Playing with Java Deep Learning (DJL), tutorial-02 & tutorial-03 combined to a standalone executable script / published by https://github.com/dacr/code-examples-manager #250223b8-c123-4c71-9f24-c57f207d371e/21bd6b7b6e927abcd9d65c706084c5b5e4fcf8b8

djl-tutorial-full.sc

// summary : Playing with Java Deep Learning (DJL), tutorial-02 & tutorial-03 combined to a standalone executable script
// keywords : djl, machine-learning, tutorial, ai, @testable
// publish : gist
// authors : David Crosson
// license : Apache NON-AI License Version 2.0 (https://raw.githubusercontent.com/non-ai-licenses/non-ai-licenses/main/NON-AI-APACHE2)
// id : 250223b8-c123-4c71-9f24-c57f207d371e
// created-on : 2021-03-05T09:23:01Z
// managed-by : https://github.com/dacr/code-examples-manager
// run-with : scala-cli $file

// ---------------------
//> using scala "3.4.2"
//> using dep "org.slf4j:slf4j-api:2.0.13"
//> using dep "org.slf4j:slf4j-simple:2.0.13"
//> using dep "ai.djl:api:0.28.0"
//> using dep "ai.djl:basicdataset:0.28.0"
//> using dep "ai.djl:model-zoo:0.28.0"
//> using dep "ai.djl.mxnet:mxnet-engine:0.28.0"
//> using dep "ai.djl.mxnet:mxnet-model-zoo:0.28.0"
////> using dep "net.java.dev.jna:jna:5.13.0"
// ---------------------


// inspired from https://docs.djl.ai/jupyter/tutorial/03_image_classification_with_your_model.html

System.setProperty("org.slf4j.simpleLogger.defaultLogLevel","debug")

import java.awt.image._
import java.nio.file._
import java.util._
import java.util.stream._
import ai.djl._
import ai.djl.basicdataset.cv.classification.Mnist
import ai.djl.ndarray.types._
import ai.djl.training._
import ai.djl.training.dataset._
import ai.djl.training.initializer._
import ai.djl.training.loss._
import ai.djl.training.listener._
import ai.djl.training.evaluator._
import ai.djl.training.optimizer._
import ai.djl.training.util._
import ai.djl.basicmodelzoo.cv.classification._
import ai.djl.basicmodelzoo.basic._
import ai.djl.ndarray._
import ai.djl.modality._
import ai.djl.modality.cv._
import ai.djl.modality.cv.util.NDImageUtils
import ai.djl.translate._

val modelPath = "build/mlp"
val modelDir = Paths.get(modelPath)

if (!modelDir.toFile.exists()) {
  println("----------------- Prepare MNIST dataset for training")
  val batchSize = 32
  val mnist = Mnist.builder.setSampling(batchSize, true).build
  mnist.prepare(new ProgressBar)

  println("----------------- Create your Model")

  val model = Model.newInstance("mlp")
  model.setBlock(new Mlp(28 * 28, 10, Array(128, 64)))

  println("----------------- Create a Trainer")

  val config = new DefaultTrainingConfig(Loss.softmaxCrossEntropyLoss())
    //softmaxCrossEntropyLoss is a standard loss for classification problems
    .addEvaluator(new Accuracy()) // Use accuracy so we humans can understand how accurate the model is
    .addTrainingListeners(TrainingListener.Defaults.logging() : _*)

  val trainer = model.newTrainer(config)

  println("----------------- Initialize Training")
  trainer.initialize(new Shape(1, 28 * 28))

  println("----------------- Train your model")
  val epoch = 5
  EasyTrain.fit(trainer, epoch, mnist, null)

  println("----------------- Save your model")

  Files.createDirectories(modelDir)

  model.setProperty("Epoch", String.valueOf(epoch))

  model.save(modelDir, "mlp")
}

// ========================================================================================


println("----------------- Load your model")
val model = Model.newInstance("mlp")
model.setBlock(new Mlp(28 * 28, 10, Array[Int](128, 64)))
model.load(modelDir)

println("----------------- Create a Translator")
val translator = new Translator[Image, Classifications] {
  override def processInput(ctx:TranslatorContext, input:Image):NDList = {
    // Convert Image to NDArray
    val array = input.toNDArray(ctx.getNDManager(), Image.Flag.GRAYSCALE)
    new NDList(NDImageUtils.toTensor(array))
  }

  override def processOutput(ctx:TranslatorContext, list:NDList):Classifications = {
    // Create a Classifications with the output probabilities
    val probabilities = list.singletonOrThrow().softmax(0)
    val classNames =
      IntStream
        .range(0, 10)
        .mapToObj(_.toString)
        .collect(Collectors.toList())
    new Classifications(classNames, probabilities);
  }

  override def getBatchifier():Batchifier = {
    // The Batchifier describes how to combine a batch together
    // Stacking, the most common batchifier, takes N [X1, X2, ...] arrays to a single [N, X1, X2, ...] array
    Batchifier.STACK;
  }
}

println("----------------- Create Predictor")
val predictor = model.newPredictor(translator)


println("----------------- Run inference")

0.to(9).foreach { num =>
  val url = s"https://mapland.fr/data/ai/images-numbers/$num.png"
  val img = ImageFactory.getInstance().fromUrl(url)
  //img.getWrappedImage()

  val classifications = predictor.predict(img)
  println(s"*** result for $num ($url)")
  println(classifications)
}