Sometimes you want to run a parameter sweep. A parameter sweep is a method of evaluating a function (or a program) with a pre-determined set of parameters. The examples below will clarify what this means.
To run a parameter sweep, use the
-m) flag and pass a comma separated list for each
dimension you want to sweep.
To run your program with the 3 different schemas in schema config group:
Here is sweep over the db types (mysql,postgresql) and the schemas (warehouse,support,school). Output does not contain the configuration prints.
Hydra supports other kind of sweeps, for example a range sweep: x=range(1,10) or a glob: support=glob(*).
See the Extended Override syntax for details.
The sweeping logic is implemented by a simple sweeper that is built into Hydra. Additional sweepers are available as plugins. For example, the Ax Sweeper can automatically find the best parameter combination!
A Launcher is what runs your job. Hydra comes with a simple launcher that runs the jobs locally and serially. Other launchers are available as plugins for launching in parallel and on different clusters. For example, the JobLib Launcher can execute the different parameter combinations in parallel on your local machine using multi-processing.