programming

[real sierra]

[sage crag]

shimimimi

[clear sedge]

shimi shimi ya shimi ya shimi yah

[obsidian mantle]

[sage crag]

[tender river]

[olive sable]

drank swalalala

[obsidian mantle]

Where do they get probabilities from

[sage crag]

depends on the encoder and decoder

look into RANS and TANS for different ways of getting the probabilities

[obsidian mantle]

Wait how could Microsoft patent something that already exists

rANS

Or is that forum I stumbled on lying

[sage crag]

no they did patent it

its expired though

you can use it dont worry

[clear sedge]

just don't look up patents

they can't hurt you if you don't know they exist

[tender river]

[clear sedge]

unless it's nintendo
then they can and will use military forces against you

[jagged turtle]

bold to assume it will only be militarily

[maiden geyser]

just live somewhere where they don't operate

[real sierra]

bathroom

lock door

this is why you book seats ahead of time

[faint sandal]

no bathroom visit allowed during turbulence

[midnight sigil]

[silent cloak]

https://tenor.com/view/spider-man-norman-osborn-surprised-you-cant-do-this-to-me-gif-24182893

[rigid snow]

during takeoff and landing too i heard one plane had to goaround because someone ran to the restroom during landing

recently

[real sierra]

[clear sedge]

now i love linux
but i love doggos more

i'm sitting at #2

[faint sandal]

they're hungry

they're wolves

[clear sedge]

they're just big doggos

they can just eat the linux guy and i can wait until they finish first

then i can sit with them

problem solved

[opaque sigil]

[real sierra]

question

[opaque sigil]

Answer

[real sierra]

//! For Windows(MSVC) compatibility
#if defined(_MSC_VER)
    #define DLLEXPORT __declspec(dllexport)
#else
    #define DLLEXPORT
#endif

this is in chessapi.h for MSVC compatibility

[opaque sigil]

Yes that's what you want

[real sierra]

there's an issue to add this to bitboard.h, do i have to worry about it already being defined

e.g.

#ifndef DLLEXPORT
    //! For Windows(MSVC) compatibility
  #if defined(_MSC_VER)
      #define DLLEXPORT __declspec(dllexport)
  #else
      #define DLLEXPORT
  #endif
#endif

[opaque sigil]

Uhhhh

[nocturne olive]

So what happened, was the Chess API exploded?

[real sierra]

didnt have time to make changes yesterday

im taking a look at open PRs and issues today first

[sage crag]

do this unless you have a reason to override a user-set dllexport

[real sierra]

i'll change chessapi.h to use this as well

oh also

i guess bitboard.h should have

#ifdef __cplusplus
extern "C" {
#endif
...
#ifdef __cplusplus
}
#endif

[sage crag]

ye

[real sierra]

also adding #pragma once to it, shouldnt be needed but good practice

that should be all the easy compat changes for the header files

[midnight sigil]

ligma

[clear sedge]

#pragma once is non-standard

[real sierra]

already in use in chessapi.h with no issue

[opaque sigil]

it's supported by all compilers that matter

[real sierra]

wontfix

[nocturne olive]

It's interesting how batch size effects training speed
Up to a point higher batch size will be at least the same speed, but eventually it starts to slow down, but also not linearly
Quite interesting

[clear sedge]

not using header guards

disappointed

[tender river]

shiro how are you a java dev

you are given a more verbose option and you just ignore it

[olive sable]

[rigid timber]

this isn't enterprise software tbh, no point in making it too verbose, regular unreadable is enough

[real sierra]

all this token saving over the last couple weeks has wiped my java skillset

[rigid timber]

long names are still one token

[real sierra]

well yeah, i still have that part of my java skills

have you seen my macro names

[rigid timber]

no

[real sierra]

#define best_score_eq_best_score_gt_score_best_score_or_score_sc best_score = best_score_gt_score best_score : score;

[clear sedge]

oh god

[rigid timber]

[sage crag]

java skill inversely proportional to code length

[rigid timber]

that's a Java dev alright

[olive sable]

drawing tablets work on linux right?

[real sierra]

i like it because my code looks the same after i add the macro

[sage crag]

at least its not
bseqbsgtsbsssc

[rigid timber]

that would be your average C dev

[real sierra]

true

[sage crag]

x86 instruction

[real sierra]

#define cbclose_return_rv_sc_cbclose_void } return_rv_sc } void

[rough bloom]

mostly
need drivers of course but anything common should be supported

[clear sedge]

most mainstream distros provide drivers for drawing tablets
fedora comes with a few preinstalled :3

[sage crag]

cbrsccbcv

[real sierra]

cat stepped on your keyboard

[rigid timber]

or it's a widely used C stdlib function for string manipulation

[opaque sigil]

There's probably an instruction with that name

[clear sedge]

average python dev naming scheme

[real sierra]

@shadow sinew apologies for the ping, i wanted to see if you would be free to get the PR resolved before a planned breaking api change? sorry for all the running in circles, that's on me

[tender river]

python has pretty average names

other than all the dunders

[clear sedge]

not python itself
python devs

for some reason they love shortening variable names to hell and back

literal opposite of java devs

[tender river]

variable names are fine to shorten tho

they're local

just name them whatever you want

[opaque sigil]

Half the "python devs" are just mathematicians

[real sierra]

im a mathematician and i like C and Java

python is ok tho

[rigid timber]

don't you know? writing numpy all the time gets tiring, gotta save 3 characters

[nocturne olive]

Meanwhile me: Kotlin

[clear sedge]

if you're using non-self-documenting identifiers, you're either a psychopath or a code golfer

[tender river]

its true though

[nocturne olive]

And I'm a vocal synthesist

[rigid timber]

autocomplete

[real sierra]

plot twist: everyone here is both

[tender river]

np. faster than num<tab>.

