Struct TrainingConfig 

pub struct TrainingConfig {

    pub metric: Metric,
    pub epochs: usize,
    pub batch_size: usize,
    pub learning_rate: f64,
    pub beta1: f64,
    pub beta2: f64,
    pub penalty: f32,
    pub verbose: bool,
    pub patience: Option<i32>,
    pub loss_reduction: LossReduction,
    pub k_neighbors: usize,
    pub min_desired_loss: Option<f64>,
    pub timeout: Option<u64>,
    pub normalized: bool,
    pub minkowski_p: f64,
    pub repulsion_strength: f32,
    pub kernel_a: f32,
    pub kernel_b: f32,
    pub neg_sample_rate: usize,
    pub cooldown_ms: u64,
    pub figures_dir: Option<PathBuf>,
}

Configuration for training the UMAP model.

Deprecated: Use UmapConfig instead. This type is provided for backward compatibility and converts to/from UmapConfig.

Fields

metric: Metric

The distance metric to use for training the model.

epochs: usize

The total number of epochs to run during training.

batch_size: usize

The number of samples to process in each training batch.

learning_rate: f64

The learning rate for the optimizer.

beta1: f64

The Beta1 parameter for the Adam optimizer.

beta2: f64

The Beta2 parameter for the Adam optimizer.

penalty: f32

The L2 regularization (weight decay) penalty.

verbose: bool

Whether to show detailed progress information during training.

patience: Option<i32>

The number of epochs to wait for improvement before triggering early stopping.

loss_reduction: LossReduction

The method used to reduce the loss during training.

k_neighbors: usize

The number of nearest neighbors to consider.

min_desired_loss: Option<f64>

Minimum desired loss to achieve before stopping early.

timeout: Option<u64>

Maximum training time in seconds.

normalized: bool

Normalize distance output.

minkowski_p: f64

Minkowski p parameter.

repulsion_strength: f32

Weight applied to the repulsion term.

kernel_a: f32

UMAP kernel parameter a, fitted from min_dist and spread. Controls the width of the kernel: q = 1 / (1 + a * d^(2b)).

kernel_b: f32

UMAP kernel parameter b, fitted from min_dist and spread. Controls the decay shape: q = 1 / (1 + a * d^(2b)).

neg_sample_rate: usize

Number of negative samples per positive edge per epoch.

cooldown_ms: u64

Milliseconds to sleep at the end of every training epoch.

0 (the default) disables the pause and gives maximum throughput. Increase this value to reduce GPU utilisation at the cost of longer training time (e.g. cooldown_ms = 5 for ~10–20 % GPU headroom).

figures_dir: Option<PathBuf>

Directory where loss-curve and embedding snapshot plots are written.

Defaults to None, which resolves to "figures" in the current working directory. Set to a writable PathBuf when the process runs on a read-only filesystem.

Implementations

impl TrainingConfig

pub fn builder() -> TrainingConfigBuilder

Creates a new builder for constructing a TrainingConfig.

Trait Implementations

impl Clone for TrainingConfig

fn clone(&self) -> TrainingConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for TrainingConfig

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

Formats the value using the given formatter.

impl From<&TrainingConfig> for UmapConfig

fn from(config: &TrainingConfig) -> Self

Converts to this type from the input type.

impl From<&UmapConfig> for TrainingConfig

fn from(config: &UmapConfig) -> Self

Converts to this type from the input type.

impl From<TrainingConfig> for UmapConfig

fn from(config: TrainingConfig) -> Self

Converts to this type from the input type.

impl From<UmapConfig> for TrainingConfig

fn from(config: UmapConfig) -> Self

Converts to this type from the input type.