Monty.Frameworks.Models.GoalGeneration (ExTbpMonty.Core v0.1.0)

Goal-state generators: what a learning module wants to look at next.

Mirrors tbp/monty/frameworks/models/goal_generation.py, which holds one error and two generators:

...GoalGeneration.ParentLMNotProvided — ported: the error a generator raises when asked about its learning module before it has one;
...GoalGeneration.GraphGoalGenerator — ported: the base generator, which checks whether the goal it last emitted was reached and emits the next one; on its own it emits nothing, meaning "just be more certain";
...GoalGeneration.EvidenceGoalGenerator — ported: the generator an evidence-based module uses to pick where to look, from the mismatch between the graph of the object it believes in and the graph of the one it cannot rule out.

The behaviour these implement is Monty.Frameworks.Models.AbstractMontyClasses.GoalGenerator.

Summary

Functions

from_config(config)

The generator a config entry names, built from its own keys.

Functions

from_config(config)

@spec from_config(map()) :: {:ok, struct()} | {:error, term()}

The generator a config entry names, built from its own keys.

Monty instantiates the entry through Hydra and hands the learning module the object; a module built from a config has to instantiate the entry itself, which is what this does. A class this port has not written is an error rather than a silent nil, so building a module cannot quietly drop the goals the experiment asked for.