The second important data structure adopted from SPRLIB is the way that datasets are stored. A dataset consists of a dataset header with a linked list of all samples attached, see (DATASET). The header contains general information about the dataset such as the creation date, a description string, dimensions of input and output vectors (the targets in neural network context), number of samples and space to hold additional data that the user can use for his own purpose. See figure 2.1 for a graphical representation of the data structure, and (DATASET) for the structure as it is implemented in C.
Attached to the header is a linked list of data points, called samples in SPRANNLIB, that holds the actual data, pairs of input and corresponding targets, together with control information. As in the data set header, space is reserved for the user to store his own information, if needed (figure 2.1 and SAMPLE).