Tutorial

You installed mujpy. Find out below how to * run the simplest fit from the graphic interface * explore the potential in the automatic demos

Basic GUI editor usage

Get a GUI

  • Linux:

    • activate the venv. Choose the root of your data analysis, say mujpy/sample1. Create subfolder data and download your data files in it. [1]

    • cd back to sample1 and launch jupyter lab. A tab opens in your default browser.

  • Windows:

    • Choose the root of your data analysis in the Windows filesystem, say C:\Users\myhome\mujpy\sample1 and place your data in subfolder data. [1]

    • Start a Ubuntu terminal, activate the venv and cd to the Windows root folder, /mnt/c/Users/myhome/mujpy (or mujpy/sample). Launch jupyter lab --no-browser. Copy one of the two urls that jupyterlab provides and paste it in a Windows browser. A tab

Now launch a Python3 notebook. Write

%matplotlib qt

in the first cell (hover the green box for a code copy widget). This activates the qt backend for graphic popup windows, fine for smallish screens. If you prefer tabbed graphics type instead

%matplotlib ipympl

Then move to the second cell and type

from mujpy.mudashed import dashed
the_dash = dashed()

Rename this notebook e.g. Mudashed.ipynb. From now on you can use this same notebook to run all your fits.

Use it

Assume you are running an \(\alpha\) calibration in TF geometry, e.g. gps run 822, 2021. Click on the first cell and press twice \(\blacktriangleright\) in the notebook bar. Each time you want to restart from scratch Mudashed do the same). Three rows of widgets appear

_images/Mudashed-0.png

You can now retrieve run 822 from the Fetch data tab. Check that Group 0 is OK for you (it is, for 822). Then press DL and choose a prototype data file. Now type the run number (no leading zeros) in the run list box and press Enter. A new row of widgets appears. New info has appeared in the fist two widget rows, explore them. Notice that most widgets have tips shown when the mouse hovers on them (with browser in focus!).

_images/Mudashed-1.png

Type your chosen model acronym in the box, e.g. almg, calibrate \(\alpha\), and press Enter. The model editor appears.

_images/Mudashed-2.png

  • Component al, on the left side has one parameter \(\alpha\). Select a guess value, say 1.0. Forget Flag and Function, for now.

  • mg is a precessing component with Gaussian damping. Parameter names are self evident, B is mT, \(\phi\) degrees. The figure has easonable guess values. Chech the FR box, fit range and binning, and press Fit to get your minimization.

The gui switched to the log tab to show a printed log and a plot appeared. Notice that one mg component is not good enough, explore the plot and the log! Read below and see Reference for more details on the model editor (but be patient, it’s WIP!)

Automatic demos

If you want to explore further mujpy potentials open an Ubuntu terminal, activate the venv, move to a suitable empty folder (e.g. cd /tmp) and run

python -m mujpy.tests.tests

The tests automatically produce 16 distinct types of mujpy fits on a standard gps TF experiment (runs 822-834). The model chosen is the sum of two relaxing cosines, a Gaussian component, mg, and a Lorentzian component ml, identified by the model acronym: mgml. Follow below to understand their features:

  • A plot opened: you see a TF asymmetry with its fit. Check the note below, and kill the window when done.

Note

This is a simple A0 fit. Additional helpers catch goodness of fit: the residues, lower panel, an info box, top right, and \(\chi^2\) histogram of the fit standard deviations, with the expectation curve, bottom right. More details in the terminal output. All the information is stored in log (ASCII), csv and json format, for plotting and reproducibility. When you kill the plot you also get a test summary.

  • Another plot opened: same scenario, the A20 fit. Two orthogonal groups of detectors have been independently fitted. The log reports two reduced \(\chi^2\) values.

Note

This is an animation: you see both fits, in an infinite loop. Click on the plot to toggle stop/start. The terminal shows two compact logs (replicated in subfolder cache/): uniquely named parameters and their best values are also added to a specific csv (subfolder csv/). More details elsewhere in these docs.

  • B1 fit: 4 independent A1 fits on a list of runs. Approaching a phase transition below 12.0 K

Note

The animation loops on four independent fits. Actually the \(\chi^2\) is not very good at 12.0K, something else is going on…

  • B20 fit: 8 independent A20 fits on the same list of runs. Discover in which order from the plot and check on the log.

  • A21 global fit: same run and groups as A20, but a single cost function. Log lists the best values of the Minuit parameters (f,A34,A21 etc) before the usual componet parameters, derived from them (here, a few details are hidden …).

  • B21 global fits: 4 independent A21 fits for the list of runs. Check.

  • C1 global fit: list of 4 runs, only one group. Check that there is a single reduced \(\chi^2\) and list of Minuit parameters. The log shows also the components for each asymmetry (for each run).

Note

Each frame of the animation reports estimated \(\chi^2\) statistics for that asymmetry, in the info box. The only true reduced \(\chi^2\) is in the log.

  • C2 global fit, the full Monthy. A single cost funtion for 2 groups and 4 runs.

But there are further 8 fits. The previous ones had fixed \(\alpha\) ratios (calibrated somehow). The next ones, model almgml, include \(\alpha\) as a fit parameter for each group, in the very same sequence. Component al is listed with the others but it’s not additive, it’s treated differently. That’s how \(\alpha\) is calibrated. With A20, B1, B20 fits you get calibrations for each asymmetry, with C1, C2 fits you get a single value per group (be it the correct one or some average if conditions change!).

Now you can also re-run each test on its own. List them pasting

python -c "from mujpy.tests.tests import ParameterizedRunner as PR; rnr=PR(); rnr.list_tests()"

in the terminal. Execute e.g. test 0 (A1 almgml) by

python -c "from mujpy.tests.tests import ParameterizedRunner as PR; rnr=PR(); rnr.test_single(index=0)"

Warning

the test timings includes your watch time on the plot! It is not very significant.