#suggestions
1 messages · Page 5 of 1
good point but i guess it would look too boring
pp counter
it'll be different values though because pp is different on akatsuki rx
^
they have a dll for akatsuki rx
oh that's quite cool
btw i made one rn :D
looking good
@covert mirage thx for the fonts :D
:c
jk pp counter is looking good
if vedal uses your overlay, then it would be better using same fonts for the counter
yeah but there is one problem
pp counters are integrated in the overlay
so wtf now

GG @queen shard, you just advanced to level 3!
@wide grotto stfu
yeah stfu bot
you tell him
-find a way to increase fps so the ai can handle faster and more complicated maps at the cost of training speed
yeah it is
is that the reason why she is so shaky
GG @covert mirage, you just advanced to level 3!
no it is the reason why she is less shaky
oh no..
more fps = more shake
more pixels = more places to move = more shaky
h
oh
oh yeah
makes sense
is there a way to smooth her movement
is there a way to smooth her movement
increase neuro-cam resolution?
but that would take a lot of data i guess
I don't think that would help tbh
GG @forest forge, you just advanced to level 8!
would her training speed increase if instead of doing
1 auto 1 play
she did
something like 5 auto 1 play
i thought she just calculates what auto is doing
yeah but playing is more fun to watch
the neuro's skin is just for her,she won't upload it?
is this some streamcompanion skin or something
its a osu!Sync widget
the new one is streamcompanion
ohhh
includes a pp calc too
pp calc 
GG @gritty comet, you just advanced to level 3!
wait what
wait what
no, the one with the pp display plugin and stuff
Gimme the forum link
is dead
osu! forums » osu! » Development » Completed Projects » [DEAD] osu!Sync | Share your beatmaps
Lol
everyone switched to streamcompanion
That's the one I was talking about rn
sync probably won't work on new builds of osu
Noooo
it already doesn't work in cuttingedge
osu!Sync works fine
hmm
On cuttingedge too
I've never gotten it running on cuttingedge
streamcompanion works way better
and doesn't depend on injection so you don't have to restart osu if you forget to run it first
An official plugin got updated a few days ago, so its surely not dead
which one
ortdp i think
the only problem I have with Sync is that I have to start a beatmap to make it work
is rtppd official tho
i thought yes
it is uploaded on official Sync github account
strange then
so yeah
it has a separate forum thread too
Oof peak of skill was 5 hours ago
sync was made by a German guy but is very Chinese now
Or just lucky map
maybe someone else took over
but if that's the case idk why it's called dead
if you want to learn C# to help osu!tools just go for streamcompanion
ortdp isn't even open source
streamcompanion is friendlier
it is open source lol
GG @queen shard, you just advanced to level 4!
it pretends to be
but it only includes one file
and won't compile
and the dll is obfuscated so you can't easily use dnspy
oh crap
xd
gross
idk, streamcompanion doesn't let you create own custom widgets
also c# 👀
yea
plugins seem fine to me
JAVASCRIPT OMEGAROLL
be smart, use Electron 
i don't think u know what smart means
:D
fuck sc distracted me from castle wars
bruh moment
anyway C# is the best if you disagree you are a revisionist
also c# is becoming multi-platform with .NET Core anyway
can we please not spam #suggestions without any? 
changing channels is hard
understandable
GG @bright whale, you just advanced to level 1!
will move
ok
request bot like mikuia
When I say I'd use a linear activation function, I mean something like this:
tanh assumes a normal distribution for the positioning of circles onscreen with a mean at the center, but this just isn't the case.
suggestion: blank out leaderboard, spectator and notification areas
suggestion: wipe out osu account on every update
add a pp counter?
already suggested, there is none that work with akatsuki's rx
fork streamcompanion and use akatsuki's oppai instead
though streamcompanion has a C# implementation of the performance calculator so vedal or someone else would either have to change it to use akatsuki's values, or run it in the background with Process.Start or something
alternative you could just say fuck it and use bancho's pp values since that's what most people play anyway
🤷
put a link to the skin neuro uses in panels or with a !skin command?
