negative infinity. A column with no finite values shows no bars; a constant finite column places its finite observations in the middle bin.
- Large studies: scatter points are evenly downsampled to at most 1,500 per feasibility group (feasible, infeasible, and unverified). Histograms and correlations use the full columns, not the downsampled scatter points.
- Labels: row and column labels rotate automatically when space is limited.
- Selection linkage: Scatter Matrix does not participate in cross-widget selection. It neither highlights selections made in other widgets nor generates its own selection. It has no point-hover details or click-to-open Trial Detail; use All Trials to inspect individual trials.
Correlation Coefficient Cells (Upper Triangle)
The coefficient is the
Pearson product-moment correlation
of the two variables' raw values. It ranges from (negative linear
correlation) through (no linear correlation) to (positive linear
correlation); a larger absolute value means stronger linear co-movement. These
cells include all feasibility groups, regardless of Show Infeasible. Degenerate
or undefined results are displayed as 0.00, not as a separate missing-value
indicator.
Pearson correlation measures linear co-movement, so a curved (nonlinear) 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 relationship between the two variables.
- Curved or nonlinear pattern: a curved cloud suggests a nonlinear relationship — use PDP Chart 2D for a detailed view.
- Circular cloud: no obvious systematic relationship between the variables.
- Clusters: distinct groups of points suggest separate regions of the sampled variable space; they do not by themselves establish discrete categories.
- Triangular layout: each variable pair is plotted only once; the mirrored upper-triangle cell contains its correlation coefficient instead.
- Looking for important parameters: a strong pattern between a parameter and an objective is a useful lead, not proof of importance or causality. Check Importance Chart for further analysis.