Hi Fabian,
On 12/20/2017 11:18 AM, Fabian Klopfer wrote:
1. Have there been a try in feeding the crossover/genetic operator
results in the inner MOSES loop into some deep net to figure out
relevant features, advanced genetic operators or some hints on how to
exactly cross over/ mutate the genome(s)?
No. The most advanced thing we've done with MOSES in terms of
meta-learning for feature selection is described in this paper
https://arxiv.org/abs/1703.06990
MOSES also allows to use MOSES itself dynamically perform feature selection.
2. Is there something like a minimal example to PLN (based reasoning) or
sth like a formal intro to PLN in a nutshell?
See
https://github.com/opencog/atomspace/tree/master/examples/rule-engine
then move onto
https://github.com/opencog/opencog/tree/master/examples/pln
(not all pln examples work, I think amusing-friend,
chicken-feet-or-pizza, crazy-happy and which-product don't)
There's also the tutorial
https://wiki.opencog.org/w/Hands_On_With_OpenCog#Pattern_Matching
https://wiki.opencog.org/w/Hands_On_With_OpenCog#Probabilistic_Logic_Networks_.28PLN.29
3. How are MOSES and the pattern miner connected/related? Does MOSES use
the pattern miner?
For now they aren't (well both are connected in spirit but not in code).
The idea is to have both rely on the URE
https://wiki.opencog.org/w/Unified_rule_engine
URE-based pattern mining is in the work, URE-based will code after.
The reason for moving as much intelligence on the URE is because it is
controllable by "control rules" that can ultimately be learned, see
https://github.com/opencog/opencog/tree/master/examples/pln/inference-control-learning
4. How are moses procedures linked to the atomspace/where are they saved?
You can either invoke MOSES with the right option to output scheme
programs that can be loaded into the atomspace. Or convert combo problem
into scheme, see
https://github.com/opencog/moses/blob/master/moses/comboreduct/main/combo-fmt-converter.cc
or moses --help for outputting directly in scheme format.
Note that output don't reflect all information are would be required to
reason about the models, such as the definitions of the training data,
fitness function and probably more.
Nil
Sorry if those questions have already been answered somewhere else and
thanks in advance
FK
--
You received this message because you are subscribed to the Google
Groups "opencog" group.
To unsubscribe from this group and stop receiving emails from it, send
an email to [email protected]
<mailto:[email protected]>.
To post to this group, send email to [email protected]
<mailto:[email protected]>.
Visit this group at https://groups.google.com/group/opencog.
To view this discussion on the web visit
https://groups.google.com/d/msgid/opencog/b321a93f-2d95-4f7c-a266-85a9b3576c5b%40googlegroups.com
<https://groups.google.com/d/msgid/opencog/b321a93f-2d95-4f7c-a266-85a9b3576c5b%40googlegroups.com?utm_medium=email&utm_source=footer>.
For more options, visit https://groups.google.com/d/optout.
--
You received this message because you are subscribed to the Google Groups
"opencog" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To post to this group, send email to [email protected].
Visit this group at https://groups.google.com/group/opencog.
To view this discussion on the web visit
https://groups.google.com/d/msgid/opencog/0a5e5a40-cc6f-1d0e-0421-ceec48a44b1e%40gmail.com.
For more options, visit https://groups.google.com/d/optout.