You can add an HTML snippet that reads the pp directly from akatsuki, and display it on stream. If you want a non-invasive way of showing her pp on stream
GG @pure sphinx, you just advanced to level 4!
wouldn't show live pp count for the current map tho
https://akatsuki.pw/static/rxprofile.js maybe this could be used to show her current total PP on her profile
https://akatsuki.pw/api/v1/users/scores/recent?mode=0&p=1&l=20&rx=1&id=35270 this for her most recent plays
https://akatsuki.pw/api/v1/users/scores/best?mode=0&p=1&l=20&rx=1&id=35270 this for top plays
I'm guessing that means no
im still reading it, so idk yet
akatsuki uses a different algorythm for RX
so calculating total should be easy
yeah we know that's the issue we had
since we wanted pp counter, but streamcompanion would have to be modified to use akatsuki's formulae
suggestion: make a #suggestions-discussion channel so people stop spamming #suggestions with things that aren't
for now move to #574720535888396290
seeing as the training system uses epochs couldnt you just divide 100 by the amount of epochs and use that output as the training counter
wouldnt be supoer accurate but it would work
super*
that's what I did before
but I'm using libraries for training now
so I'm not in direct control of events between epochs
wait if you play with the leaderboard in local only, since you dont send the score because offline ,its always going to be empty so it doesn't show ? or I'm miss remembering
it will see my plays but yes, it just often details to global
@forest forge since it got quite alright, how about you test it against a map it hasn't trained against yet?
did you apply my training suggestion?
and offset the current frame and target x,y to the next frame?
for training
suggestion: disable ingame notifications
@pure sphinx I don't think it would help tbh, why would it help?
I have disabled in-game notifications btw
@honest mason will do soon
because unless your image processing is faster than your framerate
you'll be telling your AI to go to x,y based on the current frame, but current frame = past by the time the network returns the value
also, have you done validation playthroughs?
checking if she's actually improving, by playing maps she hasn't trained on
Well, it's obvious she learnt in one way or another
GG @honest mason, you just advanced to level 5!
I can see loss values and she is improving trust me
I don't get why your osu client keeps saying "You're now playing relax", "You're now playing normal" ... etc
akatsuki server
:|
maybe to see the scores she do
scores
live pp counter + change the ready to play sound, its spooky as heck
I'm taking read to play sound suggestions
@forest forge try showing this in the overlay somewhere too
osu! forums » osu! » Development » osu!StreamCompanion - live pp display / ingame overlay!
it gives an overlay like that
someone first has to modify streamcompanion to match the akatsuki rx pp calculations
ah
thats an issue
wait, doesnt this run off of the client?
even if it shows inaccurate readings for the server, better then nothing
¯_(ツ)_/¯
also cant read osu forums while on akatsuki client
that overlay is bad imo
i meant i cant read the post xd
oh, rip
GG @covert mirage, you just advanced to level 1!
osu!StreamCompanion- as name might imply, this program is directed for osu! streamers.
Features:
Grabbing song title ANYWHERE you are in osu!
Built-in IRC twitch chat bot on TODO
Customisable map outputs
Key counter - how many times you clicked these 2 osu! keys since...(beginning of the stream?)
Play/Retry counter
Mods display
Twitch ==> osu! chat Relay on TODO
link the github repo
thats the repo
ok gonna learn C# just to make is possible to create custom pp counters
oof
you can customise how the counter looks
the ss i took was with default settings
ah ok
I wanted to create something like this
doesnt it output details to a file?
cant you read the files and use that?
