opencog-dev team mailing list archive

["Joel Pitt" <joel.pitt@xxxxxxxxx>]

On Tue, Apr 29, 2008 at 2:31 PM, Cassio Pennachin <cassio@xxxxxxxxxxxxx> wrote:
>  I've suggested just calling it Attention or something like that.  We
> definitely don't want all different cognitive dynamics in the same
> directory.  If a root for all cognitive dynamics is needed, CogDynamics is
> better than CogAgents -- a dynamic is usually a collection of agents and
> related code, and an Agent by itself is often meaningless.

Yup, I'll do this soon - using CogDynamics or something similar.
However so far all the dynamics I've been working on are purely
MindAgent based with minor additions to the AtomSpace.

> > p.s. FYI, on a related note, I've got SQL save and restore working.
> > I've got this
> > vague daydream of someday being able to fetch atoms, "on demand", from
> > SQL, and, when RAM is tight, saving the least important ones to SQL.  But
> > for that, I need to have some way of telling apart atoms that are in hot
> demand
> > from the stale ones.  .. and I'm thinking that your attention ranking
> would be
> > just the ticket for that.

Cool about the SQL storage!

Yup, basically each atom should have an AttentionValue object, and
with that is an STI  (Short Term Importance) value. This should be
accessed through the AtomSpace getSTI() method (and changed using the
setSTI, as STI needs to be conserved).

There is also the Long Term Importance and the Very Long Term
Importance values which are more for guiding the forgeting process.

Basically lower STI atoms are less important and can potentially be
swapped out. Although as Cassio mentioned, to do this effectively one
needs to identify unimportant clusters instead of just individual
unimportant atoms.

J