[rigid timber]

shruge

[obsidian mantle]

Sometimes i make a 5 word sentence variable and i feel like its too long. Sometimes i leave it unchanged

[nocturne olive]

I like long variable names usually

[tender river]

what if i'm both

[rigid timber]

by what, a couple hundred milliseconds maybe, not like typing speed usually limits the speed at which you can code anyway

[tender river]

also identifiers are selfdocumenting by their assignment

[clear sedge]

gotta save yourself two keystrokes to make someone else / your future self check all your imports and variables for what the hell np means

[opaque sigil]

It adds up

[obsidian mantle]

Is there a limit to variable name length

[nocturne olive]

I think my Chess bot has a major lack of long and dumb variable names

[opaque sigil]

There's always a limit

[rigid timber]

pretty sure there isn't unless you can store data in the variable name, then it's 64 bits

[tender river]

maybe if your future self has dementia

i'm against wildcard imports

i'm not against arbitrary renames

[olive sable]

the drawing tablet has mouse acceleration

[rigid timber]

what

[tender river]

as long as you can check what some name means, it doesnt matter if its short or long as long as its vaguely relevant

[tender river]

tablet smoothing?

you can turn it off

somehow

[olive sable]

if i move faster it goes further

[nocturne olive]

Average Chess bot import table import chess.api.*

[tender river]

disable relative mode maybe

[tender river]

fine in code golfing

[nocturne olive]

Silly

[clear sedge]

python is especially painful in that regard. you can't tell where something is declared at times unless you spend minutes reading your code

[rigid timber]

time to do the same with JS

[olive sable]

it seems to only do that for toush inputs

the pen is pretty acurate to the width of my monitors

[tender river]

just remembered an evil hack i made

[real sierra]

ok

time to fix my chess bot

tokens can come later

[olive sable]

that does mean it moves 3x as fast horizontally cuz i have 3 monitors

[clear sedge]

i have a golfed beers on the wall in js somewhere in my dms

let me find it

[tender river]

https://github.com/chayleaf/notnft/blob/f827912c850cb66cf489a69078ff55658f59b96f/dsl.nix#L13

[olive sable]

oh it has bluetooth apparently

ill probably never use that

[clear sedge]

$=console.log,i=99,k=1,a=_=>`${i||"nN"[k^=1]+"o more"} bottle${i-1?"s":""} of beer`,c=" on the wall",d=".",e=", ",f="\n";for(;i;)$(a()+c+e+a()+d+f+"Take one down and pass it around"+e+a(--i)+c+d+f);$(a()+c+e+a()+d+f+"Go to the store and buy some more"+e+a(i=99)+c+d)```
look at this beauty from the year of our lord 2023

[rigid timber]

hell yeah

[tender river]

shiro why does <x> not support ' in identifiers bad language

[nocturne olive]

On mine I'm waiting for a fix to a probably API-based bug

[midnight sigil]

I thought this was #gaming for a sec

[rigid timber]

[tender river]

theres a reason people dont use superLongVariableNames in math

keeping them short makes your code concise

when you dont see the code, thats when you gotta have descriptive names

interface boundaries

and the more generic your code, the less descriptive names you even can give to variables

what are you gonna name it, "argument1"? yeah just name it "x"

[real sierra]

i single-handedly carrying 50% of for loops

[tender river]

[nocturne olive]

Meanwhile my Chess bot has 0 actual for loops

forEach function more efficient

[midnight sigil]

:neuroTrue: place the definition in comments

[tender river]

[nocturne olive]

And then there's fun countOwnPieces(board: Board) = PieceType.entries.sumOf { ChessApi.getBitboard(board, color, it).value.countOneBits() }

[midnight sigil]

[clear sedge]

you're kind of strawmanning, i never said always use long names
my argument is usually it's best to use descriptive names, x, y, z may very well be descriptive for something like a vector

[real sierra]

descriptive =/= long

[tender river]

i do v' = push element v not vWithElement = push element v

[real sierra]

sometimes i use idx where i might be unclear w.r.t. other variables in use

[tender river]

i use idx outside of the context of loops

[nocturne olive]

sumOf { } is just more Kotlin STD goodness, it sums up and returns the return values of running the given integer returning block on each element of the given iterable

[clear sedge]

call me a psycho, but i use index

[tender river]

v' is not as descriptive as vWithElement, but its more readable

v prime

same goes for intermediate variables in data processing, i'd rather give them no names at all and use point free style but sometimes i have to give them names in which case i'll just do x or y

[nocturne olive]

Basically this is very small

var sum = 0
for (type in PieceType.entries) {
  sum += ChessApi.getBitboard(board, color, it).value.countOneBits()
}
return sum

[tender river]

some_func_that_returns_result().map(|x| x.do_something()), not some_func_that_returns_result().map(|some_func_result| some_func_result.do_something())

[clear sedge]

PieceType.entries.forEach { ... } might save you some tokens

[nocturne olive]

Already got fun countOwnPieces(board: Board) = PieceType.entries.sumOf { ChessApi.getBitboard(board, color, it).value.countOneBits() }

[clear sedge]

o

yes yes

my mistake

[nocturne olive]

Kotlin STD peak

[tender river]

and most variables are equally as useless imo only exist because of language skill issues not letting you chain functions in a nice way

(most not all)

[clear sedge]

if you can reasonably say that you can infer what a variable name means, that is descriptive enough

[tender river]

i cant infer what x means from x alone

[clear sedge]

the context surrounding it is also important

[tender river]

but i can look 4 tokens to the left and see what it means

thats my point

no need for long names when the assignment is right there

so you can literally just see what it is

[clear sedge]

then it's descriptive enough

i'm arguing for descriptive

not for long

[tender river]

x not descriptive

[clear sedge]

it's descriptive given the context

it's literally f(x) = ...

[nocturne olive]

Meanwhile I must do var scoreCpy = score because score is not mutable

And later legalMoves[move] = scoreCpy

[clear sedge]

this is some java single-index array level bs

[tender river]

oh i remembered go

[real sierra]

HOW DOES ANY OF THIS WORK ANYMORE

[nocturne olive]

Well it's iterating over the elements of a MutableMap<Move, Int> to score them

[real sierra]

MY TREE IS A CIRCLE

[rigid timber]

pretty sure trees are supposed to be planted in dirt, not be a circle...

