Skip to main content

Module classifiers

Module classifiers 

Source
Expand description

First-class genome classifier assembly.

A classifier is packaging around two memory areas joined by associative_memory. It is stored under the top-level genome key classifiers, parallel to brain_regions. It is not a brain region and is not exportable as a circuit.

External inputs are ordinary episodic connections selected by connectivity rule:

  • classifier_sample lands on the sample memory
  • classifier_class lands on the class memory and creates one class output whose shape matches that class input

The sample memory accepts any number of those inputs, including a mix of memory and interconnect areas. The class memory does the same. Pattern identity is the memory area’s own episodic hash. The classifier does not add a second input rule on top of that.

Structs§

ClassPairFiring
One class input/output pair observed on a single burst.
Classifier
First-class classifier record persisted in the genome.
ClassifierClassBinding
One class input and the output area created for it.
ClassifierMapping
Directed mapping owned by a classifier assembly.
ClassifierMappingEffect
Result of recording one mapping edit on a classifier.

Constants§

CLASSIFIER_ASSOCIATIVE_MORPHOLOGY
Morphology used for sample-memory → class-memory bind.
CLASSIFIER_CLASS_MORPHOLOGY
Connectivity rule: episodic input into the classifier class memory.
CLASSIFIER_SAMPLE_MORPHOLOGY
Connectivity rule: episodic input into the classifier sample memory.
CLASS_MEMORY_REPLAY_BURSTS
Bursts from the class-memory replay neuron firing to the class output firing.
MEMORY_NEURON_FIRING_LATENCY_BURSTS
Bursts between a memory area receiving its upstream input and its memory neuron firing.

Functions§

class_output_area_name
Visible name of one class output.
class_output_reports_error
True when this class output should raise the classifier error.
classifier_answer_delay_bursts
Bursts from a class input firing to its class output answering it: the hop into class memory (at least one burst), plus the memory neuron’s firing latency, plus the replay onto the output. With the default one-burst rule delay this is 3.
classifier_mapping_rule
Mapping rule written for one classifier-owned edge.
classifier_reports_correct
True when at least one class input fired and every class output answered it without error.
classifier_reports_error
True when any class output of this classifier raises the error.
classifier_reward_eligibility_bursts
Bursts from co-activity on the associative mapping until the answer reaches it as reward: the answer delay, one burst for the correct or error area to fire, and one for the modulator to commit its signal.
morphology_is_episodic_input
True when a mapping rule is an episodic encode into a memory area.