private val model = Sequential.of(
    Input(28, 28, 1),
    Flatten(),
    Dense(300),
    Dense(100),
    Dense(10)
)
fun main() {
    val (train, test) = Dataset.createTrainAndTestDatasets(
        trainFeaturesPath = "datasets/mnist/train-images-idx3-ubyte.gz",
        trainLabelsPath = "datasets/mnist/train-labels-idx1-ubyte.gz",
        testFeaturesPath = "datasets/mnist/t10k-images-idx3-ubyte.gz",
        testLabelsPath = "datasets/mnist/t10k-labels-idx1-ubyte.gz",
        numClasses = 10,
        ::extractImages,
        ::extractLabels
    )
    val (newTrain, validation) = train.split(splitRatio = 0.95)
    model.use {
        it.compile(
            optimizer = Adam(),
            loss = Losses.SOFT_MAX_CROSS_ENTROPY_WITH_LOGITS,
            metric = Metrics.ACCURACY
        )
        it.summary()
        it.fit(
            dataset = newTrain,
            epochs = 10,
            batchSize = 100,
            verbose = false
        )
        val accuracy = it.evaluate(
            dataset = validation,
            batchSize = 100
        ).metrics[Metrics.ACCURACY]
        println("Accuracy: $accuracy")
        it.save(File("src/model/my_model"))
    }
}