cu

Documentation

Overview

Package Cu implements the learning stage of the Neurlang classifier on CUDA

Index

• Constants
• type HyperParameters
  • func (h *HyperParameters) H() *learning.HyperParameters
  • func (h *HyperParameters) SetLogger(filename string)
  • func (h *HyperParameters) Solve(d datasets.SplittedDataset) (int, *hashtron.Hashtron)
  • func (h *HyperParameters) Training(d datasets.Splitter) (*hashtron.Hashtron, error)

Constants

const INPUTS = 16

const RESULTS = 50

Types

type HyperParameters

type HyperParameters struct {
	learning.HyperParameters

	CuCutoff        uint32 // the switchover point to cuda. Smaller problems go to cuda
	CuMemoryBytes   uint64 // statically set memory
	CuMemoryPortion uint16 // how many percent of gpu memory to use. 2=half, 3=third
	CuErase         bool   // pre-erase memory set
	CuArenaBytes    uint64 // set memory for solution arena
	// contains filtered or unexported fields
}

func (*HyperParameters) H

func (h *HyperParameters) H() *learning.HyperParameters

func (*HyperParameters) SetLogger

func (h *HyperParameters) SetLogger(filename string)

SetLogger sets the output logger file where hashtron golang code programs are written

func (*HyperParameters) Solve

func (h *HyperParameters) Solve(d datasets.SplittedDataset) (int, *hashtron.Hashtron)

Solve directly solves a single hashtron on a splitted dataset d. It outputs the size of solution and the trained hashtron if successful. Most callers should use Training instead.

func (*HyperParameters) Training

func (h *HyperParameters) Training(d datasets.Splitter) (*hashtron.Hashtron, error)

Training trains a single hashtron on a dataset d. It outputs the trained hashtron if successful, or an error.