I am working on documentation on Flmatrix (matrices containing floating point numbers). Any comments are welcome:

http://soegaard.github.io/tmp/flmatrix/flmatrix.html

The Github repo is here: https://github.com/soegaard/flmatrix

huh. it says your library is in… the base package??

that’s a new one

Good catch!

Just wrote @defmodule[flmatrix] .

do you also need an info.rkt file?

Sure - it hasn’t been submitted to the package server yet.

they’re still needed to list dependencies and scribble docs

(I was looking for it to see what deps your lib uses)

Not much. Search for require in flmatrix.rkt

I think an info file will fix the base problem too.

Yup. That’s so weird though. Why does scribble think a non-package module is in the base package?

¯_(ツ)_/¯

I’m surprised that these are not linked automatically

needs for-label imports, probably

(for-label racket/base flmatrix)

Well, I need to write the documentation for them first…

oh, yeah that would probably help

I started with the Quick Tutorial.

Yeah, but also the link for define. (for-label racket/base) is the thing needed here

Now I see. I’ll add a for-label.

Is there a conversion function from the current math library?

No. But that’s good idea.

Nit: nxn/mxn is hard to read IMO

Noted. I’ll need something better.

out of curiosity, is there something like this library but for fixnum matrices?

does math have that already?

Nit: grammar error in “Since BLAS and LAPACK exists in multiple versions, so a little care is needed.”

Typo: CLASS -> CBLAS

Hope the Kronecker product is supported :wink:

And speaking of it, I do hope that https://github.com/racket/math/pull/41 will be merged for the next release

@notjack Not that I know of. math/matrix can handle matrices over fixnums, but there isn’t anything specialized.

I added kron to flmatrix for you :slightly_smiling_face:

Oohh. I missed it

Thank you!

Does the division into “levels” make sense?

“median” should be “medium”

By ‘action’ you mean ‘sequence of stepwise actions’?

Also, branching factor is normally how many stepwise actions you have at each step. So if you always have, say, 4 possible actions at each step, the branching factor is 4, and you have 4^t sequences of actions of length t.

Then you have the effective branching factor which depends on how much you prune during search.

Filtering afterwards doesn’t sound very efficient ;)

@soegaard2 is probably the best person to review that PR by @sorawee

it’s not, but I think there’s some good generalized ways I can do the filtering without needing to know anything domain specific about the state space

this is an example of the direction I’m thinking in: https://docs.racket-lang.org/planning/The_Multiset_State_Representation.html?q=multiset-con#%28def._%28%28lib._planning%2Fmultiset%2Fcondition..rkt%29._multiset-condition-ignore-frequencies%29%29\|https://docs.racket-lang.org/planning/The_Multiset_State_Representation.html?q=multiset-con#%28def._%28%28lib._planning%2Fmultiset%2Fcondition..rkt%29._multiset-condition-ignore-frequencies%29%29

(I’m trying to make a passable general-purpose planner, not a great Sokoban planner)

(and users won’t be expected to provide their own heuristics)

@samth @sorawee I gave my thumbs-up a while ago. But I forgot to do it in the the Github thread (we talked over Slack).