Monitor a Campaign Live#
The LivePlotBO driver serves a dashboard for measurements and Bayesian
optimization results stored in Tiled. Start the standalone server with:
python -m AFL.automation.shared.liveplot
The default port is 5096. Open http://localhost:5096/liveplot to load
the campaign belonging to the newest timestamped Tiled run. Enter an
AL_campaign_name and use Refresh to select a different campaign.
The default source preset expects the color-matching workflow’s
OT2Prepare.prepare, RGBCamera.capture_rgb,
SeabreezeUVVis.measure, and DoubleAgentDriver.predict records.
Configuration#
Supply launcher overrides to pin a default campaign instead of discovering the latest one, or to change source names and variables. An ID entered in the dashboard takes precedence over this configured default. For example:
{
"campaign_id": "color-matching-082626-b3b865b5",
"objective_direction": "minimize",
"target_image": {
"entry_id": "QD-92c4c5f4-4a67-40c9-b528-a440eed2cd31",
"variable": "img_rgb"
},
"target_spectrum": {
"entry_id": "QD-d0832f67-15a4-43b4-8123-931010ec56bf",
"x_variable": "wavelength",
"y_variable": "extinction"
}
}
composition_source accepts either a metadata metadata_path or an
xarray variable. image_sources and spectrum_sources are lists, so a
dashboard can combine multiple instrument types. Each source selects records
with driver_name and task_name and may customize its displayed name and
dataset variable names.
The ordered dashboard_layout list controls every dashboard row. Each row
sets a title, column count, and a list of semantic panels. Supported panel
kind values are image, gallery, spectrum, spectra,
composition, progress, and design_space. source names an actual
top-level configuration block rather than a generated figure. For example, a
BO design-space panel uses "source": "bo_source" and therefore draws the
dataset variable explicitly named by its variable field, for example
"variable": "composition". Spectrum panels similarly declare
x_variable and y_variable. The BO driver uses the layout’s variables
when loading composition, score, best-composition, and suggested-composition
data. A state of
latest, suggested, best, or target selects a particular view
where a kind has multiple views. title sets the panel’s plot title.
Reorder, remove, or add panels without editing the driver or browser assets.
Multi-spectrum options belong directly on a spectra panel. Set
color_by to score, sample_order, none, or a composition key
such as stock_Red (composition.stock_Red is also accepted). Rendering
uses consistent built-in styling and a Viridis color scale.
Component names are read directly from the BO xarray dataset’s configured
component coordinate. composition_plot.component_names is only an optional
fallback for datasets that genuinely have no such coordinate. Display text can
be changed independently with component_labels and strip_prefixes.
These settings apply consistently to current, next, and best composition bars
and to every design-space axis or distribution title.
The dashboard loads once when opened and only updates when Refresh is pressed. Image arrays are fetched lazily while spectra and Plotly figure data are assembled by the driver. Missing or malformed sources are reported in the dashboard without hiding plots that could be constructed successfully.