dguest / scharm-fitter Goto Github PK

This package includes all the code to calculate CLs upper limits for direct production of supersymmetric charm quarks decaying to charm jets.

The aim is to provide:

1. A common script that can read yields (in a textfile format) and construct workspaces to test various signal points.
2. A script to run the fit on an existing workspace and save the fit results to a textfile.

Less important goals may include:

• Scripts to perform sanity checks on the inputs / outputs,

What's not currently included:

• The suite of HistFitter scripts like YieldsTable, SysTable, etc.
• Plotting code.

Running install.py install will add a .pth file to your local python installation. It will also run make in the src/ directory to build the HistFitter fitting functions. All top level scripts are in the scripts directory:

• susy-fit-*: try the -h flag to get help.
• susy-fit-test.py: this segfaults on some computers I use, even though it's doing very little. I suspect it has something to do with a bad pyroot install, but if it fails, everything else here will as well.

Example inputs are in example_data/. If scripts/ has been added to your PATH, the following command should produce some workspaces:

cd example_data
susy-fit-workspace.py inputs.yml -c configuration.yml -s

Adding the -s flag will produce more workspaces, including the _afterFit.root and _upperlimits.root files.

Input files should be formatted as follows:

nominal_yields:
  REGION_NAME:
    BACKGROUND: [YIELD, STATISTICAL_UNCERTAINTY]
    ....
    scharm-SCHARM_MASS-LSP_MASS: [YIELD, STATISTICAL_UNCERTAINTY]:
  REGION_NAME:
    ....

yield_systematics:
  SYSTEMATIC:
    REGION_NAME:
      BACKGROUND: [YIELD]
    ....

relative_systematics:
  SYSTEMATIC:
    REGION_NAME:
      BACKGROUND: [DOWN, UP]
  ....

The REGION_NAMEs are completely arbitrary, since the fit treats all regions identically (except when the signal region is blinded and the MC SM sum is used in place of real data).

Some of the SYSTEMATIC categories under yield_systematics will be treated in special ways:

• The b-tagging systematics (names starting with b, c, u, or t and ending with up or down, i.e. bup, udown etc...) will be added in quadrature before fitting. (will probably add warnings if all these backgrounds aren't found).
• Other names that end in up or down will be paired to give an asymmetric uncertainty.
• Any other (unpaired) systematics will be entered as a symmetric uncertainty centered on the nominal value.

Signal points must be of the form "scharm- scharm mass - lsp mass". The masses can be arbitrary integers. The BACKGROUND names are arbitrary.

Values given by YIELD and STATISTICAL_UNCERTAINTY should be absolute. Relative uncertainties are specified with 1.0 meaning "no variation", e.g. a variation that is expected to fluctuate by 20% in each direction would be [0.8, 1.2].

An additional "fit config" file is required as an input to the workspace creation routine. This is formatted as:

CONFIG_NAME:
  control_regions: [REG1, REG2, ...]
  signal_region: SIG_REGION
  combine_tagging: TRUE_OR_FALSE

This file will be created (although not necessarily with sensible regions) if it doesn't exist. The option combine_tagging tells the fitter whether it should add the flavor tagging systematics in quadrature.

• The code may not be very robust to incorrectly formatted files. Invalid yaml won't get through, but no promises about anything that doesn't conform to the above schema.
• The workspace creation routine prints a lot of errors of the form: ERROR argument with name nom_something is already in this set. These are probably just harmless overwrites, since the nominal value shouldn't be use in the fit, but it should be checked. For now these errors are being filtered from the output stream.
• Workspace creation produces about 5 files, only one of which we seem to need. Right now I'm deleting the others, should make sure this is safe.
• Figure out whether the sample.ActivateStatError() function is needed in our case. If so, figure out how to keep it from crashing.