Hi!
In order to perform connections weight normalization, I am using the GetConnections()
function to obtain the input connections for every neuron. That is, I use a code like
for i in range(number_of neurons)
conn[i] = GetConnections(my_source, my_target[i], ...)
Later in the code, I use conn[i].weight to perform my normalization.
It implies that for big networks (number_of_neurons=10k) I have to call GetConnections()
around 10k times. Hence, the execution time is huge.
Could it be possible that you enhance GetConnections() (or define a new function) in order
to return the connections for all the neurons in the target in array with the same
longitude as the number of neurons in target instead of the number of connections? That
is, indexed by neurons instead of indexed by connections.
I understand that present implementation of GetConnections() visits all the connections
and return only the input connections to target. I do not want to tell developers how to
do it :-) but I believe it could be implemented by visiting all the connections (related
to target) just ONCE and copy each one to the corresponding output array indexed by
neuron. This way, it would be a single call to GetConnections() (plus the time needed to
internally manage the assignment to the output array), instead of 10k times.
It would be very much appreciated.
Thanks a lot in advance!
Xavier