Error bands in charts

By default, Neptune displays the minimum and maximum values of the downsampling range as the shaded area on charts. For chart widgets in dashboards and reports, you can disable this behavior or define your own custom error bands.

Types of error bands

Neptune doesn't compute the error bands. Instead, you can configure the already logged metrics to show as custom error bands for the Y-axis series. For example, you can log your own confidence interval as a series and then visualize it in the Neptune app.

The following types of error bands are available:

Type	Description
Auto min-max	Default setting. The bounds represent the minimum and maximum values of the downsampling range. For details, see the rendering algorithm.
Symmetrical	The bounds of the error band are relative to the main metric. In a chart, the shaded area is the result of the main metric ± the band metric.
Two-sided	Two band metrics provide the absolute bounds of the error band. In a chart, the area between the upper bound metric and the lower bound metric is shaded.

note

• If the error band metrics don't match the steps of the the main metric, Neptune displays approximations.
• Chart legend doesn't show metrics that represent error bands.
• Smoothing doesn't affect custom error bands.

Log error bands

To log the main metric and the error band metrics, use the log_metrics() function:

Symmetrical configuration

from neptune_scale import Run


run = Run(experiment_name=...)

for step in epoch:
    # your training loop
    run.log_metrics(
        data={"ci_mean": 84.14, "ci_var": 0.18},
        step=step,
    )

Two-sided configuration

from neptune_scale import Run


run = Run(experiment_name=...)

for step in epoch:
    # your training loop
    run.log_metrics(
        data={"ci_mean": 84.14, "ci_upper": 85.9, "ci_lower": 82.3},
        step=step,
    )

Show custom error bands in charts

To add custom error bands to a chart:

1. In a dashboard or report, add a new chart widget or edit an existing one.

2. In Advanced settings, enable Custom error bands.

3. For each Y-axis series plotted on the chart, use the dropdown to configure the bounds:

   • For Symmetrical, select an additional metric that represents the bounds.
   • For Two-sided, select two additional metrics that represent the upper and lower bounds.

note

• You can't configure custom error bands for metrics that are the result of custom expressions or dynamic selection.
• You can't configure such series as error bands either.

Hide error bands in charts

To hide error bands in charts:

1. In a dashboard or report, add a new chart widget or edit an existing one.
2. In Advanced settings, enable Custom error bands.
3. For each Y-axis series plotted on the chart, from the dropdown, select Disabled.