The sum of the log-outputs of unit c of the network evaluated using a SAMPLE
with Output.Label = c divided by the number of samples, i.e. the log-likelihood.
FUNCTION
The function checks whether the net is an MLNET network (see NET-flags).
If so, it evaluates the network for each sample in dset, using eval_ff_net,
and finds the output unit which represents class sample->Output.Label (i.e.
with unit->InOutIndex = sample->Output.Label). The output value of that
unit is added to a sum.
Finally, the sum is normalized by dividing it by the total number of samples
in dset.