itll be slower though, granted
i think i have seen some issue about adding a WebSocket interface
that would be much better than reading from file or requesting the API every 50 ms LOL
wait a sec
surely you could just change the font and colours to make it look how you want
then get the overlay to cover what you dont want to be shown
its not that far of a leap tbh
you could get most of the way there by covering parts of the counter up
static black image with obs should do the trick
¯_(ツ)_/¯
still won't be even close to the image I sent
already most of the way there by the visualiser transparent
some overlay tricks should be enough for the rest tbh
anyways html/js/css widgets are better
fair
GG @covert mirage, you just advanced to level 2!
you could just fork the repo and make whatever changes you want 🤷
no, because I don't know C# well enough :D
its getting Close Enough ™
the blue area could just be the overlay and the red area could show actual values
giving you a lot more to customise how it looks
there's no combo on the overlay and I does not have animations I had on my counter >:c
¯_(ツ)_/¯
oof, that pp counter looks uglier then this
apologies
its cool
@forest forge reeeee a radeon vii would prob serve better for workstation type loads then a 1080 ti
and you wont have to sell your house for it
fair point
but solely for neuro sama a radeon vii would probably do better then a 1080 ti
for games it does seem to suck a bit though
hmm I thought a 1080 ti would be better
its kinda subjective tbh
a 1080 ti is a good balance
dont get me wrong its far from bad
but for more work station oriented stuff it does a tiny bit better then a 1080 ti
for price obviously the Radeon vii would be better
is the jump from your 2x 980's to a 1080Ti worth it tho?
i dont want to influence you for / against selling your house lmao
oh and the other thing is that I can't use cuda on amd
amd uses openCL
@pure sphinx I only have 1 currently
not what your twitch thingie says :v
yeah I don't want to have to port it
also full disclosure: may or may not be an amd fanboy
¯_(ツ)_/¯
regardless ima just sit here and be jealous while melting my cat on my rx 570
understandable
btw, iinstead of going 1080 Ti, go 2070
this ^
1080 ti performs better for machine learning
yet*
@covert mirage not yet, it's for the future
™
i think you won't be able to use the tensor cores just like that?
tensorflow probably can
maybe slightly increase resolution @forest forge
Maybe add an !np bot since people ask for map links. Or display it on screen with osu stream companion or something
sug: make the donation-text bigger, we're no ants 【=◈︿◈=】
Thoughts
I think it will be too op if she will get 1000 pp everytime without choking
well
the point is
to get her to that stage
and how long it takes her and how she progresses
after that
she can start training on different games too
more anteaters
https://discordapp.com/channels/574720535888396288/574725266346344470/575731475127861289 you should pin this one too
scoreboard is messing neuro-sama up at the beginning of plays. I don't know how to fix it, but I'm pretty sure it is focusing on it in the beginning
shouldn't make too much of an impact
show the gameplay on stream from a spectators perspective, so we can see combo, score, etc, but feed neuro-sama the raw footage from the client?
train while watching
suggestion: take a screenshot when it's one miss or S or fc
suggestion: make the automatic inputs click the switched servers messages at the bottomright after selecting mods
if the gameplay on stream is from a spectator account, would it also be possible to change song while training?
also you could use streamcompanion to show the combo/score/etc on stream without it being in-game
^
maybe use a different language like javascript
cos as god tier as python is
it is quite slow
depends on the usecase
I really doubt it's better for this
PyPy often beats node as well
i don't know how good it is , but if you want python and C level of performance there is an extension called Cython
https://cython.org/
that's much different
I don't know if the libraries vedal uses are compatible with it
ye
I mean he could just be a chad and rewrite it in C++ 😘
pog
gop
you have to understand
there's a reason most people pick python for machine learning
also Cython has a much different syntax
Because there are many liberies and prewritten stuff to allow integration to be as quick as possible
But I mean you could it in C++
C++ is easy enough to write in
nice you have found out that tensorflow is partly written in C++
now you should understand why using C++ offers literally no benefit over python
gonna need some benchmarks for that
Fuck
The site is down
🙀
That is the article I saw a few years back
nice now maybe some benchmarks relevant to this usecase?
