Struct TrainConfig 

pub struct TrainConfig {

    pub noise_asset_root: String,
    pub training_sources: Vec<String>,
    pub validation_sources: Vec<String>,
    pub destination: String,
    pub log: String,
    pub simulation_size: usize,
    pub simulation_peak_radius: f32,
    pub simulation_harmonic_decay: f32,
    pub simulation_frequency_wobble: f32,
    pub captured_oversample_factor: usize,
    pub mha_heads: usize,
    pub dropout: f64,
    pub trunk_hidden_size: usize,
    pub model_epochs: usize,
    pub model_batch_size: usize,
    pub model_workers: usize,
    pub model_seed: u64,
    pub adam_learning_rate: f64,
    pub adam_weight_decay: f32,
    pub adam_beta1: f32,
    pub adam_beta2: f32,
    pub adam_epsilon: f32,
    pub no_plots: bool,
}

The training configuration used for all training, inference, and hyper parameter tuning.

Fields

noise_asset_root: String

The source directory for the noise assets used to generate simulated items.

training_sources: Vec<String>

The directories (or the special sim source) that provide training samples. Must contain at least one entry.

validation_sources: Vec<String>

Optional directories (or sim) that provide validation samples. An empty list triggers an 80/20 split of the training pool.

destination: String

The destination directory for the trained model.

log: String

The log directory for training.

simulation_size: usize

Simulation data set size.

simulation_peak_radius: f32

Simulation peak radius.

simulation_harmonic_decay: f32

Simulation harmonic decay.

simulation_frequency_wobble: f32

Simulation frequency wobble.

captured_oversample_factor: usize

The number of times to replicate captured samples when constructing datasets.

mha_heads: usize

The number of Multi Head Attention (MHA) heads.

dropout: f64

The dropout rate applied to attention and trunk layers.

trunk_hidden_size: usize

The hidden size of the model’s MLP trunk.

model_epochs: usize

The number of epochs to train for.

model_batch_size: usize

The number of samples to use per epoch.

model_workers: usize

The number of workers to use for training.

model_seed: u64

The seed used for training.

adam_learning_rate: f64

The Adam optimizer learning rate.

adam_weight_decay: f32

The Adam optimizer weight decay.

adam_beta1: f32

The Adam optimizer beta1.

adam_beta2: f32

The Adam optimizer beta2.

adam_epsilon: f32

The Adam optimizer epsilon.

no_plots: bool

Suppresses the training plots.

Implementations

impl TrainConfig

pub fn new( noise_asset_root: String, training_sources: Vec<String>, validation_sources: Vec<String>, destination: String, log: String, simulation_size: usize, simulation_peak_radius: f32, simulation_harmonic_decay: f32, simulation_frequency_wobble: f32, captured_oversample_factor: usize, mha_heads: usize, dropout: f64, trunk_hidden_size: usize, model_epochs: usize, model_batch_size: usize, model_workers: usize, model_seed: u64, adam_learning_rate: f64, adam_weight_decay: f32, adam_beta1: f32, adam_beta2: f32, adam_epsilon: f32, no_plots: bool, ) -> Self

Create a new instance of the config.

Trait Implementations

impl Clone for TrainConfig

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Config for TrainConfig

fn load_binary(data: &[u8]) -> Result<Self, ConfigError>

Loads the configuration from a binary buffer.

impl Debug for TrainConfig

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for TrainConfig

fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>

where D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Display for TrainConfig

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

Formats the value using the given formatter.

impl Serialize for TrainConfig

fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>

where S: Serializer,

Serialize this value into the given Serde serializer.