Struct DataView 

pub struct DataView<'a> {
    pub dataset: &'a Dataset,
    pub total_instances: usize,
    pub feature_columns: Vec<Vec<usize>>,
    pub possible_split_indices: Vec<Vec<usize>>,
    pub label_freq: Vec<usize>,
    pub sort_by_heuristic: bool,
    pub heuristic_values: HeuristicValues,
    pub bitset: Bitset,
}

A subset of the dataset: the instances that reach one node of the tree.

Each feature column is kept sorted by value, so candidate thresholds are the positions where the value changes. Splitting a view partitions every column in one linear pass, preserving the order.

Fields

dataset: &'a Dataset

The full dataset the view refers into.

total_instances: usize

Number of instances in the full dataset.

feature_columns: Vec<Vec<usize>>

For each feature, positions into dataset[f] of the instances in the view, in increasing value order.

possible_split_indices: Vec<Vec<usize>>

For each feature, positions in feature_columns[f] where the value changes: the candidate thresholds.

label_freq: Vec<usize>

Number of instances of each class in the view.

sort_by_heuristic: bool

Whether features and splits are ordered by Gini.

heuristic_values: HeuristicValues

Gini ordering of features and splits, when sort_by_heuristic is set.

bitset: Bitset

The instances in the view; the cache key of the subproblem.

Implementations

impl<'a> DataView<'a>

pub fn root(dataset: &'a Dataset, sort_by_heuristic: bool) -> Self

The view of the whole dataset. The dataset’s features must be sorted.

pub fn get_dataset_size(&self) -> usize

Number of instances in the view.

pub fn get_feature_number(&self) -> usize

Number of features.

pub fn get_sorted_feature(&self, f: usize) -> &Feature

The full, sorted column of feature f.

pub fn get_feature_indices(&self, f: usize) -> &[usize]

Positions into Self::get_sorted_feature of the instances in the view, in value order.

pub fn get_labels_freqs(&self) -> &[usize]

Number of instances of each class in the view.

pub fn get_num_labels(&self) -> usize

Number of classes in the dataset.

pub fn get_possible_split_indices(&self, f: usize) -> &[usize]

Candidate thresholds of feature f, as positions in the sorted column.

pub fn get_max_splits(&self) -> usize

Largest number of candidate thresholds of any feature.

pub fn ordered_possible_splits(&self, feature: usize) -> &[usize]

Indices into the candidate thresholds of feature, best Gini first. Empty unless the heuristic ordering is enabled.

pub fn features_best_score(&self) -> &[(f64, usize)]

(best Gini, feature) pairs, best first when the heuristic ordering is enabled and in feature order otherwise.

pub fn initialize_split_parameters( &self, feature_index: usize, split_point: usize, left_freq: &mut [usize], right_freq: &mut [usize], )

Fills the class histograms of both sides of a split of feature_index at split_point. Both slices must start zeroed.

pub fn split(&self, sf: usize, split_point: usize) -> (Self, Self)

Splits the view on feature sf at position split_point of its sorted column. The left view holds the instances before the position.

pub fn len(&self) -> usize

Number of instances in the view.

pub fn is_empty(&self) -> bool

Whether the view has no instances.