this is machine learning, not general purpose C++
the libraries are already mostly written in C++ with Python being the description layer
they're highly optimized already
tbh I wanna see if vedal can write it in c++ as a challenge xd
if you can point to machine learning-specific benchmarks that indicate C++ has a substantial enough benefit over Python in this specific application, and not in general, then I'd be happy to agree
coz I'm not gonna deny C++ is in general faster than Python that's obvious
who cares
GG @shadow knoll, you just advanced to level 1!
not me
I want to see him doing things useful instead
also your point was "C++ is faster" which doesn't even apply
ok who cares
My final thing, if I cannot provide evidence to my point then surely that makes your point just as invalid as mine as there is no evidence to ether side of the argument?
But anyway
Hope you have a good day my dude ❤
no, because your point is "This is how it is" and my point is "you are wrong"
your job is to prove you aren't
I've already given you plenty of reason you are
take some fresh air 😚
christ u guys sound like me and my best friend arguing
I mean I couldn't care less lol
my ex friend had an infatuation with C++ so I'm used to taking about how shit it is
mad cus bad xd
"I couldn't care less" says the person who is very bitter I said they were wrong
:~)
lmao ok
I think he should write his code in self leveling concrete
That would be cool
agree
when the blokes start arguing about which code is better and all you can do it print("Hello world")
🙏 🙏 🙏 🙏

i would say python is one of the better ones for machine learning
but i like js for oop
js is hilarious for oop
now there's something I can agree on

imagine having a shiba inu with a corn cob on its head for a profile picture
what your saying fam you watch swing out sisters calm down
imagine commenting on anime in a discord dedicated to an anime game
very smooth brain

imagine being such a beta male you use poggers in chat
LMAO
poggers
poggers
GG @shadow knoll, you just advanced to level 2!
THAT MADE MY DAY
thanks glad I could desalinate you xx
lul wut
GG @gritty comet, you just advanced to level 4!
:^)
oh pog
look we don't need to argue about code we all know it should be made with rockstar like ok mate
GG @lone creek, you just advanced to level 1!
hell yeah
Or we should have written it in assembly
^
^
^
^
probably ahs been said but an estimate time of the learning phase might be nice
@shadow knoll You could probably do a benchmark of it on your system for c++ and python if you need to prove a point tbh. And anyways, you guys are all wrong. tensorflow would obv be the most efficient in KOBOL
you could calculate.. how long it takes in average based on the lenght of the song (using previous data for reference)
You could ask it politely
i have a feeling that since the autobot spin so fast when you take screeshot of the game the cursor is always near the same place, that might screw up some of the learning on the spinner
GG @eternal jacinth, you just advanced to level 1!
That used to be a problem but not anymore
neurocam is 60 (ish) fps now?
yes
Have a huge watermark over play area when neuro is learning from auto
dont forget to add www.bandicam.com, unregistered hypercam 4, and EZ Screenrecorder
xd
I have got really weird and stupid idea
I just thought a little bit
That's too stupid
Ok, i will say
Heartbeat based on hardness spikes lol
... no
I knew
Yes, it's 3 am
hr would do nothing
Suggestion: make all the numbers a 1x1 png so the spinns are more consistant
But the sound is?! aaaaa
I just thought to point this out somewhere, the unknown might be fun tho
im too poor to sub so insert poor man sub here
free amazon prime trial smh
o n o
train individually for each map. It seems that it is worse at honesty today after having training with other maps. I think if you want it to FC a map, then it should just train for more epochs on a single map
GG @river silo, you just advanced to level 1!
training in individual maps leads to learning anti-patterns that make it worse for playing other maps (for example, it could learn something like that certain 2 black pixels on the upper right corner plus one black pixel on the upper left means it has to aim to the center, which might make it FC a certain map it is training, but it is not a real pattern that leads to playing maps in general).
GG @tight swift, you just advanced to level 2!
something needs to be done about the bot playing other game modes
I thought maps were already grouped by mode to avoid this idk what happened
Suggestion: i sub only channel.
