The previous sections elaborated how you can use the Spark framework with its MLlib for ML problems. In most complex tasks, however, deep learning models provide better performance. Spark supports SparkDL, a higher-level API working over MLlib. It uses TensorFlow at its backend, and it also requires TensorFrames, Keras, and TFoS modules.
In this section, we will make use of SparkDL for classifying images. This will allow you to get acquainted with the Spark support for the images. For images, as we learned in Chapter 4, Deep Learning for IoT, Convolutional Neural Networks (CNNs) are the de facto choice. In Chapter 4, Deep Learning for IoT, we built CNNs from scratch, and also learned about some popular CNN ...