Skip to content

DASHBOARD.md

Latest commit

 

History

History
115 lines (80 loc) · 6 KB

DASHBOARD.md

File metadata and controls

115 lines (80 loc) · 6 KB

TAS-AI Dashboard Guide

The TAS-AI Dashboard provides real-time visualization and control of autonomous neutron scattering experiments. It is implemented with Dash and runs as a lightweight web app (served locally, viewable in any browser).

Quick Start

# Start the dashboard (serves http://127.0.0.1:8050/)
python -m tasai.dashboard.app --port 8050

Then open http://127.0.0.1:8050/ in your browser (or use the tasai-dashboard console script which wraps the same entry point).

Dashboard Layout

The Dash application renders a single-page control room with the following regions:

  • Control bar – Start/Pause/Stop/Reset buttons, export button, queue utilities, and the help shortcut
  • Dispersion + Parameters – Streaming plot of ω(H,H,0) overlaid with measurements plus a parameter card with J₁, J₂, D (or Sunny order parameters) and χ²/N
  • Model weights & information gain – Side-by-side cards showing posterior probabilities, Bayes factors, and cumulative/instantaneous information gain
  • Acquisition heatmap – 2D (H,E) intensity/score map; click anywhere to create a candidate measurement
  • Measurement queue – Table of pending measurements with acquisition scores and statuses
  • Event log & toasts – Time-stamped log plus dismissible notifications for warnings/errors

Controls

Control Description
▶ Start, ⏸ Pause, ⏹ Stop Manage the autonomous loop. Pause keeps the queue intact; Stop clears the active measurement.
↺ Reset Clears measurements and queue, returning the planner to its initial state.
Add to Queue Push the current H, K, L, E, and count-time values (taken from the sliders/inputs) onto the queue.
Batch Add Paste multiple whitespace-separated lines (H K L E count) to enqueue several points at once.
Queue ↑/↓/🗑/Clear Reorder or remove queued measurements directly in the browser.
Export Download the current measurement log as CSV (includes timestamps, intensities, and fitted values).
? Help Opens a modal with tips and shortcut reminders.

Dispersion & Parameter Cards

  • The Dispersion plot overlays the current fit (solid line) with the latest measurements. Point colors encode the source (AI-selected, forced coverage, manual). Error bars show propagated uncertainty.
  • The Parameter card lists the current MAP estimates, 1σ uncertainties, χ²/N, and the total measurement count. Colors shift from amber to green as uncertainties shrink.

Model Weights & Evidence

  • Horizontal bars show the posterior probability of each candidate Hamiltonian.
  • A Bayes factor badge summarizes the evidence ratio vs. the nearest competitor (same interpretation as in the manuscript).
  • Once the winning model exceeds 95 % probability, the card highlights that the discrimination goal has been satisfied.

Information Gain

  • Cumulative bits – Running total of the Shannon information gathered.
  • Last step – How much the most recent point contributed.
  • Rate – Bits per minute to track efficiency.
    The progress bar turns green when the target entropy (set in the settings panel) is met.

Measurement Queue

Columns: queue index, (H,K,L), energy transfer, count time, acquisition score, and status.

  • 🟢 Measuring – the spectrometer is currently at this point.
  • 🟡 Queued – waiting its turn.
  • ✅ Complete – data has been folded into the fit.
  • 🟠 Paused – pending user confirmation (e.g., coverage seeds).

You can drag rows to reorder them or use the arrow buttons for precise moves. Clicking a row also surfaces its metadata in the sidebar.

Acquisition Heatmap

When the Sunny/analytic backend provides precomputed scores, the central heatmap displays score × intensity across H and energy. Clicking a pixel fills the Add-to-Queue inputs, allowing you to seed tactical points with a single tap.

Event Log & Notifications

A scrollable log at the bottom lists each measurement, posterior update, or warning with timestamps. Toast notifications (top-right) mirror major events so you can step away from the keyboard and still see when TAS-AI pauses or finishes.

Keyboard Shortcuts

The web app installs a small key listener that forwards a few essential shortcuts to the corresponding buttons:

Key Action
Space Toggle Pause/Resume (prefers Pause if available, otherwise Start)
Esc Stop immediately
A Focus the temperature/energy input so you can type a new value
C Clear the queue
E Open the export modal
R Reset the experiment
? Open the help modal

All other interactions (adding/removing points, editing settings, switching acquisition functions) are performed through the on-screen controls and are touch/mouse friendly.

Command-Line Options

python -m tasai.dashboard.app [OPTIONS]
# or
tasai-dashboard [OPTIONS]
Option Default Description
--host 127.0.0.1 Interface to bind (use 0.0.0.0 inside Docker or on remote machines)
--port 8050 TCP port for the HTTP server
--debug False Enable Dash hot reload / debugger (developer use only)

Troubleshooting

  • Dashboard does not start – Ensure Dash dependencies are installed (pip install dash dash-bootstrap-components plotly pandas).
  • Blank page – Confirm you’re visiting the URL printed to the terminal (default http://127.0.0.1:8050/). Refresh or try a different browser if the assets fail to load.
  • Slow performance – Close duplicate dashboard tabs, lower the queue size, or run the CLI without --debug.
  • Cannot reach the instrument – Verify the proxy server is running (python -m tasai.proxy_server.server) and that your firewall allows the configured host/port.

For additional questions open an issue on GitHub or run tasai-dashboard --help for the latest CLI options.

Dashboard version 1.0 | Built with Dash
⚡