classifier

Documentation

Overview

Package classifier is a MNIST-based digit classifier. It loads a pre-trained model and offers a Classify method that will classify any image, by first resizing it to the model's input size.

To use it, create a Classifier with New(), and then simply call its Classify method with a 32x32 image.

This is an example of how to serve a model for inference.

Index

• type Classifier
  • func New(checkpointDir string) (*Classifier, error)
  • func (c *Classifier) Classify(img image.Image) (int32, error)

Types

type Classifier

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

Classifier holds the MNIST model compiled. It will use XLA with GPU if available or CPU by default. But the backend can be configured with GOMLX_BACKEND.

func New

func New(checkpointDir string) (*Classifier, error)

New creates a new Classifier object that can be used to classify images using a MNIST model.

func (*Classifier) Classify

func (c *Classifier) Classify(img image.Image) (int32, error)

Classify takes a 28x28 image and returns a MNIST classification according to the models. The returned class is from 0 to 9.

TODO: resize image to 32x32, used by the model.