Statistics reference
This page contains detailed documentation on statistics utilities provided for use in Cumulus studies.
Specific stats modules
General usage guidelines
You can invoke a statistic task from your study’s manifest the same way that you would run SQL or python files - the only difference is that you point it at another toml file, which allows stats configs to have different input parameters depending on the analysis you’re trying to perform.
In your manifest, you’d add a section like this:
[statistics_config]
file_names = [
"psm_config.toml"
]
We’ll use this as a way to load statistics configurations. Since some of these statistical methods may be quasi-experimental (i.e. perform a random sampling), we will persist these outputs outside of a study lifecycle.
The first time you run a build against a study with a statistics config that has not previously been run before, it will be executed, and it should generate a table in your database with a timestamp, along with a view that points to that table. Subsequent updates will not replace that data, unless you provide the --statistics argument. If you do, it will create a new timestamped table, point the view to your newest table, and leave the old one in place in case you need to get those results back at a later point in time.
When you clean a study with statistics, by default all statistics artifacts will be ignored, and persist in the database. If you want to remove these, the --statistics argument will remove all these stats artifacts.