[real sierra]

this is

so confusing

[tender river]

https://github.com/golang/go/issues/60078 closed

[clear sedge]

how is your tree a circle

what did you do

[nocturne olive]

Wuh

[tender river]

shiro circular trees are the best

if you wanted non-circular trees you could just rust

boring

[clear sedge]

rust (derogatory)

[rigid timber]

rust (oxide)

[real sierra]

ok ok so

[clear sedge]

my professional opinion? rust is fun but c is funner for me

[tender river]

rust (plant disease)

[real sierra]

consider the following game state tree, where you can make a move (indicated by letter) and the state is represented by the unordered set of moves preceeding it

[rigid timber]

[real sierra]

the problem with this tree is that the state AB occurs twice, but we regard them as separate entities

this means the work we did analyzing the first occurrence is repeated

which is not ideal

[clear sedge]

[real sierra]

to resolve this, we might consider merging the states:

but then a new problem arises, which is that this is no longer a tree, it is a graph

nodes no longer have a concept of "parent", as they can have multiple parents

[tender river]

circles in circles i go around

[clear sedge]

i'm still waiting for the now-diamond with two wings to turn into a circle

[tender river]

looking for something that cant be found

[olive sable]

the problem then becomes you have to start working with indices if you weant to reuse stuff

[real sierra]

this causes a number of problems, such as with backpropogation; if we were to naively propogate results up to each parent node from AB, they would be double-counted at the root

so now i have to rewrite my backpropogation functions to keep a list of visited nodes

[rigid timber]

just add infinite more nodes

[real sierra]

so i have an O(n^2) backpropogation now

[tender river]

use hash that includes parent hash

[warped frost]

I accidentally unmuted this server

Lemme just

[real sierra]

i plan to

[tender river]

because two boards with the same position that come after one another are NOT equivalent

[clear sedge]

duh doi just download more ram

[tender river]

they are different

[clear sedge]

[warped frost]

Bye chat

[tender river]

because the later the position the closer you are to draw

[clear sedge]

bye vixen

[real sierra]

this is a fair criticism but probably worth addressing later

[olive sable]

My deskspace has become too small

[real sierra]

depth should be an easy thing to integrate to the hash

[tender river]

thats medium

[real sierra]

or halfmove counter, for your particular example

[olive sable]

that is medium yes

supposedly

[trim valve]

monitor arm time

[olive sable]

dont those cost a lot?

[real sierra]

i mean actually

is it bad to double count a node

[olive sable]

especialy for triple

[real sierra]

maybe thats not even an issue

if the node occurs twice and is a bad result, it should be counted doubly bad, once for each branch in which it occurs

[trim valve]

I mean

it depends

nice ones are expensive

but also you wouldn't need a triple one, just get multiple arms

[olive sable]

ah

lets see

[real sierra]

question

on a one-to-one basis, is realloc or malloc more expensive

i have a case where i can either dynamically expand an array or add items to a linked list

[tender river]

bench it

[clear sedge]

the only way to tell is to bench it

[real sierra]

too much work gonna find someone on stackoverflow who did it already

[rigid timber]

pretty sure it also depends on the code around it so their result may not be exactly relevant to you

[real sierra]

On a modern OS the realloc will win by a huge margin because it can remap a buffer to a new address

[olive sable]

monitor arms are made to go against gravity so i cant mount one upside down right?

[real sierra]

different use case but relevant principle

[opaque sigil]

i mean, malloc can reuse buffers too

[clear sedge]

won't the malloc equivalent be at least roughly two syscalls though?

one for free, one for malloc

[real sierra]

ultimately i wish i just didnt have to do this

[clear sedge]

[opaque sigil]

who says free is a syscall

[real sierra]

but i really dont have an upper bound on transposition table parent count

[clear sedge]

free isn't a syscall??

[opaque sigil]

it depends

[sage crag]

free is a syscall

[tender river]

munmap is

[sage crag]

but libc free doesnt necessarily free

[tender river]

[opaque sigil]

libc malloc/free do a fuckton of micro-managment of memory to avoid calling to the os at all costs

[sage crag]

because syscalls are really slow

[tender river]

syscall overhead is pretty big and can only allocate in pages

[opaque sigil]

[real sierra]

[clear sedge]

[real sierra]

reminds me of some object pooling code i wrote once

[sage crag]

on a pure stdout throughput test fprintf (which locks, writes to, and unlocks a buffer) hugely outperforms the write syscall, which doesnt even do locking

[tender river]

:idathinking:

[sage crag]

like, if write reaches 20 MiB/s, fprintf reaches 1GiB/s

[tender geode]

Their words not mine

Happy birthday Shiro!

[real sierra]

[clear sedge]

std::endl vs '\n' moment

[sage crag]

even more if you swap fprintf for fprintf_unlocked

[sage crag]

no not really

[clear sedge]

doesn't std::endl force a flush?

[opaque sigil]

so does '\n' for stdout

[tender river]

i have this in my cp template

#ifdef LOCAL
  #include "lib/local.hpp"
#else
  #define VDBG(...)
  #define DBG(...)
  #define endl '\n' // remove this line for interactive tasks
  #undef assert
  #define assert(expr) (expr) || (__builtin_unreachable(), 0)
#endif

[tender river]

flush is a bit different

[sage crag]

[clear sedge]

Inserts a newline character into the output sequence os and flushes it as if by calling os.put(os.widen('\n')) followed by os.flush().
https://en.cppreference.com/w/cpp/io/manip/endl

[real sierra]

thats a fun thing i discovered while making the C API

[clear sedge]

'\n' wouldn't necessarily flush, no?

[tender river]

\n to cout doesnt flush

[real sierra]

in C, printf containing a '\n' usually flushes

but not always

[opaque sigil]

am i misremembering something

maybe i am

[real sierra]

cute chess happens to be one of those "not always" cases

which is why the api has calls to flush after every important printf

[opaque sigil]

io is annoying

[tender river]

and the classic ios_base::sync_with_stdio(!1); cin.tie(0); cout.tie(0);

[sage crag]

phrr

[real sierra]

📣 phrrr

[obsidian mantle]

soo rn i cant use bitboard.h

in cpp

[real sierra]

[tender river]

extern c

[real sierra]

how come

[obsidian mantle]

or can i

[real sierra]

i pushed a change today

[opaque sigil]

it worked fine for me

[real sierra]

did that break or fix your issue

[obsidian mantle]

oh i didnt update

what issue

[real sierra]

the one you just described

[obsidian mantle]

no i never touched bitboard yet just wondering

[real sierra]

there's a now chessapi.h and a new bitboard.h

the latter should support C++ and MSVC DLLEXPORT now

[rigid timber]

yay

[clear sedge]

i like how steam just shows me a big blue delete icon next to proton experimental

[opaque sigil]

ah i remember now, stdout is only line-buffered when it's a tty

you should try playing the new proton experimental

[clear sedge]

i physically can't it just has the delete icon

[tender river]

i only use lts protons sorry

[nocturne olive]

https://www.youtube.com/watch?v=fxFL_BGev8M
This true?

