Neet.Network

Sigmoid

modSig :: Double -> Double

Modified sigmoid function from the original NEAT paper

Network

data Network

Sparse recurrent network, like those made by NEAT

Constructors

Network

Fields

netInputs :: [NodeId]

Which nodes are inputs

netOutputs :: [NodeId]

Which nodes are outputs

netState :: IntMap Neuron

netDepth :: Int

Upper bound on depth

Instances

Show Network

Neuron

data Neuron

A single neuron

Constructors

Neuron

Fields

activation :: Double

The current activation

connections :: IntMap Double

The inputs to this Neuron

yHeight :: Rational

Visualization height

neurType :: NodeType

Type, used in pushThrough

Instances

Show Neuron

Construction

mkPhenotype :: Genome -> Network

Updates

stepNeuron :: IntMap Double -> Neuron -> Neuron

Takes the previous step's activations and current inputs and gives a function to update a neuron.

stepNetwork :: Network -> [Double] -> Network

Steps a network one step. Takes the network and the current input, minus the bias.

snapshot :: Network -> [Double] -> Network

Steps a network for at least its depth

Output

pushThrough :: Network -> [Double] -> [Double]

Specialized snapshot for nonrecurrent nets. Will break on recurrent connections.

getOutput :: Network -> [Double]

Gets the output of the current state