Psi4 Dashboard

A multi-page Dash web app for exploring Psi4 benchmark data — timer breakdowns, SCF iteration counts, and parallel scaling — across Psi4 versions and core counts.

Pages

Route	Page	What it shows
/	Home	Landing page.
/tests	Tests	Per-test line charts of a timer metric (wall/user/system time) across versions. Non-leaf timers expand into a stacked-area breakdown of their child timers.
/timers	Timers	A selected timer's metrics across tests, over versions.
/scf	SCF	Filled-area plots of SCF iterations per label across versions, each with a collapsible per-accelerator breakdown.
/parallelism	Parallelism	Wall-time-vs-core-count charts per timer, for a chosen version and test.

The Tests, Timers, and Parallelism pages share a level slider (to walk the timer hierarchy), a selection dropdown, and — for the timer pages — a metric tab strip. Slider position and selections are reflected into the URL query string, so a shared link restores the view.

Data

All data is read at startup from a sibling psi4dashboard-data repo, at the path hardcoded as DATA_DIR in data.py (../psi4dashboard-data/data/, relative to where you launch the app). Each loader recursively searches that directory for its files:

Dataset	Source files	Grain
timing	timer.json	one row per timer per test run
scf	scf_iterations.json	one row per (test run, SCF label)
parallelism	<test>.json.n<cores>.out	one row per timer per core count

Malformed or empty source files are skipped; if every file for a dataset is skipped, that dataset loads as an empty frame (the app still starts). See the module docstring and create_*_df parsers in data.py for the full per-column schemas.

Layout

dashboard.py     App entry point: loads the datasets and builds/runs the Dash app.
data.py          Data loading, parsing, and runtime state (module-global DataFrames).
callbacks.py     Callback logic and chart-building helpers shared across pages.
components.py    Reusable Dash layout builders (slider, dropdown, tabs, chart grid).
theme.py         Shared Plotly dark-theme styling.
pages/           One module per route (home, tests, timers, scf, parallelism).
assets/          CSS served automatically by Dash.

Requirements

• Python 3
• dash, plotly, pandas, packaging

Install into a fresh virtual environment or conda environment, e.g.:

pip install dash plotly pandas packaging

Running

Check out the psi4dashboard-data repo alongside this one so that ../psi4dashboard-data/data/ resolves, then from the project root:

# Defaults to Dash's built-in host/port (http://127.0.0.1:8050)
python dashboard.py

# Or bind an explicit address
python dashboard.py --address 127.0.0.1:8050

The app runs with debug=True, so it hot-reloads on source changes.

License

Released under CC0 1.0 Universal (public domain dedication).