[opaque sigil]

[sage crag]

Java is the most used language but C is the language java is written in

c

[clear sedge]

is java the most used?

[sage crag]

by popular metrics

that arent great

[clear sedge]

first i've heard of it

[sage crag]

[opaque sigil]

sth sth 3 billion devices

[tender river]

java is in every intel me and in java cards

so maybe

[clear sedge]

that's exciting and terrifying

[tender river]

the vm approach worked really well for them

and nobody really made a vm standard that could compete

wasm is kind of getting there?

[real sierra]

hblang will overtake jvm

[tender river]

hblang not vm

hbvm vm

[clear sedge]

hbvm

[real sierra]

semantics

[clear sedge]

[tender river]

shiro i'm gonna right bitshift you

[clear sedge]

is that not against the server rules??

[real sierra]

barrel shift..?

[clear sedge]

DO A BARREL ROLL

[tender river]

rule 16. you can shr a shiro

[real sierra]

shro

[clear sedge]

shi

means rain in my native language

[real sierra]

[clear sedge]

you must be my long lost twin

[fierce ether]

javac is made in java

[real sierra]

i actually dont know if id rather do a linked list or realloc for this array

[clear sedge]

isn't openjdk's javac made in c?

[sage crag]

"rustc is made in rust"

[real sierra]

linked list has token overhead of the recursive call

[sage crag]

self hosted compilers are secondary implementations

[real sierra]

realloc has token overhead of reassigning the realloc'd buffer

[fierce ether]

the standard library is also made in java

[clear sedge]

the standard library uses jni

[sage crag]

what are you going to write the standard library in except the language its written for

[tender river]

eventually bootstrap compilers become secondary

as soon as self hosted compilers reach a fixpoint

a better argument would be llvm

but doesnt apply to java

[sage crag]

stupidly, frankly

[real sierra]

[clear sedge]

it's the year of the graalvm everyone!!!

[real sierra]

i feel like that's every compiled language developer's dream

[fierce ether]

the JVM is also made in c++

[sage crag]

its a fun goal but not useful for most languages

[tender river]

just compile 10 years worth of compilers and you're done bootstrapping from source it's not that hard

[clear sedge]

it's a useful metric, it means your language can make a language

[sage crag]

[tender river]

actually self hosting is one of the most popular projects to make in a language

langdevs optimizing for langdev

[sage crag]

"useful" <=> "popular"

[olive sable]

surely

[tender river]

its reasonably useful since it instantly makes you your language's user and you don't have to maintain a compiler and a second big project as a test suite

[obsidian mantle]

why are you guys yapping and not mastering your bots

[real sierra]

holy

[tender river]

https://cdn.discordapp.com/attachments/703535223702552578/1284111550227152940/bounce.gif

[real sierra]

im working on my bot rn

i gotta catch up

my tokens

[tender river]

im pickling bot can wait

[real sierra]

have you even started your bot

i thought you were the one saying a month might not be enough time

[obsidian mantle]

i have a question
what do i do if my bot beats random bot every time?
give it to stockfish to analyze moves?

[tender river]

ye because no time i have picklesto pickle

[obsidian mantle]

(its not beating yet but all i need is checkmate mechanism)

[real sierra]

it's really up to you, but there's a few ways you can try testing

[tender river]

(im procrastinating on stuff i have to actually do)

[obsidian mantle]

i was wondering if we could exchange our bots for testing

is it allowed

[real sierra]

you can play games against it yourself, but get all of your moves from lichess or something

[real sierra]

sure, if you can talk someone into doing it

i dont trust people here not to reverse engineer my code for fun

[nocturne olive]

I should maybe test if at least with error catching to prevent crashing from the bug in the piece getting stuff my bot does well

[real sierra]

i also dont think anyone wants to run a random binary i upload

and im not sharing the source code

[nocturne olive]

There's actually a ton of overhead from just priting errors so I have to remove that

[fierce ether]

the rewrite has brought me down to 756 tokens

[real sierra]

when I wanted to test my bot I just recruited random people from genchat to fight it

[fierce ether]

but the bot is stupid

[real sierra]

lots of room for new features

[lilac lantern]

Hey superbox, funny question when you have a few minutes, is your neuros.click ftp server up

[nocturne olive]

Silly
I definitely need a better way to test mine at some point

[nocturne olive]

Oh good catch, something had crashed it

[trim valve]

I am tempted to make a matchmaking thing and throw it on a cheap vps

[lilac lantern]

Hopefully wasn't me

[real sierra]

i found it valuable, playing against my own bots doesn't show me any weaknesses

[trim valve]

but unfortunately people who don't write in rust, c or c++ exist and need bespoke ways to run

[rigid snow]

ok

[trim valve]

so I have the mildly cursed idea to basically make a "proxy" mode

[nocturne olive]

True once I have a better bot I should try that
Right now though I have to figure out what causes it to be too slow/stall

[tender river]

gcc is also compiled

[trim valve]

which would also absolve me of running anything

[real sierra]

[tender river]

sorry i meant ghc

[real sierra]

yeah im planning to deprecate threads.h

[nocturne olive]

Weird

[tender river]

[real sierra]

we're just gonna go back to pthread.h and windows.h

[nocturne olive]

It works within IntelliJ but seems to break in CuteChess

[real sierra]

uci is made to be run over the web

[trim valve]

which is probably the better solution anyways as long as noone tries to be sneaky and throw in stockfish

[real sierra]

in theory you just let people connect and print moves to you

[trim valve]

ye

[nocturne olive]

And I have no clue why

[real sierra]

over a network connection of some kind

[rigid snow]

telnet

[real sierra]

idk how the end user gets the match results tho

[trim valve]

maybe I should throw in 32 core stockfish on max level to make people think I have a good bot

[real sierra]

the bot will consume all stdin so

i guess you just mybot | tee | telnet ... ?

