Generator Configuration

Per-Scenario Configuration

To enable SIERRA to generate experiment definitions based on the --scenario you specify, you need to:

Create generators/scenario_generator_parser.py in your --project directory.

Within this file, you must define the ScenarioGeneratorParser class with the following signature:

import typing as tp

class ScenarioGeneratorParser():
    def __init__(self):
        ...

    def to_scenario_name(self, args) -> tp.Optional[str]:
        """
        Parse the scenario generator from cmdline arguments into a
        string. Should return None if ``args`` is None (stage5).
        """
        ...

    def to_dict(self, scenario_name: str) -> str:
        """
        Given a string (presumably a result of an earlier cmdline parse),
        parse it into a dictionary of components: arena_x, arena_y,
        arena_z, scenario_tag

        which specify the X,Y,Z dimensions of the arena a unique tag/short
        scenario name unique among all scenarios for the project, which is
        used which creating the SIERRA runtime directory structure.
        """
        ...

Create generators/scenario_generators.py in your --project directory.

Within this file, you must define at least one function (and presumably one or more classes representing the scenarios you want to test with), which is gen_generator_name(), which takes the --scenario argument and returns the string of the class name within scenario_generators.py that SIERRA should use to generate scenario definitions for your experiments:
```
def gen_generator_name(scenario_name: str) -> str:
    ...
```
Each generator class within scenario_generators.py must define the generate() function like so:
```
class MyScenarioGenerator():
    ...

def generate(self):
    ...
```

Per-Experimental Run Configuration

In order to hook into SIERRA stage 1 experiment generation (doing so is optional), you need to:

Create generators/exp_generators.py in your --project directory.
Define a ExpRunDefUniqueGenerator class in this file, overriding the generate() function with your customizations. Your class really should be derived from a platform generator (e.g., PlatformExpRunDefUniqueGenerator) and override the generate() function, though you don’t have to.