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

Copy Markdown View Source

Three limits in one policy: matching, exploring, and when to start exploring.

Mirrors tbp/monty/experiment/recognition_strategy.py::StepLimit. Which limit applies depends on what the model is doing:

  • exploring — exploring_steps against num_exploring_steps;
  • training — matching_steps against max_train_steps, and once min_train_steps have matched it also asks to start exploring, which is how a training episode moves on to the next view of an object it knows;
  • evaluating — matching_steps against max_eval_steps.

Summary

Functions

call(arg1, counter, state)

The limit that applies to what the model is doing right now.

new(options)

A policy with the three limits.

Functions

call(arg1, counter, state)

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

The limit that applies to what the model is doing right now.

new(options)

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

A policy with the three limits.

Monty raises ValueError for a negative minimum or a non-positive maximum; here that is {:error, message} with Monty's own words, checked in the same order Monty checks them.