[trim valve]

i was probably gonna write a little wrapper program regardless

because I think it'd be useful if every bot wasn't just "My Bot"

[real sierra]

about that

[rigid snow]

[real sierra]

you can actually configure the fields sent in the uci handshake

i figured ppl didnt wanna waste tokens on it

[tender river]

entry point just means it wants a main

[real sierra]

so i just did it this way

[trim valve]

i can't edit chessapi.c really

[real sierra]

my point was more that if people are connecting over a web interface with their bots

they can change this on their own

[rigid snow]

ye i'm showing that it compiles with /std:c17. that is a linking error not compilation

[real sierra]

this name will be sent over UCI to the gui host

so you would get a bot name

[trim valve]

actually random question but would it make sense to throw an ifndef before those so people can pass -D to gcc

[tender river]

i kinda expected c++ to not work with it c++ is not a superset of c

[real sierra]

sure, that can be done

[opaque sigil]

it works with gcc and clang

this is just msvc being msvc

[trim valve]

not entirely sure if -D is the right thing but

[nocturne olive]

It also works in regular terminal
Something is wrong with CuteChess, potentially caused by the errors happening due to the API returning invalid stuff

[tender river]

its fine because the c++ standard doesnt support many of c's new features

[trim valve]

ideally it'd be runtime

actually potentially have chessapi.c read a config file from the cwd or something idk

[real sierra]

[real sierra]

that sounds like a pain

and also not really that important

[trim valve]

yeah

[real sierra]

-D is probably plenty

[opaque sigil]

it works which is all that matters (pls don't use any of the weird new c features shiro)

[tender river]

please do

[opaque sigil]

does anyone even rely on it working in a c++ context

[rigid snow]

uhh node bindings

that's about it

[rigid timber]

so how would that work with bindings that use a prebuilt shared library?

[rigid snow]

well, done "correctly" at least, using their build system,

i can just not do that i guess

[olive sable]

[opaque sigil]

talking about node-gyp? or whatever it was called

[rigid snow]

yeah

[olive sable]

well idk if ill even participate cuz i have a lot of work now

but ill try

[tender river]

you can link to c code without compiling it as c++ code

[real sierra]

i have to rebuild the library on my end when people submit anyway

[olive sable]

i bought the 3 monitor arms btw, shoudl arrive tomorrow. while i was at it i bought a larger mousepad

[real sierra]

ill just recompile libchess for each submission

[rigid timber]

fair

[real sierra]

but locally, just rebuild it if you wanna rename your bot (or yourself?)

[opaque sigil]

i hope you made sure they can carry enough weight too

[rigid timber]

if you wanna put in the work

[tender river]

better option

use env vars

[real sierra]

better option

recompile for each bot

minimal effort zero cost

[rigid snow]

shiro compute millionaire

[tender river]

shiro slave labor

zero cost

[real sierra]

libchess takes like 0.1 seconds to compile

we're gonna have what, 50 bot submissions? max?

5 seconds compile time

[rigid timber]

Shiro slave labor... hey Shiro calculate all moves for my bot kthx

[real sierra]

i have to hand check each source file anyway, i promise making libchess will not be the bottleneck

[olive sable]

they do 10kg each supposedly. so that probably means 10kg with sag and 5kg without sag, and my monitors weigh 4.2kg according to lg

[real sierra]

you will have one of the worst bots of all time with this strategy

i am horrendously bad at chess

[opaque sigil]

should be good then

[rigid timber]

but it'll be special

[tender river]

i promise you env vars are easier, you forgot about bindings

[real sierra]

[real sierra]

bindings just use libchess in their directory anyway

wdym

[tender river]

[trim valve]

I'm still mildly trying to figure out how I'd pass those vars to my rust bindings

[real sierra]

bindings devs making life harder for people

well if ur bindings dont run on my machine thats on u

[olive sable]

the mosuepad arrives in oktober tho

[tender river]

they do but they use their own libchess

[real sierra]

their own libchess?

[tender river]

(simply because i dont wanna make users set LD_LIBRARY_PATH so i link statically)

[real sierra]

[rigid timber]

Rust bindings compile it as part of the program for example

[trim valve]

(I also link statically)

[real sierra]

everyone compiles it statically

[nocturne olive]

So somehow recursion = 2 works but recursion = 3 breaks in CuteChess

[real sierra]

thats wild

ok i guess it has to be an env variable then

[rigid timber]

no, just some people

[real sierra]

bindings devs not following the spec and requiring the spec to change instead of adjusting the bindings

bad practice

[rigid timber]

then again at least for Rust it should be possible to make it a compile-time const somewhere in a config so it doesn't take tokens...

[tender river]

i did try to use .so but stack was being weird this is just better ergonomics

[nocturne olive]

And it doesn't seem to be a matter of it just being slow, unless running under CuteChess has insane overhead

[real sierra]

env variable will work

[nocturne olive]

Hopefully it's not some shadow effect of the weird get piece error

[real sierra]

i just have to figure out how to get env variables in c

[opaque sigil]

getenv

crazy name i know

[rigid timber]

it doesn't have an insanely short name?

I'd expect gete or ge

[tender river]

i wouldnt

[rigid timber]

but 3 character savings

[rigid snow]

https://github.com/cmake-js/cmake-js

[obsidian mantle]

wait aa second

what happens when i get color from empty square

do i get white

[rigid snow]

computer explode

[tender river]

i'm concerned

cmake is celebrated as an improvement

[opaque sigil]

-1

[tender river]

just how bad is gyp

[rigid snow]

ikr

[rigid timber]

bad

[rigid snow]

i mean cmake would actually be able to compile c as c and would not shove /std:c++20 everywhere it can

[obsidian mantle]

and its considered "true" in boolean checks

now i see what was my problem

[opaque sigil]

didn't shiro say c++ stuff would be compiled with c++23

[tender river]

c

[rigid snow]

msvc flag 20 is latest msvc supports

[tender river]

c++ is not specified

[rigid snow]

and yes

[tender river]

[rigid timber]

well it also has "latest", which is something between 20 and 23 apparently

[nocturne olive]

Howwwwww?????
This thing thought promoting to a knight and causing a draw by insufficient mating material was a good idea
And this is supposed to be the bot with recursion
It's literally dumber than the bot with no recursion
Either the bug with get piece also corrupts board states or my multipliers and score numbers completely suck

By the way what happens if a move is made on a board object that is already in draw in memory?

