Struct TrainingSummary 

pub struct TrainingSummary { /* private fields */ }

Bounded-memory summary of a training process.

Tracks counts, recent/best errors, model switches, resets, and load failures. Does not store raw samples.

Examples

use rill_ml::diagnostics::TrainingSummary;

let mut summary = TrainingSummary::default();
summary.record_sample();
summary.record_error(0.5).unwrap();
summary.set_baseline_error(0.8).unwrap();

assert_eq!(summary.total_samples(), 1);
assert!(summary.beats_baseline().unwrap());

Implementations

impl TrainingSummary

pub fn new(config: TrainingSummaryConfig) -> Result<Self, RillError>

Create a new training summary with the given configuration.

pub fn record_sample(&mut self)

Record that a sample was processed.

pub fn record_rejection(&mut self)

Record that an input was rejected (invalid, non-finite, etc.).

pub fn record_error(&mut self, error: f64) -> Result<(), RillError>

Record an error from a prediction.

The absolute value is taken, so signed errors are accepted. Updates the recent error (EW mean) and the best (minimum) error.

pub fn set_baseline_error(&mut self, error: f64) -> Result<(), RillError>

Set the baseline error for comparison.

pub fn record_switch(&mut self)

Record that the active model was switched.

pub fn record_reset(&mut self)

Record that the model was reset.

pub fn record_load_failure(&mut self)

Record that a state load failed.

pub const fn total_samples(&self) -> u64

Total samples processed.

pub const fn rejected_samples(&self) -> u64

Samples rejected due to invalid input.

pub fn recent_error(&self) -> Option<f64>

Recent error (EW mean of absolute errors), or None if no errors recorded.

pub const fn best_error(&self) -> Option<f64>

Best (minimum) error observed, or None if no errors recorded.

pub const fn baseline_error(&self) -> Option<f64>

Baseline error for comparison, or None if not set.

pub const fn model_switches(&self) -> u64

Number of times the active model was switched.

pub const fn reset_count(&self) -> u64

Number of times the model was reset.

pub const fn load_failures(&self) -> u64

Number of state load failures.

pub fn beats_baseline(&self) -> Option<bool>

Whether the model is currently beating the baseline.

Returns None if either recent error or baseline error is unavailable.

pub fn reset(&mut self)

Reset all summary statistics.

Trait Implementations

impl Clone for TrainingSummary

fn clone(&self) -> TrainingSummary

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for TrainingSummary

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

Formats the value using the given formatter.

impl Default for TrainingSummary

fn default() -> Self

Returns the “default value” for a type.