Monty.Experiment.RecognitionStrategy.ObjectRecognition (ExTbpMonty.Core v0.1.0)

Monty's own object-recognition policy: three ways to finish an episode.

Mirrors tbp/monty/experiment/recognition_strategy.py::ObjectRecognition, and is what termination/any1_objectrecognition_t1000_e500_tot6000 configures. In Monty's order:

the mode's matching limit is reached and the model is not exploring — done, and time the modules out beside it;
the total step ceiling is reached — done, and time the modules out;
the model says it is done — done, and nothing to time out.

The order matters because the first two also time out the modules: an episode stopped by a limit still owes an answer, while an episode that recognised its object already has one.

Summary

Functions

call(model, counter, state)

The first of the three conditions that holds, in Monty's order.

new(options)

A policy with a matching limit per mode and a total ceiling.

Functions

call(model, counter, state)

@spec call(map(), Monty.Experiment.RecognitionStrategy.RecognitionCounter.t(), map()) ::
  Monty.Experiment.RecognitionStrategy.RecognitionResult.t()

The first of the three conditions that holds, in Monty's order.

new(options)

@spec new(keyword()) :: {:ok, {module(), map()}} | {:error, String.t()}

A policy with a matching limit per mode and a total ceiling.

Monty raises ValueError for any limit that is not positive; here that is {:error, message} with Monty's own words, checked in the same order.