Machine learning algorithms often need large datasets for training. E.g. the MNIST dataset consists of 100.000 records. Each record represents a little greyscale image of size 28x28 pixels of a handwritten number and the respective number as integer value , i.e. 28x28 = 784 integer values in the range interval [0,255] and one integer value for the number value.
Assume you want to train a neural network in PAL using AMDP (ABAP managed database procedures) in order to predict the number. For this task you could e.g. use the PAL algorithm "Multilayer Perceptron", which is a neural network. In order to supply the input neurons with data a dataframe is needed, i.e. an ABAP-DDIC structure containing the following fields: