FitTeD tutorials¶
Eight notebooks. Read 00–06 in order; 07 is a reference to dip into. Every one ships with its outputs already computed, so you can read them all without running anything — and every code cell is real, so you can re-run them and get the same figures.
The tutorials¶
| # | Tutorial | What it covers | Runs in | Needs manyTDE |
|---|---|---|---|---|
| 00 | Introduction and installation | What FitTeD does, how to install it, a first light curve | seconds | no |
| 01 | Data loading basics | Data_Set, manyTDE, your own arrays, cutting and rebinning |
seconds | optional |
| 02 | Observer-frame data input | Fluxes, AB magnitudes, X-ray fluxes, and the cosmology FitTeD does for you | seconds | no |
| 03 | Model setup and parameters | The disc parameters, and physical vs sampled variables | seconds | no |
| 04 | Basic fitting workflow | Priors, best_fit(), and why which data you include matters more than the optimiser |
~1 min | no |
| 05 | MCMC fitting with emcee | Running a chain, sizing it, checking convergence, saving and resuming | ~5 min | no |
| 06 | Complete workflow: AT2019dsg | All six bands, end to end, on a real TDE | ~35 min | yes |
| 07 | Model options reference | Every accuracy and performance option, and when to change it | ~3 min | optional |
Timings are for a laptop with a handful of cores; the MCMC notebooks parallelise across walkers, so more cores is faster.
Suggested path¶
If you are new to the package, read 00, 01, 03, 04, 05 in that order. That is the complete arc — install, data, model, fit, posterior — and none of it needs optional dependencies. Then 06 for the full multi-band analysis, which does need manyTDE.
Do not skip 03
It explains the difference between the parameters the physics uses (m_disc in solar
masses, incl in degrees) and the parameters the sampler moves (log_m_disc,
cos_incl). Getting these confused is by far the most common way to end up with a
chain that will not start, and the fix — model.pack_parameters() — takes one line
once you know about it.
If you have your own photometry, add 02.
Before you spend a week of CPU time, skim 07. One of the options in it can silently ruin a long-baseline fit.
A note on what these show¶
Tutorials 04, 05 and 06 fit real data and report what they actually get, including where that is unflattering: chains that have not converged are reported as unconverged, and notebook 04 shows a fit landing an order of magnitude from the right answer before showing what fixes it.
Notebook 06 is explicit about two data-preparation traps that bias results — cutting late-time data, and dropping negative fluxes from host-subtracted photometry. Both are easy to do by accident, and both push the answer the same way.
Running them¶
The notebooks add the parent directory to sys.path, so they work from a bare checkout as
well as from an installed copy. If FitTeD is installed (pip install -e . in the repository
root) the installed copy takes precedence.
To re-execute the whole set from the command line, use the script rather than doing it by hand — it runs them in order and then checks that every published notebook came out with its outputs intact:
Drafts¶
drafts/ holds notebooks that are not part of the published set and have never been
executed. See drafts/README.md. They are excluded from the website, and the docs build
fails if any of them is ever published by accident.
Requirements¶
- Python 3.10 or newer
- FitTeD and its dependencies (see Tutorial 00)
- Jupyter
- manyTDE for Tutorial 06:
git clone https://github.com/sjoertvv/manyTDE && cd manyTDE && pip install -e .
Getting help¶
model.what_pars()prints what any model instance expects, in order — start there when a parameter vector is rejected.- The paper has the physics: arXiv:2408.15048
- Example scripts:
fitted/examples/
Citation¶
Mummery et al. (2025), Fitting transients with discs (FitTeD): a public light curve and spectral fitting package based on evolving relativistic discs, MNRAS 544, 2225. arXiv:2408.15048 · ADS
BibTeX is in cite_fitted.bib in the repository root.