[real sierra]

make_move does not check legality

the move will be performed

[rigid timber]

everyone gets moves from chess_get_legal_moves tho

[nocturne olive]

Then that can't be the bug either

[real sierra]

so its generally not an issue

[nocturne olive]

Yeah except when there's that one broken castling move

[real sierra]

i thought there was an issue open for the castling stuff but it vanished

[nocturne olive]

But that probably has nothing to do with the weird bug with getting pieces that's causing me annoyances

Maybe I should do recursive move evaluation after all the non-recursive evaluation so it doesn't end up flipping the score negative

[rigid timber]

thanks Copilot, very cool

[trim valve]

[opaque sigil]

so who's writing the chess engine for the tournament

[rigid snow]

stockfish contributors

[trim valve]

incredible google pull

[opaque sigil]

[trim valve]

now I can truly realize my llm chess bot

[final tinsel]

1024 tokens, so itll need to be a small language model

[fierce ether]

ok i finally got the api to work omg

119 tokens is next gen

[rigid snow]

10 word vocab

[tender river]

tokens

[real sierra]

?

im just gonna use cute chess

compile all bots

[tender river]

JS bindings for shiro-chess, a fast chess engine in Rust

[real sierra]

add to 500 line cutechesscli command

tournament

start

dont use pc for a week

[obsidian mantle]

i get crazy numbers with this, something isnt right

so i get bitboard which is super big number for example 2,305,843,009,213,693,952
and it doesnt look like / or % by 8 is going to give something alright

i need to divide by 0b100000000

[rigid timber]

what are you trying to do exactly?

[obsidian mantle]

or am i wrong

i want to get coordinates of a piece to get distance

[real sierra]

you need chess_get_index_from_bitboard on that

[final tinsel]

git fetch origin
git checkout 7-create-the-lrzip-compression-benchmark-program

[obsidian mantle]

oh right

[real sierra]

that will give you a number 0..63

[final tinsel]

ope

wrong tab

[obsidian mantle]

im so bad

[real sierra]

[rigid timber]

fatal: not a git repository (or any of the parent directories): .git

[obsidian mantle]

would this work though

[rigid timber]

no

[obsidian mantle]

why

[trim valve]

maybe

it sounds like a very cursed version of bitshifting right

[obsidian mantle]

a2 is exactly this 0b100000000

[final tinsel]

From github.com:[FILTERED]
 * [new branch]      7-create-the-lrzip-compression-benchmark-program -> origin/7-create-the-lrzip-compression-benchmark-program
branch '7-create-the-lrzip-compression-benchmark-program' set up to track 'origin/7-create-the-lrzip-compression-benchmark-program'.
Switched to a new branch '7-create-the-lrzip-compression-benchmark-program'

worked for me

[obsidian mantle]

might save tokens

[rigid timber]

well a2 might be, yeah, but are you going to do that for every square individually?

[obsidian mantle]

i am not trying to shrink tokens yet, will do everything as explicit as i want
then when im done i will do the shrinking

[obsidian mantle]

"y" which is a number is alright
but i feel like getting x will be worse

a is 1
b is 2
c is 4 and so on

i need to get log2?

[obsidian mantle]

oh ye first i mod by this

then log2

by mod i mean % is it called "mod" in your language?

[tender river]

its called mod in english

[obsidian mantle]

oh

[tender river]

or in maths rather

[obsidian mantle]

esl

[tender river]

(full name is modulo but its shortened to mod in maths and programming)

[final tinsel]

i heard full name as modulus

[tender river]

modulus is what you divide by

modulo is the operation

[final tinsel]

[tender river]

a modulo b, not a modulus b

[fierce ether]

ok bots done

944 tokens

its a little stupid but it wins vs itself in 29 moves

[tender river]

it loses to itself in 29 moves

[final tinsel]

play it vs stockfish

[fierce ether]

glass have full

lmao

[final tinsel]

glass is completely full actually, 50% air and 50% water

also minor spelling error

[fierce ether]

glass have full

[final tinsel]

that sentence stunlocks my brain every time i read it i keep forgetting what im doing and needing to reread what i just did

[trim valve]

smh noob, I can win vs myself in two moves

clearly it will never be agi

[fierce ether]

doing it now

lose in 13 moves

lol

[trim valve]

this has thoroughly made me consider writing it

on one hand, would be cool

on the other, I have a cover letter to write

[nocturne olive]

This bot seems like a dead end
I may just have to go and figure out how real Chess bots are done and start from scratch

[trim valve]

[nocturne olive]

Ver8 was my peak for MK1

As soon as recursion was added it broke

And became stupid

[obsidian mantle]

no way

10 mates in a row

how do i run 1000 games

i want statistics

[nocturne olive]

Tournament

[trim valve]

there's a setting for concurrent games in tournaments too

(do not set it larger than floor(threads / 2) otherwise cute chess segfaults )

[real sierra]

makes sense

[final tinsel]

aw look at the lil baby om

[real sierra]

i dont know if i'll do concurrent games for the final tournament or not

i dont want one bot thinking really hard to potentially impact the others

it shouldnt but

you never know

[final tinsel]

[trim valve]

I can throw so many slow simd instructions at shiro's cpu that it downclocks and slows down my opponent

[real sierra]

SIMD explosion gambit

⚡ ⚡

[trim valve]

sadly you have no avx512 so its not incredibly easy to do

[final tinsel]

big brain

[trim valve]

(I say this having experienced ffmpeg downclock me to 4.2GHz at full tilt)

[nocturne olive]

I guess I'll wait for fixes and try a new bot from scratch

[nocturne olive]

How silly
My CPU can sustain 5.1GHz all-core with an undervolt, hopefully more once I get my new AIO

[trim valve]

does your cpu have avx512

[obsidian mantle]

my tournament freezes
i think my bot disconnects after several games

[trim valve]

because that is a different beast entirely

[fierce ether]

i would show yall the image but my king just decided to roam around lmao

[nocturne olive]

12th gen so no unfortunately

[trim valve]

idk how but I was drawing 190W for my cpu

which uh

[opaque sigil]

avx512 go brrrrr

[obsidian mantle]

what can possibly cause my bot to freeze in tournament?

