package data
Ordering
- Alphabetic
Visibility
- Public
- Protected
Type Members
- trait BatchStream[+I] extends AnyRef
- case class NonEmptyBatch[I](batch: I) extends StreamControl[I] with Product with Serializable
- case class Peek(label: String) extends Module with Product with Serializable
- case class SWALoopState(model: Seq[STen], optimizer: Seq[STen], epoch: Int, lastValidationLoss: Option[Double], minValidationLoss: Option[Double], numberOfAveragedModels: Int, averagedModels: Option[Seq[Tensor]], learningCurve: List[(Int, Double, Option[Double])]) extends State with Product with Serializable
- case class SimpleLoopState(model: Seq[STen], optimizer: Seq[STen], epoch: Int, lastValidationLoss: Option[Double], minValidationLoss: Option[Double], minValidationLossModel: Option[(Int, Seq[Tensor])], learningCurve: List[(Int, Double, Option[Double])]) extends State with Product with Serializable
- case class SimpleThenSWALoopState(simple: Option[SimpleLoopState], swa: Option[SWALoopState]) extends State with Product with Serializable
- sealed trait State extends AnyRef
- sealed trait StreamControl[+I] extends AnyRef
- trait TrainingCallback extends AnyRef
- trait ValidationCallback extends AnyRef
Value Members
- object BatchStream
- object BufferedImageHelper
- object DataParallel
- case object EmptyBatch extends StreamControl[Nothing] with Product with Serializable
- case object EndStream extends StreamControl[Nothing] with Product with Serializable
- object GraphBatchStream
- object IOLoops
Contains a training loops and helpers around it
Contains a training loops and helpers around it
The two training loops implemented here are:
- lamp.data.IOLoops.epochs
- lamp.data.IOLoops.withSWA implements Stochastic Weight Averaging
- object Reader
- object SWA
- object StateIO
- object StreamControl
- object Text
- object TrainingCallback
- object ValidationCallback
- object Writer
Binary serialization for Tensor with primitive Double, Float, Long
Binary serialization for Tensor with primitive Double, Float, Long
The layout of binary format is as follows: - The first 6 bytes are "LAMP" - The next unsigned byte is the major version - The next unsigned byte is the minor version - The next 4 bytes form a little endian integer as HEADER_LENGTH - The next HEADER_LENGTH bytes form an UTF-8 string as the header. - The header is a valid JSON object with the following fields:
- v: numeric positive integer is the version of the header structure
- shape: numeric array
- datatype : string, either "double", "long", "int", "float", "byte" - The header is padded with spaces (0x20) such that HEADER_LENGTH+12 is divisible by 16. The count of spaces are included in HEADER_LENGTH. - The next width * shape.reduce(_ * _) bytes form a little endian primitive array.