Struct RowVariants 

pub struct RowVariants<'a> { /* private fields */ }

Rows that differ from one base row in a few features: variant v is base with base[features[j]] replaced by values[v * features.len() + j], for every j. BoostedModel::predict_margin_variants_into and BoostedModel::predict_variants_into predict all of them bit for bit as BoostedModel::predict_margin_row_into and BoostedModel::predict_row_into predict each materialized row, but walk every tree once per 64 variants instead of once per variant: a search scoring hundreds of candidate values of a few inputs against one fixed context. NaN is a missing value, in base and values alike.

use hessboost::model::RowVariants;
use hessboost::prelude::*;

let x: Vec<f32> = (0..120).map(|i| (i % 40) as f32).collect();
let y: Vec<f32> = x.chunks(3).map(|row| row[0] - row[2]).collect();
let dtrain = DMatrix::from_dense(&x, 40, 3)?.with_labels(&y)?;
let model = train(&TrainingParams::default(), &dtrain, 5)?;

// Three candidate values of feature 2 against one fixed row.
let base = [4.0, f32::NAN, 1.0];
let candidates = [0.0, 10.0, 30.0];
let variants = RowVariants::new(&base, &[2], &candidates)?;
assert_eq!(variants.len(), 3);
let mut margins = [0.0; 3];
model.predict_margin_variants_into(&variants, Iterations::Best, &mut margins)?;
for (&value, &margin) in candidates.iter().zip(&margins) {
    let mut row = base;
    row[2] = value;
    assert_eq!(margin, model.predict_margin_row(&row, Iterations::Best)?);
}

Implementations

impl<'a> RowVariants<'a>

pub fn new( base: &'a [f32], features: &'a [usize], values: &'a [f32], ) -> Result<Self>

The variants of base in which the features features (distinct indices) take the values values: one row of features.len() values per variant, row-major. The rows are read in place, never copied; the model checks base and the indices when it predicts.

Errors

HessboostError::InvalidParameter (variants) when features is empty or lists a feature twice, HessboostError::DimensionMismatch when values is not whole rows of features.len() values.

pub fn len(&self) -> usize

The number of variants.

pub fn is_empty(&self) -> bool

Whether there are no variants.

Trait Implementations

impl<'a> Clone for RowVariants<'a>

fn clone(&self) -> Self

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl<'a> Copy for RowVariants<'a>

impl<'a> Debug for RowVariants<'a>

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.