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"))
   }
}