@river silo That honesty play was with a completely new network with no training as a test
@glass stone yeah oops I fell asleep after setting it to play a certain map and without search criteria
you can group it without search tho
that way even if search goes wrong you won't end up in a different mode
won't help with random titanic flute maps crashing osu tho
yeah that's what I usually do
oh
someone mentioned in twitch chat that you can restrict maps by star rating in osu!settings
if you don't want neuro-sama to be playing 2*s when the search does an oopsie
I think is a good idea to make a NN for playing Poker Texas Hold'em. But 100% is illegal @forest forge
why would it be illegal, it doesnt have to be on an online game
Holy shit yeah
GG @honest mason, you just advanced to level 6!
@forest forge could play texas holdem on tabletop sim or a designated game/make it himself
probably make it myself
i am currently programming a Poker. 😃 (i am begginer)
which poker
GG @glass stone, you just advanced to level 13!
Texas Holdem when you have 2 cards and 5 community cards
I know how the game works ~_~
xd i just told it clearly so everyone can understa d
Snaking sliders so there is less clutter on the screen
Something to signify which stage of training she is on
GG @plain agate, you just advanced to level 1!
Twitch chat betting where players can guess the acc of each play.
y tho
Fun
I could implement it, but can not tell beforehand if the next map will be ranked / submitted to gather the acc
If you want betting on absolutely every map, then vedal has to combine it with streamcompanion
He should use stream companion anyway like we've been suggesting for literally years
this
:3c
bruh
The real snorlax
remove Why don't you run it from GPU? (most are aware now) and When will you turn off relax? (it's off now) from FAQ
uh lool
don't remove gpu
people still join and comment on the "cpu temperature"
even though it's the gpu
too late
sug: translate the faq to German
nein
and mod @glass stone
now thats a good suggestion
uwu
neri where actually are you from
I'm UTC-3 atm, -4 non-dst
but I usually stay awake late into the night
just awake now due to family events this weekend
yeah its good because I need people awake when my current mods are not
I'm also a NEET so I'm on pmuch always
chat is barely spammed atm but could use a mod for sure
modless times are pretty big sometimes
NEW ranked 1000pp cookiezi play!!
alright neri dont kill anyone
ok uwu
whats your twitch?
congrats pog
nericat
ill take that as a yes
yeah that is my profile pic
yeah I added you
I'll prob be back home in like 2 hrs
everyone gets banned
I'll start with vedal

i want neuro-sama to play smiling
sorry she's a neural network
I don't think she actually can smile
unless @forest forge
adds a function for her to smile
she smiles in our hearts
if ur not joking I think they meant this https://osu.ppy.sh/beatmapsets/450363#osu/966243
osu! » beatmaps » Lime - Smiling (Speed Up Ver.)
I meant vedal idk what's going on anymore
idk if this has any plausibility
also any updates on song requests
VIP to Royally_Jelly for high presence and good answering support
someone in chat wants a 300 counter too, like how we have 0x100 0x50 0xMiss
using DT would be an interesting way to decrease the actual time spent on training imo
the training only captures at a set fps though so there would be less training data
just double the fps ez
suggestion: make neuro-sama post updates to discord ( #574928403623444495 channel)
It might never learn how to play DT with high OD, the fps the net runs at is not high enough.
disable everyone and here
^
done
probably going to do a short relax pp stream in which we try and set some good plays, any map requests?
ROr, hONEsT TEa, biG bLACk, fREEdom DivE
How about some cold green eyes dt fl
best friends?
DrunkenSteiN
Last Goodbye
Shukusai no Elementalia, Soukai Rock, Caffeine Fighter, Now Loading, Ame to Kusari, Natsukoi Hanabi, Cycle Hit
and Harumachi Clover ofc cuz why not :>
(insert any mazzerin map here)
promethean kings +NC

Centipede
owoo
GG @vast hound, you just advanced to level 2!