[trim valve]

bot crashing

[obsidian mantle]

how do i force a crash to check if cutechess detects a crash

[trim valve]

there's a setting like "recover crashed engines" that forefits if a crash happens

[obsidian mantle]

because rn it doesnt look like crash it just doesnt make a move (and not processing it)

OOOOH

i think i know why

[trim valve]

oh yeah that was an issue at some point

just replace it with for (;;) {

[opaque sigil]

wasting a token

[obsidian mantle]

FUCK now random bot does the same

because i built it from example bot

[obsidian mantle]

is while true better or no

i think true is 1 token and ;; is 2?

oh someone mentioned it didnt see

[trim valve]

shruh

surely goto is less tokens

ok time to write a chess server and instantly implode

[real sierra]

bred trying to crowdsource move data

[trim valve]

do I look like I have the storage requirements for that

actually I wonder how bad that would be

probably not very

[real sierra]

you could probably do a usefulness heuristic and only store a move if its hash has occurred more than a certain number of times

[obsidian mantle]

[trim valve]

shrug

[real sierra]

that lets you trim storing lots of data for rare positions

[trim valve]

I get 40 gigs of storage to play with (realistically less because of the os but shh)

surely

[real sierra]

grandmaster

[trim valve]

I will be fine

[real sierra]

[trim valve]

shrio about to submit a bot that plays random moves on alt accounts to bloat my numbers

[real sierra]

me submitting two random bots to play 1 million games against each other on bred's server

wtf

[trim valve]

smh shiro

[stark needle]

i minimized my nn down to 88 NN parameters shiro i cant with these optimizations😭

fp4

[trim valve]

at least I don't have to run the bots myself

[real sierra]

you're taking part in the contest?

[stark needle]

i am taking part in the contest

[opaque sigil]

do ternary

[olive sable]

hi shadow

[amber fractal]

At least you can store it in 64bit if that doesn't consume too many tokens to desplit

[trim valve]

oh wait huh

[obsidian mantle]

i lost to it
(im bad at chess)

[trim valve]

they made the intel servers the same price and specs as the arm servers at hetzner

neat

smh why are the amd ones more expensive, I can't project my brand loyalty into vps purchasing

[stark needle]

my nn will be the most overtrained net in existence

trained on 100M games but only 88 nn params

😭

[trim valve]

overtraining isn't real, its called being goated

[tender river]

✅ more training means better network

[trim valve]

god I really should buy a new domain soon

[tender river]

are you gonna just stop training midway quitter mindset

[amber fractal]

As long as it isn't stuck in a suboptimal layout

I'm still not sure if I could pull off a submission but I can try to.

[real sierra]

training

[stark needle]

gemini is so carrying with the nn minification

i'm still trying to get it to compile

[tender river]

if you allocate 256 tokens for nn data thats up to 16384 bits which is ~10k params for a ternary network

[amber fractal]

The best method to play mhm

[tender river]

i have no idea how one would train such a network but i trust shadow to pull it off more than other participants

[olive sable]

true

[rigid snow]

you train it like any other network

[tender river]

gradient descent with 1.5 bits per weight

[rigid snow]

or alternatively you train it on sane precision and quantize later

[olive sable]

idk what that means

[amber fractal]

I think you have to do both because you'd lose too much data to quantizing

Your already starting very small

[olive sable]

wont it have to run on cpu btw?

[rigid snow]

network so small it cancels out

also simd

[amber fractal]

-# Quantizing is basically less precise data in exchange for less data storage, it just gets converted to floats later

[real sierra]

im gonna lose my mind

[stark needle]

the NN in question

[real sierra]

my duplex neighbours have synchronized their smoking schedule with when i open my window apparently

[real sierra]

i should just start dropping bricks out my window

[rigid snow]

and ternary in this case means -1, 0, 1

the three possible weights

[rigid snow]

you should synchronize your smoking schedule with them

[stark needle]

sam thank you for this deep statement

[olive sable]

no problem, im here till 1am or so

my numlock isn't tho

[trim valve]

how many floors

[olive sable]

i couldnt be bothered to install and configure that numlock exstension

ill just suffer

[real sierra]

only 1 floor

[amber fractal]

Shattered brick

[rigid snow]

https://cdn.discordapp.com/attachments/1072697081443131476/1239322682571489392/Nonko.gif

[real sierra]

i was feeling kinda sick so i took some nyquil

now im

holy

how do i code like this

feels like 5am

[trim valve]

caffine

[amber fractal]

Code on a random project unrelated to your current one mhm

[nocturne olive]

Some people are doing NNs now?

[cosmic sphinx]

caffeine after a sedative?

[real sierra]

many people doing NNs

[nocturne olive]

And meanwhile my bot is completely useless when it has recursion
I may have to start from scratch and look up actual techniques used in actual Chess engines

[cosmic sphinx]

im sorry but what the hell is that NN supposed to do with that amount of params

[faint sandal]

88 #neurotic-neurons damn

one was enough

[real sierra]

honestly maybe i need to do a rewrite too

bot starting to feel stapled together

not enough foresight

[nocturne olive]

I have no idea what I'm gonna do but I'm gonna have to start over from nothing and do something completely different

[sage crag]

discord.com/app

[real sierra]

i have a pretty good idea of what i need to do

[stark needle]

this challenge makes me want to start eating bricks

[nocturne olive]

I have no idea at all what I need to do
Right now I'm waiting on any possible API fixes and changes and then maybe I can start building a new bot

[sage crag]

shiro are you using hack font

where is your programmer font

[frank jacinth]

hai

[nocturne olive]

Funny number go

[rigid snow]

i see a#5

[frank jacinth]

i saw rook a2

i just didnt like it

[rigid snow]

yeah a2 is too low

i don't like it either

[sage crag]

@opaque sigil https://qwerasd205.github.io/AnnotationMono/

[frank jacinth]

bro its yung elo wtf 😭

[opaque sigil]

[nocturne olive]

Yeah it's a pitch

[rigid snow]

i know we talked about a#5 a lot

[nocturne olive]

Well there's a lot more of it now

[rigid snow]

[nocturne olive]

One of the datasets was copied and pitched up an octave

In the RVC inference stage so no quality loss from pitching up

[rigid snow]

mhm

[nocturne olive]

There's now apparently 4 examples all the way at G6

Though those are likely garbage noise

[obsidian mantle]

its time to update api, i finally need the bitboard.h

[nocturne olive]

That's crazy

[nocturne olive]

HUHHHH??
This model literally was just spawned into the world and it already has insanely low loss

[fierce ether]

