Config Registration

After defining a new task and/or optimizer, you need to register it with judo so that it can be used in the GUI and show up in the dropdown menus. This can be done by running the following in a script:

from judo.cli import app
from judo.optimizers import register_optimizer
from judo.tasks import register_task

if __name__ == "__main__":
    register_optimizer("my_optimizer_name", MyOptimizer, MyOptimizerConfig)
    register_task("my_task_name", MyTask, MyTaskConfig)
    app()

If you instead want to use the judo CLI command to start the app, you can register the task and optimizer using a hydra config. We provide a convenient interface to do so. Consider this example:

defaults:
  - judo  # you must have this default!

task: "my_cylinder_push"
optimizer: "my_cem"

# these custom tasks/optimizers are defined in the judo/examples folder
# they are nearly identical to the regular cylinder push task and cem optimizer,
# but we add an extra parameter to each of their configs for demonstration
custom_tasks:
  my_cylinder_push:
    task: judo.examples.example_task.MyCylinderPush
    config: judo.examples.example_task.MyCylinderPushConfig
  # we can keep passing in more custom tasks here...

custom_optimizers:
  my_cem:
    optimizer: judo.examples.example_optimizer.MyCrossEntropyMethod
    config: judo.examples.example_optimizer.MyCrossEntropyMethodConfig
  # we can keep passing in more custom optimizers here...

Similarly, you can run the benchmarking program on custom registered tasks/configs in a similar way, replacing app with benchmark either programmatically or in the hydra YAML file.