next up previous
Next: Learning algorithms Up: Supporting network functions Previous: Low level supporting functions

High level supporting functions

The most important group of high level functions is to read and write a network to disk file. A network is written to a disk file using an ASCII format, creating a machine independent representation. The network structural organization together with all the weights are stored in one file. If network files are very large or not loaded very often, the library enables adaptive Lempel-Ziv compression of these files on Unix machines. A general network loading function makes this transparent for the user. The loading time will be longer due to the run time decompression needed. Functions used for writing and reading networks are fprintf_compressed_network and fprintf_network, fscanf_compressed_network and fscanf_network and load_network. More detailed information about these functions can be found in the reference part.

A function to check the network consistency (check_all_network), to determine if a loaded network is valid, runs various tests checking the topology of the network. Some heuristic rules are applied to warn about parameters with unrealistic values. Many network paradigms require random initialization or random perturbation of the weights. Two functions are available for that purpose, unif_rand_net and unif_jogg_net. As explained in previously a history mechanism allows the user to record the evolution of different parameters. A number of functions are available to control this. This option should be used with care because it can lead to very large networks, especially if stored to disk.

The last set of functions is to monitor all network parameters. These are all ASCII based formatted printings of all structures found in the network memory structure, for example status_net and status_link.



Created by LaTeX2HTML on Thu Mar 5 16:02:50 MET 1998