feature-training/org.pointyware.disco.training.entities

Package-level declarations

Types

BatchStatistics

interface BatchStatistics : Statistics

EpochStatistics

interface EpochStatistics : Statistics

Exercise

data class Exercise(val input: Tensor, val output: Tensor)

An individual training or test datum.

LayerStatistics

interface LayerStatistics : Statistics

Measurement

sealed class Measurement<T : Any>

Measurements can be taken from just about any part of a system.

OrganicTrainer

class OrganicTrainer(val pruningPeriod: Int) : Trainer

An organic trainer attempts to train a model of progressive complexity by growing layers in width and depth, as well as adding functional modules where linearities allow growth without disrupting learned features.

SampleStatistics

interface SampleStatistics : Statistics

SequentialStatistics

class SequentialStatistics(val measurements: List<Measurement<Float>>) : EpochStatistics, BatchStatistics, SampleStatistics

This Statistics type collects information for a SequentialNetwork.

SequentialTrainer

class SequentialTrainer(val network: SequentialNetwork, val cases: List<Exercise>, val lossFunction: LossFunction, val optimizer: Optimizer, val statistics: Statistics, acceptableError: Double = 0.001) : Trainer

A trainer for a SequentialNetwork using a list of Exercise instances to evaluate against the given lossFunction, updating the model parameters using the provided optimizer.

Snapshot

data class Snapshot(val epoch: Int, val measurements: Map<Measurement<Float>, Map<Float, Float>>, val timestamp: Instant = Clock.System.now())

Statistics

interface Statistics

A statistics object receives updates throughout training in order to collect information to assess training performance.

Trainer

interface Trainer

A Trainer defines a workflow for training a model.