Difference between revisions of "Calibration Train"
From GlueXWiki
(→Runs to Use) |
(→Runs to Use) |
||
| Line 24: | Line 24: | ||
# Other big sets of runs: | # Other big sets of runs: | ||
#* Fall 2014: 1516-1520, 1769-1777, 1900, 2179-2184, 2205, 2228, 2407-2409, 2416-2420 | #* Fall 2014: 1516-1520, 1769-1777, 1900, 2179-2184, 2205, 2228, 2407-2409, 2416-2420 | ||
| + | #* Look for groups of runs with similar conditions | ||
# Others | # Others | ||
| + | |||
| + | |||
This also raises the question: what are the good runs? | This also raises the question: what are the good runs? | ||
| − | One | + | One proposal: |
| + | * Tag each run in monitoring database with one of three values: | ||
| + | *# 0 - non-production quality | ||
| + | *# 1 - production quality | ||
| + | *# 2 - production quality, used for calibrations | ||
| + | * Idea is that > 1 means that the run is good to use for physics | ||
| + | ** Finer grained information can be stored in RCDB | ||
| + | * To determine production quality, develop quality metrics for each subdetector, use combination of quality metrics and eye test | ||
Revision as of 18:11, 19 May 2015
Contents
Proposal
- Use case: Streamline tape library usage, provide common environment for production and development
- Example: Users develop plugins on their favorite files/run, use this for running over larger data
- Run every week (Wednesday to avoid conflict with monitoring?)
- Uses subset of runs
- Users provide:
- DANA plugin
- Optional post-processing scripts to be run after every/all runs
- Curated in SVN
- Results stored in standard location(s)
- Possible results: ROOT files, images, calibration constants, web pages
- Uses SWIF? (buzzword compliance)
Runs to Use
Several possibilities:
- Large/Popular Runs
- Fall 2014: 1514, 1769, 1777, 1787, 1803, 1807, 1810, 1825, 1847, 1852, 1854, 1871, 1872, 2138, 2206, 2207, 2209, 2223, 2397
- Spring 2015: 2931, 3079, 3179, 3180, 3183, 3185
- Other big sets of runs:
- Fall 2014: 1516-1520, 1769-1777, 1900, 2179-2184, 2205, 2228, 2407-2409, 2416-2420
- Look for groups of runs with similar conditions
- Others
This also raises the question: what are the good runs?
One proposal:
- Tag each run in monitoring database with one of three values:
- 0 - non-production quality
- 1 - production quality
- 2 - production quality, used for calibrations
- Idea is that > 1 means that the run is good to use for physics
- Finer grained information can be stored in RCDB
- To determine production quality, develop quality metrics for each subdetector, use combination of quality metrics and eye test
