#machine learning

1 messages · Page 1 of 1 (latest)

ember loom
#

make it so good players can opt in to having a bot trained based on replays of their gameplay for people to fight and train with, and you can access these through a menu

will be hard to code, especially because you have 2 players both doing actions, but might be possible

t2+ only so we don't just have 90k bots

#

also they can opt out later if they want to do paid coaching or something

formal plover
#

Wait wait wait

#

Pause

#

Bot that plays like you

#

And you can fight yourself and see your weaknesses

#

The AI itself shouldn't be that hard to make

#

And your own replays are already stored on your PC

#

I'm not sure if this is possible, but it could also be trained on your own computer

#

So it's free

#

AND it can be part of the training/practice bot and coaching that has already been suggested

#

And you can upload yourself or share your bot with other people

#

And for example, try fighting Marlow's bot

#

This theoretically should be possible even if they don't add it, just from your own replays files

#

I might make this 😭

#

Since we can just train it off our own files

#

Omg there are so many possibilities for projects with pvpr

ember loom
#

(im ht5 and im good)

formal plover
#

Imagine a khalih bot in vanilla 💀

#

Just maces you all day

#

(I'm still traumatized)

ember loom
#

lol

#

idk if it would fully pick up the playstyle or just the timings of stuff or both but either way it could help

#

depends how you make it

hasty fox
#

someones tried this before

#

idk its really cool but

#

i think it would be a longer term thing

#

also it would probably be more like a bot that plays like a certain tier

formal plover
#

yeah there are definitely more important things rn

hasty fox
#

because it would be harder to have enough training data for each person

elfin spindle
#

@inland laurel He speaks your language

formal plover
#

it should be possible to do it independently with just our own replay files

formal plover
worldly blade
#

This is way above the Minecraft tierlist paygrade

formal plover
#

no this is awesome

#

the whole point is that this can be so much more than just a minecraft tierlist

#

it can turn mctiers into a full competitive game

merry rose
#

Machine-based learning that follows a ton of people to learn different playstyles is great, but it may be difficult to create one per person, you would need hundreds if not thousands of fights to get enough data. Cool idea

quaint kestrel
#

I’ve always love the idea tho

#

It’s you vs. you, theoretically it should end in a draw

#

But if you lose, it might be an indicator of getting worse, and vice versa for winning indicating being better

#

And yeah forcing you to play to your own weaknesses makes you more aware of them, and then seeing how “you” would respond is all the more interesting

merry rose
# quaint kestrel It’s you vs. you, theoretically it should end in a draw

People make mistakes and the bot will have difficulty making any mistakes (you can see this by xanicle's crystal bot, which is t1 in skill even while being machine based learning from a non-perfect player. It is a great practice tool though, but again I would not thing this is necessarily a way to recognize your mistakes or create a bot with your mistakes.

merry rose
quaint kestrel
#

Fair and good points

#

@grim bluff i am very upset at you for renaming me here.

merry rose
#

Lets not say such things please, even if its to a friend as a joke.

quaint kestrel
#

My bad

quaint kestrel
merry rose
#

xd

formal plover
#

also there would for sure be that amount of data for specific tiers

#

so t5 bot t3 bot yk

merry rose
#

I am refering to a per person bot, so I would have a custom Tujoro bot and you would have a Ravenclaw bot

#

That is too difficult

merry rose
formal plover
merry rose
#

Maybe, but creating a bot for every player when the majority will not seems like a waste of time

#

I still like the idea of bots for different tiers

formal plover
#

and the bot could be trained on our own replays and on our own pcs

#

(i need more projects)

#

the cv is looking empty 💀

shell crow
#

We getting taskmaster in Minecraft before gta 6

weary canyon
#

ngl this idea sounds amazing

ember loom
#

like maybe it checks your chance to backstab vs stun in axe and when you do it, and if you stun 50% of the time it will use other players' gameplay who stun 40-60%, not someone who backstabs 80% of the time

ember loom
#

i also dont have bug tester

#

doesnt exist yet i guess

mild hull
#

🤬

ember loom
#

?!?!?!??!

mild hull
#

They stole mine too

#

We do revolution

ember loom
#

should we ping hat?

mild hull
mild hull
ember loom
#

lets not spam

mild hull
ember loom
#

1 ping is enough

mild hull
mild hull
#

And we got ghosted

young thunder
#

@ember loom i have a feeling reinforcement learning would be better, people tried training rocket league bots on replays and it got nowhere (they had over 10k hours of replays), just by doing a ton of self play i think the bots would learn much faster and better than replays

ember loom
young thunder
#

its defo possible

#

i was thinking of making a cpvp RL bot for my A-level NEA

#

i have 2 years