2. e4 3.0s Qa5+ 
3. c3 7.0s b5 
4. Bd2 11s Qa4 
5. b3 29s Qa5 
6. c4 3.7s b4 
7. d5 1.7s c5 
8. Nf3 4.9s h5 
9. Bd3 24s h4 
10. h3 8.5s Rh5 
11. O-O 5.1s e5 
12. g4 9.9s hxg3 0.59s 
13. Kg2 18s gxf2 
14. Ng5 42s Rh4 
15. Rxf2 14s f6 0.56s 
16. Nf3 10.0s Rh5 0.52s 
17. h4 6.5s g5 
18. hxg5 12s d6 
19. gxf6 14s Bh3+ 
20. Kg1 10s Bg4 
21. Nh2 75s Bxd1 
22. f7+ 5.2s Kd7 
23. fxg8=Q 2.9s Be7 
24. Be2 33s Bxe2 
25. Rxe2 3.2s Kc7 
26. Na3 51s Qxa3 
27. Bc1 37s Qa5 
28. a4 7.8s Rh3 
29. Qe6 13s Rg3+ 
30. Kf1 15s Kd8 
31. Ng4 32s Rxg4 
32. Qxg4 2.9s Qa6 
33. Bg5 12s Qa5 
34. Bxe7+ 1.9s Kxe7 
35. Qe6+ 5.8s Kd8 
36. Qxd6+ 9.4s Nd7 
37. Rg2 5.1s Qb6 
38. Rg8# 10s, White mates ```

[final tinsel]

brotli sucks

lrzip is better

-# for my specific use case

lrzip was straight up ~17.3x faster and with a better compression ratio

[amber fractal]

Chat, I am trying to do anything but work on the NN framework. I really don't feel like trying to deal with a loss function without floats.
So much so I'm trying to see if I can implment what I need to with literally anything else.

[nocturne olive]

It is indeed very weird

[obsidian mantle]

vscode doesnt like this

(gcc compiles it alright)

what the hell is that "travel" doing

[lilac lantern]

@nocturne olive i think your ftp server doesn't like when i try to download a lot of stuff, appears to be down again

[nocturne olive]

This must be a bug that was introduced recently, hopefully a new FileZilla server version comes soon

[lilac lantern]

dang

i'll cool it on trying to mirror all your stuff for now then

[nocturne olive]

Try different files, and also what client are you using?

[lilac lantern]

filezilla

[nocturne olive]

Try WinSCP

[lilac lantern]

only 1 concurrent transfer

[obsidian mantle]

finally someone mentioned a program i know

[nocturne olive]

Maybe FileZilla server x FileZilla client is broken somehow

[lilac lantern]

2.1 TB over 500KB/s is gonna take a hot minute to finish

[nocturne olive]

What are you downloading? The entire subathon?

Oh no that's 1.71TB

[lilac lantern]

the whole thing, ideally

(i have a storage server i want to use)

[nocturne olive]

That's decent, has 6TB more storage than my system

Yes I have 29TB on my local machine

[clear sedge]

i have

this much

[nocturne olive]

SMOL

[amber fractal]

Would my 2TB laptop be smol

[nocturne olive]

Yes it is SMOL

[lilac lantern]

i also hit the jackpot because i bought a sff disk enclosure:

and they forgot to take out the sas SSDs

[clear sedge]

holy shit

[nocturne olive]

I don't have anything crazy like that, I need more funding to expand the NeuroSynthProject infrastructure

[lilac lantern]

Lackrack

[nocturne olive]

I only have this one machine

[amber fractal]

Counting the amount of storage I have, 2TB, 1.5, 2, 1. 0.5, 0.5. (9.5TB plus whatever I forgot existed) Man is my storage split up like crazy.

[nocturne olive]

Crazy indeed

[amber fractal]

Adding +1 for whenever I figure out the fate of this laptop's old drive

Unlike a lot of chatters here, I have the storage but not the amount of interface to use all of it. So most of it is cold spares

[nocturne olive]

I have some storage but I want a dedicated storage server of like 100 TB with parity

[amber fractal]

I'd be fine with a desktop rig to at least collapse what I have

[nocturne olive]

I just have a major lack of money

2 NeuroSynth comms from Aurum did not do much

Just paid for my new CPU cooler

[amber fractal]

Storage server is far off for me as I'm light on data usage

[nocturne olive]

I use so much data a storage server makes a lot of sense

[amber fractal]

I don't think there is much tech I'd put cash into next. Maybe just maybe GPU, but if I swap for a desktop VRAM will already be less of an issue.

[nocturne olive]

I personally would like more of just everything

Either more 3090s, or a 4090D with a 48GB mod
But it's so pricy, I would need to get a ton of NS comms and nobody wants those somehow

[amber fractal]

I feel like people mostly get the songs they want officially, so not much interest in unoffical

[nocturne olive]

How silly
NeuroSynth is really good though

There's orders of magnitude more songs than have been covered by Neuro and Evil

I'm sure there are ones people would want to see

[amber fractal]

Don't know what to say, people are just people. I'm not someone who is into marketing

[cosmic sphinx]

yes and no

the joy of it comes from them singing it live on stream

[rigid snow]

potential isn't in covers

[cosmic sphinx]

not just the fact of that cover being available for you to listen

[nocturne olive]

Kinda unfortunate
Because despite being quite good at this, I can't do that

[cosmic sphinx]

i dont think you can run vedal's channel for him LULE

[nocturne olive]

Vedal already doesn't make the karaoke anyway

[rigid snow]

i feel like neurosynth is best for unofficial songs, as in original

[cosmic sphinx]

well, he still has to use his synth model to get the voice output, but that's just part of the pipeline already

most of the work as usual is done by pb

[amber fractal]

Not biased at all, but I do agree with that

[nocturne olive]

I'm sure PB does that too

https://www.youtube.com/watch?v=49CrynNZAR0
Well there is also CHXI stuff

[rigid snow]

no way vedal touches the songs at any point, he trained the model at most

even the inference is probably done by pb it's just easier

[cosmic sphinx]

yuh, but its probably as simple as just 'select audio file' -> 'press button' -> 'download synth audio'

[nocturne olive]

No, it's 100% not

That's not how vocal synth works

That's not at all how vocal synth works

[nocturne olive]

This would be RVC, which is an outdated and useless tech stack, which we also deprecated in favor of vocal synthesis

[cosmic sphinx]

afaik thats how you do all the processing with any NN model if you want to wrap it in an app

[nocturne olive]

Nope

NeuroSynth is nothing like that either

[amber fractal]

I wait for a neurosynth orginal