Monty.Frameworks.Models.EvidenceMatching.HypothesesUpdater.DefaultHypothesesUpdater (ExTbpMonty.Core v0.1.0)

Copy Markdown View Source

The straightforward updater: accumulate evidence, start channels, displace.

Mirrors tbp/monty/frameworks/models/evidence_matching/hypotheses_updater.py::DefaultHypothesesUpdater, the updater Monty pairs with the plain displacer — no bursting. Every step it brings any channel that has just started reporting into the hypothesis space, then adds this observation's evidence to the hypotheses already there.

Starting hypotheses

Two ways, chosen by whether the experiment seeds poses:

  • informed — each node's stored pose is aligned with what the sensor sees, once per direction the sensed pose allows. This is how Monty recognises an object turned any way: the alignment tells it which rotation would explain the observation.
  • fixed — every node is tried with each configured rotation, which is what a uniform experiment means.

All of them start with the evidence the first observation already gives them: the feature evidence of the node they sit on. Nothing has been ruled out yet — every node is a hypothesis — but a node whose colour and curvature look like what the sensor just saw starts ahead of one that does not, which is what Monty's first hypothesis space is worth.

A channel that joins late

A sensor can start reporting after another channel has already accumulated evidence. Its hypotheses enter with the mean of the evidence already there, so a newcomer is not silently out of the running against hypotheses that have had several steps to climb.

Summary

Types

t()

Functions

begin_step(updater)

Nothing to prepare: this updater keeps every hypothesis it has and spends no step sampling new ones, so each step looks like every other.

displace_hypotheses(default_hypotheses_updater, hypotheses, displacement)

Move every hypothesis by displacement, as its own orientation sees it.

end_step(updater)

Nothing to wind down, for the same reason begin_step/1 has nothing to set up.

new(memory, options)

An updater over memory.

reset(updater)

Forget which channels this episode has started.

update_evidence(updater, hypotheses, features, graph_id, evidence_update_threshold)

Bring the hypothesis space up to date with this observation.

Types

t()

@type t() ::
  %Monty.Frameworks.Models.EvidenceMatching.HypothesesUpdater.DefaultHypothesesUpdater{
    displacer: term(),
    graph_memory: term(),
    initial_possible_poses: term(),
    initialised_channels: term(),
    scorer: term(),
    umbilical_num_poses: term()
  }

Functions

begin_step(updater)

@spec begin_step(t()) :: t()

Nothing to prepare: this updater keeps every hypothesis it has and spends no step sampling new ones, so each step looks like every other.

displace_hypotheses(default_hypotheses_updater, hypotheses, displacement)

@spec displace_hypotheses(
  t(),
  Monty.Frameworks.Models.EvidenceMatching.Hypotheses.t(),
  [number()] | nil
) :: Monty.Frameworks.Models.EvidenceMatching.Hypotheses.t()

Move every hypothesis by displacement, as its own orientation sees it.

end_step(updater)

@spec end_step(t()) :: t()

Nothing to wind down, for the same reason begin_step/1 has nothing to set up.

new(memory, options)

@spec new(Monty.Frameworks.Models.GraphMatching.GraphMemory.t(), keyword()) :: t()

An updater over memory.

:initial_possible_poses is :first_observation for informed sampling, or a list of 3×3 pose matrices for fixed poses.

reset(updater)

@spec reset(t()) :: t()

Forget which channels this episode has started.

update_evidence(updater, hypotheses, features, graph_id, evidence_update_threshold)

@spec update_evidence(
  t(),
  Monty.Frameworks.Models.EvidenceMatching.Hypotheses.t(),
  %{optional(String.t()) => map()},
  String.t(),
  float()
) :: {t(), Monty.Frameworks.Models.EvidenceMatching.Hypotheses.t() | nil, map()}

Bring the hypothesis space up to date with this observation.

Starts the space on the first step, adds a channel that has just begun reporting, and otherwise adds the observation's evidence to what is there. Returns nil and no telemetry when no stored channel was observed.