Main Summary View Heatmap

Main Summary View (Heatmap)

The Main Summary View is the central chart in every facet panel. This page documents what it shows and every interaction it supports, including the three selection modes.

For how the heatmap links to the other views and exports records, see Understanding the Views and Selecting and Exporting Data.

Figure: the Main Summary View with its framing histograms and the "Selection Modes" buttons in the top-right corner.

What it shows

The heatmap is a grid:

• Columns = bins of the x variable (or individual categories/list values for a categorical/list x-axis).
• Rows = bins of the y variable.
• Cell color = the color variable aggregated by color_agg (mean by default) over the records that fall in that cell. Darker = higher. See Configuration.

Axes switch to a log scale automatically when the data spans many orders of magnitude. Datetime x-axes are drawn as a time axis; categorical x-axes get rotated labels (thinned when there are many).

Hovering a cell or column reveals a label/tooltip with the bin's range and record count.

Selection modes

Three square buttons in the top-right corner, under the caption "Selection Modes", switch how clicking and dragging on the heatmap selects records. The active button is highlighted. Each button has a hover tooltip:

Button (tooltip)	Mode	What it does
Cell Selection	cell-pin	Click individual cells to toggle them in/out of the selection. Hovering a cell outlines it.
Column Selection	column-pin	Click a column to pin it (selecting all its cells). This is the default. Hovering a column zooms it slightly and draws ribbons around it and any pinned columns.
Brush Selection	2d-brush	Drag a rectangle across the grid to select a 2D region of cells.

The chosen mode is remembered across re-renders.

Column-pin (default)

• Click a column → pin / unpin it. Pinned columns keep a persistent rotated label showing their value or range, and every cell in the column is highlighted as selected.
• Hover a column → it zooms slightly larger, a hover label shows the full value/range and record count, and outline "ribbons" are drawn around the hovered column plus any pinned columns.
• Mouse leaves → the column returns to normal size and the hover label/ribbon clears; pinned ribbons remain.

Cell-pin

• Click a cell → toggle that single cell in or out of the selection (selected cells are highlighted).
• Hover a cell → the cell under the pointer is outlined. (Columns do not zoom in this mode, so cells stay put for precise clicking.)

2D-brush

• Drag a rectangle over the grid → selects all cells inside the box (a range on both axes at once). The brush rectangle stays stable across zoom.
• Click outside the box → clears the brush.
• The 2D-brush selection is shared with the framing histograms, so it stays in sync with the x/y histogram brushes (see Histograms Bar Chart and Legend).

All three modes contribute to the same exported selection — see Selecting and Exporting Data.

List / categorical x-axis extras

When the x-axis is a list column (e.g. HPC node lists) or a high-cardinality categorical, the heatmap gains additional structure below and around the grid. List values are intelligently ordered so related ones sit adjacent (by hardware naming structure when detectable, otherwise by how often values co-occur).

• Count strip — a small bar per column showing the distinct-record count, with a left axis labeled "Records".
• Sharedness strip (list x) — bars showing the fraction of each column's records that span multiple values (e.g. multi-node jobs).
• Co-occurrence arcs (list x) — when you hover or pin a column, curved arcs connect it to columns it frequently co-occurs with; arc width/opacity scales with the strength of the association, and the sharedness bars recolor to show the partners.
• Fleet overview strip (grouped list x) — a persistent mini-map of the whole "fleet" of values, labeled "Fleet (brush to zoom)". Drag a brush across it to zoom the detailed heatmap to a sub-range of values; the current detail window is reflected back onto the strip. A tiny or empty brush clears the zoom. Pinned regions survive the zoom.

Edge-case states

• Categorical overflow — a scalar categorical x-axis is capped at the top values by frequency. When some are hidden, a red note reads "Showing top N of M categories (by frequency)." (List x-axes keep all values via grouping.)
• Too few datapoints — if a facet group has too little data to draw a grid, the panel shows: "There are too few datapoints in this category… To remove this chart, please filter this category from the original dataset."

---

Next: Histograms Bar Chart and Legend · Selecting and Exporting Data · Understanding the Views