GoMNIST

README

GoMNIST

GoMNIST is a Go driver for reading Yann LeCun's MNIST dataset of handwritten digits

The MNIST dataset is included. It is a copy from Yann LeCun's MNIST page.

How to use

Install the packages:

% go get github.com/petar/GoMNIST

Import them in your code as

% import "github.com/petar/GoMNIST"

Load the MNIST training and testing data sets into memory using

train, test, err := Load("./data")
if err != nil {
	…
}

Sweep through the samples

sweeper := train.Sweep()
for {
	image, label, present := sweeper.Next()
	if !present {
		break
	}
	…
}

You can also have random access to the sets, using the methods of Set.

Author

My homepage is Petar Maymounkov. Follow me on Twitter at @maymounkov.

Documentation

Index

• Constants
• type Label
  • func ReadLabelFile(name string) (labels []Label, err error)
• type RawImage
  • func ReadImageFile(name string) (rows, cols int, imgs []RawImage, err error)
  • func (img RawImage) At(x, y int) color.Color
  • func (img RawImage) Bounds() image.Rectangle
  • func (img RawImage) ColorModel() color.Model
• type Set
  • func Load(dir string) (train, test *Set, err error)
  • func ReadSet(iname, lname string) (set *Set, err error)
  • func (s *Set) Count() int
  • func (s *Set) Get(i int) (RawImage, Label)
  • func (s *Set) Sweep() *Sweeper
• type Sweeper
  • func (sw *Sweeper) Next() (image RawImage, label Label, present bool)

Constants

const (
	Width  = 28
	Height = 28
)

Types

type Label

type Label uint8

Label is a digit label in 0 to 9

func ReadLabelFile

func ReadLabelFile(name string) (labels []Label, err error)

ReadLabelFile opens the named label file (training or test), parses it and returns all labels in order.

type RawImage

type RawImage []byte

Image holds the pixel intensities of an image. 255 is foreground (black), 0 is background (white).

func ReadImageFile

func ReadImageFile(name string) (rows, cols int, imgs []RawImage, err error)

ReadImageFile opens the named image file (training or test), parses it and returns all images in order.

func (RawImage) At

func (img RawImage) At(x, y int) color.Color

func (RawImage) Bounds

func (img RawImage) Bounds() image.Rectangle

func (RawImage) ColorModel

func (img RawImage) ColorModel() color.Model

type Set

type Set struct {
	NRow   int
	NCol   int
	Images []RawImage
	Labels []Label
}

Set represents a data set of image-label pairs held in memory

func Load

func Load(dir string) (train, test *Set, err error)

Load reads both the training and the testing MNIST data sets, given a local directory dir, containing the MNIST distribution files.

func ReadSet

func ReadSet(iname, lname string) (set *Set, err error)

ReadSet reads a set from the images file iname and the corresponding labels file lname

func (*Set) Count

func (s *Set) Count() int

Count returns the number of points available in the data set

func (*Set) Get

func (s *Set) Get(i int) (RawImage, Label)

Get returns the i-th image and its corresponding label

func (*Set) Sweep

func (s *Set) Sweep() *Sweeper

Sweep creates a new sweep iterator over the data set

type Sweeper

type Sweeper struct {
	// contains filtered or unexported fields
}

Sweeper is an iterator over the points in a data set

func (*Sweeper) Next

func (sw *Sweeper) Next() (image RawImage, label Label, present bool)

Next returns the next image and its label in the data set. If the end is reached, present is set to false.