Node contents

Struct to be the content of an terminal node of an expression tree. This struct represents a Float64 constant value.

Const(v::Float64)

Receives a Float64 value v which will be used as a constant.

The representation of the constant as string is obtained by rounding the value to 3 decimal places, and it is automatically obtained.

When using non-linear optimization, the nonlinear least squares optimization method looks for this struct specifically to optimize their values.

Struct to be the content of an terminal node of an expression tree. This struct keeps the range limits for creating a random constant with the ERC method.

ERC(lb::Float64, ub::Float64)

This struct is used to create constants in the terminal nodes. When it is selected to be a terminal, a new terminal will be created with the struct Const with a random value drawn between [lb, ub) to take the place of the ERC (Ephemeral Random Constant) at the terminal node. The string representation of the created constant is as described in Const documentation.

Struct to be the content of an internal node of an expression tree.

Func(f::Function, a::Int64)

Takes a function f and the arity a of the function. The function must always work in vectorized form (will always receive $a$ arrays with $n$ values, where each value of the array is an observation). So the function input will have a rows and n columns when this node is being evaluated in other methods.

The function's representation as a string is inferred from the name of the function passed, and when this node is used to create a node of a tree, it will have $a$ children.

Optionally, you can give a name to the function by using the 3 parameters constructor:

Func(f::Function, a::Int64, str_rep::String)

When creating new functions, protected operators must not be used. The non-linear optimization method uses autodiff to differentiate the tree, and complex functions can be problematic to automatically differentiate.

Struct to be the content of a terminal node of an expression tree. This struct represents a variable of the data set.

Var(v::String, i::Int64)

Receives a string v that will be used as the representation of the variable when printing the expression (you can use a placeholder if the database do not have column names) and a ì index that matches the column index of the corresponding variable in the observations.

Struct to be the content of a terminal node of an expression tree. This struct represents a weighted variable, that have a coefficient associated with it at the time of creation.

WeightedVar(v::String, i::Int64)

This struct represents a weighted variable, that can be adjusted with the non-linear least squares method.

The String representation is inferred as the same way when creating a $Var$ and the coefficient is inferred in the same way as a $Const$.

WeightedVar(v::String, i::Int64, w::Float64)

Additionally, you can force a specific coefficient by passing the value as the third argument on the constructor. When no value is specified, the coefficient is set to 1.0.

This weighted variable is a subtree with 3 nodes and depth 2, but in practice it is treated as a single node, as it is not of interest to make the dissociation between the weight and the variable during the GP process. By treating the weighted variable as a single node, it is not necessary to modify the crossover or mutation implementations to prevent changing the subtree.