brain (macula_tweann v0.18.1)

Brain process for real-time neural network inference.

A brain is a running neural network that accepts sensor inputs and produces actuator outputs. Unlike the process-per-neuron architecture used during training, this is a single GenServer optimized for real-time inference in applications like games and robotics.

Features

- Synchronous evaluation via evaluate/2 - Internal state persistence for LTC neurons - Visualization data for UI rendering - PubSub notifications for state changes

Usage

Start a brain with a network: Network = network_evaluator:create_feedforward(42, [16, 8], 6), {ok, Pid} = brain:start_link(#{network => Network})

Evaluate with sensor inputs: Outputs = brain:evaluate(Pid, Inputs)

Get visualization data: VizData = brain:get_viz(Pid)

Subscribe to viz updates (Erlang): brain:subscribe(Pid) receive {brain_viz, Pid, VizData} -> ... end

Summary

Functions

evaluate(Pid, Inputs)

Evaluate the brain with sensor inputs.

evaluate_with_activations(Pid, Inputs)

Evaluate and return both outputs and all layer activations.

get_network(Pid)

Get the current network.

get_topology(Pid)

Get network topology information.

get_viz(Pid)

Get current visualization data.

handle_call(Request, From, State)

handle_cast(Msg, State)

handle_info(Info, State)

init(Opts)

set_network(Pid, Network)

Replace the network (e.g., after mutation).

start_link(Opts)

Start a brain process.

start_link(Name, Opts)

Start a named brain process.

stop(Pid)

Stop the brain process.

subscribe(BrainPid)

Subscribe to visualization updates.

subscribe(BrainPid, SubscriberPid)

Subscribe a specific process to visualization updates.

terminate(Reason, State)

unsubscribe(BrainPid)

Unsubscribe from visualization updates.

unsubscribe(BrainPid, SubscriberPid)

Unsubscribe a specific process from visualization updates.

Functions

evaluate(Pid, Inputs)

-spec evaluate(pid(), [float()]) -> [float()].

Evaluate the brain with sensor inputs.

Returns the output values from the neural network. Also updates internal state and notifies subscribers.

evaluate_with_activations(Pid, Inputs)

-spec evaluate_with_activations(pid(), [float()]) -> {[float()], [[float()]]}.

Evaluate and return both outputs and all layer activations.

This is used by the learning system to record experiences. Returns {Outputs, Activations} where Activations includes all layers.

get_network(Pid)

-spec get_network(pid()) -> network_evaluator:network().

Get the current network.

get_topology(Pid)

-spec get_topology(pid()) -> map().

Get network topology information.

get_viz(Pid)

-spec get_viz(pid()) -> map() | undefined.

Get current visualization data.

Returns a map with nodes, connections, and activation levels.

handle_call(Request, From, State)

handle_cast(Msg, State)

handle_info(Info, State)

init(Opts)

set_network(Pid, Network)

-spec set_network(pid(), network_evaluator:network()) -> ok.

Replace the network (e.g., after mutation).

start_link(Opts)

-spec start_link(map()) -> {ok, pid()} | {error, term()}.

Start a brain process.

Options: - network - The neural network (required, from network_evaluator) - id - Optional identifier for this brain - input_labels - Labels for input neurons (for visualization) - viz_enabled - Enable visualization data (default: true)

start_link(Name, Opts)

-spec start_link(term(), map()) -> {ok, pid()} | {error, term()}.

Start a named brain process.

stop(Pid)

-spec stop(pid()) -> ok.

Stop the brain process.

terminate(Reason, State)

unsubscribe(BrainPid)

-spec unsubscribe(pid()) -> ok.

Unsubscribe from visualization updates.

unsubscribe(BrainPid, SubscriberPid)

-spec unsubscribe(pid(), pid()) -> ok.

Unsubscribe a specific process from visualization updates.