Internal API

General

EvoTrees.TrainNode Type

TrainNode{S,V,M}

Carries training information for a given tree node

source

EvoTrees.EvoTree Type

EvoTree{L,K}

An EvoTree holds the structure of a fitted gradient-boosted tree.

Fields

• trees::Vector{Tree{L,K}}

• info::Dict

EvoTree acts as a functor to perform inference on input data:

pred = (m::EvoTree; ntree_limit=length(m.trees))(x)

source

EvoTrees.check_parameter Function

check_parameter(::Type{<:T}, value, min_value::Real, max_value::Real, label::Symbol) where {T<:Number}

Check model parameter if it's valid

source

EvoTrees.check_args Function

check_args(args::Dict{Symbol,Any})

Check model arguments if they are valid

source

check_args(model::EvoTypes)

Check model arguments if they are valid (eg, after mutation when tuning hyperparams) Note: does not check consistency of model type and loss selected

source

Training utils

EvoTrees.init Function

init(
    params::EvoTypes,
    dtrain,
    device::Type{<:Device}=CPU;
    target_name,
    feature_names=nothing,
    weight_name=nothing,
    offset_name=nothing
)

Initialise EvoTree

source

init(
    params::EvoTypes,
    x_train::AbstractMatrix,
    y_train::AbstractVecOrMat,
    device::Type{<:Device}=CPU;
    feature_names=nothing,
    w_train=nothing,
    offset_train=nothing
)

Initialise EvoTree

source

EvoTrees.grow_evotree! Function

grow_evotree!(evotree::EvoTree{L,K}, cache, params::EvoTypes) where {L,K}

Given a instantiate

source

EvoTrees.get_best_split Function

get_best_split(::Type{L}, h∇, node, n, js, params, feattypes, monotone_constraints)

Return (gain, feat, bin) for the best split of node, or (gamma, 0, 0) if none exceeds params.gamma.

source

EvoTrees.update_gains! Function

update_gains!(::Type{L}, h∇, node, n, js, params, feattypes, monotone_constraints)

Write eligible split gains into node.gains (oblivious trees).

source

EvoTrees.predict! Function

predict!(pred::Matrix, tree::Tree, X)

Generic fallback to add predictions of tree to existing pred matrix.

source

EvoTrees.subsample Function

subsample(left::AbstractVector, is::AbstractVector, mask_cond::AbstractVector{UInt8}, rowsample::AbstractFloat, rng)

Returns a view of selected rows ids.

source

EvoTrees.split_set_chunk! Function

split_set_chunk!(
    left,
    right,
    is,
    bid,
    nblocks,
    x_bin,
    feat,
    cond_bin,
    feattype,
    offset,
    chunk_size,
)

Multi-threaded split set. Take a view into left and right placeholders. Right ids are assigned at the end of the length of the current node set.

source

Histogram

EvoTrees.get_edges Function

get_edges(X::AbstractMatrix{T}; feature_names, nbins, rng=Random.TaskLocalRNG()) where {T}
get_edges(df; feature_names, nbins, rng=Random.TaskLocalRNG())

Get the histogram breaking points of the feature data.

source

EvoTrees.binarize Function

binarize(X::AbstractMatrix; feature_names, edges)
binarize(df; feature_names, edges)

Transform feature data into a UInt8 binarized matrix.

source

EvoTrees.update_hist! Function

update_hist!(nodes, build_nodes, ∇, x_bin_T, js, ::Val{NK})

Build nodes[n].h for each n in build_nodes. Threads over (node × feature tile); tiles write disjoint h[:, :, j] slices. NK == 2K+1 is hist width (Val so the inner loop unrolls).

source