repeated staterror across channels

[andrewfowlie]

Summary

I expected that staterror modifiers with the same name, across different channels, would be treated as correlated. Instead, the second staterror modifier parameters get appended to the first lot.

This isn't what I expected to happen. Not sure if it is a bug or not.

OS / Environment

NAME="Fedora Linux"
VERSION="40 (Forty)"
ID=fedora
VERSION_ID=40
VERSION_CODENAME=""
PLATFORM_ID="platform:f40"
PRETTY_NAME="Fedora Linux 40 (Forty)"
ANSI_COLOR="0;38;2;60;110;180"
LOGO=fedora-logo-icon
CPE_NAME="cpe:/o:fedoraproject:fedora:40"
DEFAULT_HOSTNAME="fedora"
HOME_URL="https://fedoraproject.org/"
DOCUMENTATION_URL="https://docs.fedoraproject.org/en-US/fedora/f40/system-administrators-guide/"
SUPPORT_URL="https://ask.fedoraproject.org/"
BUG_REPORT_URL="https://bugzilla.redhat.com/"
REDHAT_BUGZILLA_PRODUCT="Fedora"
REDHAT_BUGZILLA_PRODUCT_VERSION=40
REDHAT_SUPPORT_PRODUCT="Fedora"
REDHAT_SUPPORT_PRODUCT_VERSION=40
SUPPORT_END=2025-05-13

Steps to Reproduce

bug.json

{
    "channels": [
        { "name": "firstchannel",
          "samples": [
            { "name": "signal",
              "data": [5.0, 10.0],
              "modifiers": [
                { "name": "k_staterror", "type": "staterror", "data": [0.1, 0.2] },
                { "name": "lumi", "type": "lumi", "data": null }
              ]
            }
          ]
        },
        { "name": "secondchannel",
          "samples": [
            { "name": "signal",
              "data": [5.0, 10.0],
              "modifiers": [
                { "name": "k_staterror", "type": "staterror", "data": [0.1, 0.2] },
                { "name": "lumi", "type": "lumi", "data": null }
              ]
            }
          ]
        }
    ],
    "observations": [
        { "name": "firstchannel", "data": [50.0, 60.0] }, { "name": "secondchannel", "data": [50.0, 60.0] }
    ],
    "measurements": [
        { "name": "Measurement", "config":
          {"poi": "lumi", "parameters": [ 
            { "name": "lumi", "auxdata": [1.0], "sigmas": [0.017], "inits": [1.0] }
          ]
          }
        }
    ],
    "version": "1.0.0"
}

Reproducer

import pyhf
import json

with open("bug.json") as f:
    spec = json.load(f)
    
ws = pyhf.Workspace(spec)
model = ws.model()

print(model.config.par_names)

Output:

['lumi', 'k_staterror[0]', 'k_staterror[1]', 'k_staterror[2]', 'k_staterror[3]']

File Upload (optional)

No response

Expected Results

I expected to have 2 k_staterror parameters (one for each bin) that were correlated across the two channels

['lumi', 'k_staterror[0]', 'k_staterror[1]']

Actual Results

['lumi', 'k_staterror[0]', 'k_staterror[1]', 'k_staterror[2]', 'k_staterror[3]']

The second k_staterror gets appended to the first, giving 4 k_staterror parameters in total.

### pyhf Version

```console
pyhf, version 0.7.6

Code of Conduct

• I agree to follow the Code of Conduct

[alexander-held]

I think the fact that staterror modifiers cannot be correlated across channels is probably on purpose (I don't think they can in ROOT HistFactory either?) but the behavior here is counterintuitive. Instead of having the bin number range in the parameter names cover 0–3 (for the four total bins) I would expect that this is an invalid model and would raise an error (which could point out that correlation is not supported). Then the modifier names would have to be different by channel.

[andrewfowlie]

sounds reasonable