Struct NeuralBlendNet 

pub struct NeuralBlendNet {
    pub w1: Vec<Vec<f64>>,
    pub b1: Vec<f64>,
    pub w2: Vec<Vec<f64>>,
    pub b2: Vec<f64>,
    pub output_names: Vec<String>,
}

A 2-layer MLP (input → hidden ReLU → output softmax) used to predict morph-target blend weights from anthropometric measurements.

Weights are stored in row-major order:

• w1: shape [HIDDEN_SIZE][INPUT_SIZE] — input→hidden
• b1: shape [HIDDEN_SIZE] — hidden biases
• w2: shape [N_OUTPUT][HIDDEN_SIZE] — hidden→output
• b2: shape [N_OUTPUT] — output biases

Fields

w1: Vec<Vec<f64>>

Rows = hidden units, cols = inputs. w1[h][i]

b1: Vec<f64>

Hidden-layer bias. b1[h]

w2: Vec<Vec<f64>>

Rows = outputs, cols = hidden units. w2[o][h]

b2: Vec<f64>

Output-layer bias. b2[o]

output_names: Vec<String>

Names of the output morph targets (same length as w2).

Implementations

impl NeuralBlendNet

pub fn new( w1: Vec<Vec<f64>>, b1: Vec<f64>, w2: Vec<Vec<f64>>, b2: Vec<f64>, output_names: Vec<String>, ) -> Self

Construct a network with explicit weight matrices.

Panics (only in debug mode)

Inconsistent dimensions trigger a panic — call from tests only.

pub fn default_body_predictor() -> Self

Build a default body-shape predictor with handcrafted weights that reflect anthropometric archetypes (not random values).

The hidden layer encodes four primitive body-feature detectors:

• Units 0-3: height patterns (tall / short / average / threshold)
• Units 4-7: weight/BMI patterns (light / heavy / moderate / obese)
• Units 8-11: age patterns (youth / middle / elder / crossover)
• Units 12-15: fitness/lean patterns (athletic / sedentary / mixed / peak)

The output layer maps these features to BODY_TARGET_NAMES softmax probabilities calibrated on anthropometric population data.

pub fn forward(&self, inputs: &[f64]) -> Vec<f64>

Run a forward pass through the network.

inputs must have exactly INPUT_SIZE elements; extra elements are ignored, missing elements default to 0.0.

Returns the softmax-normalized output vector (sums to 1.0).

pub fn predict_morph_weights( &self, height_cm: f64, weight_kg: f64, age: f64, fitness_0_1: f64, ) -> HashMap<String, f64>

Predict morph-target blend weights from anthropometric measurements.

Inputs are normalised internally:

• height_cm → (h - 170) / 15
• weight_kg → (w - 70) / 20
• age → (a - 35) / 20
• fitness → passed as-is (already [0, 1])

The returned map has exactly output_names.len() entries, with all values in (0, 1) and summing to 1.0.

Trait Implementations

impl Clone for NeuralBlendNet

fn clone(&self) -> NeuralBlendNet

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for NeuralBlendNet

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

Formats the value using the given formatter.