How do I use the parallel coordinates chart in W&B?

The parallel coordinates chart visualizes the relationship between hyperparameters and metrics across many runs. Each run is a line; each axis is a config parameter or metric. Lines that converge toward good metric values reveal which hyperparameter combinations work best.

Adding a parallel coordinates panel

1. In your workspace, click Add panel.
2. Select Parallel coordinates from the panel type list.
3. The panel auto-populates axes using the config keys and summary metrics logged by your runs.

Choosing which axes to display

Click the pencil icon on the panel to open settings. Under Columns, add or remove axes:

• Config keys (e.g., learning_rate, batch_size, optimizer) appear as categorical or continuous axes.
• Summary metrics (e.g., val/accuracy, val/loss) appear as the outcome axes — usually placed on the right.

Drag axes to reorder them. The most useful layout puts hyperparameters on the left and the target metric on the right.

Coloring lines by a metric

To make the best-performing runs visually stand out, set the Color option to a summary metric such as val/accuracy. Lines are colored on a gradient from worst (cool colors) to best (warm colors), making patterns immediately visible.

Filtering to a region of interest

Click and drag on any axis to create a selection brush. Only runs whose values fall within the brushed range on that axis are highlighted; others dim. Brushing multiple axes narrows the selection to runs that satisfy all brushed ranges simultaneously.

This is the primary way to answer questions like “which combinations of learning rate and batch size produce accuracy above 90%?”

Reading the chart

• Lines that cross between two adjacent axes indicate an inverse relationship between those parameters.
• Lines that run parallel between two axes indicate a positive correlation.
• A tight bundle of lines converging on a good metric value marks the optimal hyperparameter region.

Using parallel coordinates with sweeps

Parallel coordinates are especially powerful after a sweep. All sweep trials appear as runs in the chart, making it easy to visualize the full search space explored and identify the most promising hyperparameter region for a follow-up targeted sweep.

To see only sweep runs, apply a filter (Filter → Sweep → your-sweep-id) before opening or refreshing the panel.

---

Runs Experiments Workspace Sweeps