#clml — Public Fediverse posts

I think #CLML and @shinmera get this wrong:
You can't (well, shouldn't) do this in #CommonLisp :
(defclass my-array () ((actual-array) (element-types) (dimensions)))
because you end up with a garbage type. Instead I think:
(deftype arrayish (&optional type size) `(and (array type size) (satisfies some-predicate)))

This way you get a meaningful parameterised type that relates to array #types

Edit: Scroll down to @shinmera's commentary

#lisp y #gopher show live nowish on #aNONradio @SDF https://anonradio.net/
Topics:
(1) Lispy ai
Mastodon discussion and lots of links to current CommonLisp #ai (thanks everyone) https://mastodon.sdf.org/@screwtape/109978604825517968
- What even is #AI
- #CLML and #MGL
- #NLP ... Thanks @galdor @svetlyak40wt
- #Interlisp Medley + ai projects from xerox ai machines docs (thanks @amoroso )
(2) Chaosnet
really just my chaosnet fiddling and some reading of the #moon http://bitsavers.trailing-edge.com/pdf/mit/ai/AIM-628_chaosnet.pdf

#gopher #laundromat ???

@mnl acronyms: #CLML is a now-forked big Japanese company's machine learning common lisp package. https://github.com/mmaul/clml
#MGL is a Franz lisp GPU-accelerated machine learning toolkit that won a Google competition for such. https://github.com/melisgl/mgl
#DL is a subset of #ML that uses at least something like a two layer hidden markov model with nonlinear activation functions.
Princomp = Principle component analysis is basically just an SVD. Er..
A GPU is hw good for bitmappable tasks.
AI is... er...

@james just remembered, I want to talk about putting the #ai back in #ai programming. I want to get away from writing what feels like web utilities for unix and back to ai, both me personally and the lisp ecosystem as such.

My recent step was a phost/peertube video about #CLML princomp. And I guess I'll evolve a modern AI system out of that. There's also Northwestern's case based reasoning. #MGL GPU DL?

@scm @svetlyak40wt @galdor @left_adjoint @amoroso @mhcat @ldbeth ideas? @mnl

@galdor for example machine learning #CLML MUST have at least --dynamic-space-size 2560
https://github.com/mmaul/clml #CommonLisp #sbcl

#gopher #code #ML #AI #CommonLisp #CLML #example #princomp
Principal component analysis is so easy in common lisp!
For the gopher:
gopher://tilde.institute/0/~screwtape/215530911-princomp.lisp.sig
For the gopher impaired:
https://gopher.floodgap.com/gopher/gw.lite?=tilde.institute+70+302f7e7363726577746170652f3231353533303931312d7072696e636f6d702e6c6973702e736967

gopher.club/1/users/screwtape generally
@pizzapal

#machineLearning #CLML #ai #CommonLisp
A #princomp (eigendecomposition of covariance matrix) + #ascii #gnuplot #biplot #example

Done on an oval, it successfully finds and identifies the relative importance of the major and minor ellipse axes.

@pizzapal corrections plz
https://forth.noip.me:8443/w/3YvEi4iSWWmyHaqY8BHR6n

#procrastinating
I need to phlog/peertube/toot
"
look what running princomp does [in #CommonLisp #CLML with a dumb terminal #gnuplot ]
"
but I have to do infrastructural things for it and then it implies doing followups like eigenface which will be lots of work and make me sound sophomoric

and I have a million urgent things to do

Does anyone know a #book on ML that uses #CLML ? CLML has done exactly the best approach to #machineLearning : Foreign BLAS constructed algorithms in #CommonLisp for CPUs. (For use inside #emacs #orgmode ).

Aside, does anyone know any other #japan based #lisp treasures like this?

I wrote a tiny sh that gets sbcl to do a demonstrative #pca principal component analysis with #CLML on #sbcl

I'll start a separate #ai show named lispy brain somewhere to go with my #gopher phosts on it. But for now could I put upon the usual suspects to sanity check my phost / give me some feedback on this particular phost:
gopher.club/0/users/screwtape/j.lispy-brain.txt
@ldbeth @mhcat @scm @pizzapal
Sorry if I missed you but if you can or can't make the sh in that phost go plz respond here!