Weighted RNG Garbage: Implications and Suggestions

[primal stratus]

A brief overview on weighted RNG Garbage is that it uses the decimal as a probability to send an extra line. An example is a 2.5 attack has a 50% chance to send +1 lines. Based on this table https://docs.google.com/spreadsheets/d/16WBk3baLByaWmmqAUvMUx3lB0wTrtOWCaiXBcHbTuxs/edit?gid=0#gid=0, are the lines sent by a clear and the decimals act as a chance to send +1 lines. It was stated that the feature was changed to make damage modifiers less sudden.

However, I believe it has lead to some consequences that are more noticeable at the higher ratings. Keep in mind that it is still using the season one attack table which has more b2b bonus but it can be assumed that combo bonuses will stay the same.

TLDR: For me, sending lines should not be RNG. It affects both offense and defense by making the lines less predictable. Another is that it increases the damage output especially multipliers since the old system ignored the decimals, even if the first suggestion is used.

This was inspired by Opyu's Video
https://www.youtube.com/watch?v=0rpNtEHVlGo

Link to the table with decimals from Opyu's Video
https://docs.google.com/spreadsheets/d/16WBk3baLByaWmmqAUvMUx3lB0wTrtOWCaiXBcHbTuxs/edit?gid=0#gid=0

Implications:
Lines sent is less consistent especially for combos of any kind which includes b2b combos and multipliers which leads to unpredictable offense and defense. This can affect decision making of the players due to the different possibilities that can happen.

Examples:
A TSD to TSD Spike will send 10 lines (75%) or 11 lines (25%). If both players made the setup and one gets lucky. They can send more for free from a common spike pattern in opener and midgame.
A downstack example can be a 4 singles into a quad which can send 10-12 lines with the combo having a chance to become double garbage.
A 10 combo 4 wide can send 12-18 lines with it usually being in between.
DT and BT can send 12 (25%) to 13 (75%) lines at the start compared to C-Spins 12 (75%) to 13 (25%) lines.
And many more can be seen in the attack table.

The problem is it is not possible to reliably tell if a defensive or offensive move will do what you want it to do. Of course probability can be managed and understood, but I believe that it is unnecessary.

The problem exacerbates during margin time since the damage multiplier will give decimal lines to all attacks. This can also be seen in Season 1 b2b and is the reason of the higher lines sent in the off season tetra league and in customs.

The other implication is that on average, it increases the damage output now that decimals are being used instead of being truncated. Combos are the most affected by the change, having the chance to send +1 and becoming different cheese can become devastating, since wides and multipliers exist.

Suggestions:
For me, the first implication is a lot more problematic. Adding a counter that keeps the decimals to be used once it reaches above one. It may be visible or not or optional but the purpose is to remove the random aspect and to make for more predictable lines. A way would look like this.

Counter -> 0

Sending 1.5 lines
Counter -> 0.5
Sent 1 line

Sending 2.8 lines
Counter -> 1.3 (Add 1 to send and reduce counter by 1)
Counter -> 0.3
Sent 3 lines

Another is to make a clearer and cleaner attack table though I do not know what would be the values. Perhaps a combination will make for a better way while keeping the original goal of reducing sudden damage changes and to reduce the now extra lines sent from the usage of decimals.

Lastly, reverting is an option and if the issue is due to margin time then addressing a better margin time can be a solution also. My suggestions for that will be for another day though I do like the idea of adding a quickplay like system of telling the debuffs after a certain time.

Remember that these are all suggestions based on what I currently know about the system. If there are any mistakes that were made or other suggestions then post about it.

[tawdry monolith]

#1266852096154537986 (35 messages) was moved here.

[celest mango]

-# (↑ citation from osk about this being intentional)

[primal stratus]

I see.

[fringe sage]

Noooo I hate 4 widers

[hollow dawn]

is the garbage sent complete rng or based on seed?

oh nvm it is i saw opyus video

[broken shore]

i did suggest keeping the decimal as well but the more i think of it, the less useful it seems. if the 2 players are at different piece counts (which is very normal), the decimal will be a different number, leading to still unpredictable outcomes. yes, the number is there for you to analyze, but no normal player will memorize every decimal in the attack table, so it is about as random as before.

[steep jackal]

I do think implementation of rng is a bit damaging at higher levels when the sends are supposed to be the same. people can get cheesed out of nowhere

[primal stratus]

That is true that it only makes it less RNG, but it still will likely feel random. I originally thought of the idea to truly avoid randomness but it seems like more has to be done to allow decimals to work as they have been ignored previously. The intended goal I believe was to make damage modifiers ramp up less sudden, so going after margin time mechanics may be better.

[primal stratus]

That was my biggest concern which made me create the thread. Garbage should always send the same lines and the decimal keeping solution that me and baron made does not fix that. It can really be a problem as Tetris has always been a game where 1 line can decide it all especially in higher ranks. Fixing the table combined with the decimal keeping may work so that only when margin time is on or damage modifiers are used will the mechanic be active.

Reminder that the extra lines sent due to decimal usage still needs to be addressed. Honestly, I think about just reverting it to the original or to remove decimals from the attack table entirely to let modifiers affect sending in a clearer way.

[past iron]

I was reading osk's reply - this change allows us a lot more freedom and leeway with using modifiers, multipliers, etc as damage is less atomic. This also introduces inconsistencies in attack, examples from @primal stratus above. imo I don't see an issue with margin time, although modifying it can result in even longer matches.

Agree with @primal stratus. Defending opponent's attack by performing the same attack may introduce a line of cheese which has a huge impact in gameplay. I don't like the idea of introducing RNG, since a line of cheese can be a huge deciding factor on a player gaining a disadvantage.

[cinder glade]

In more competitive play, any amount of unnecessary RNG can greatly ruin the experience, and this is exactly that type of unnecesssary RNG. For more casual players, having seemingly random attacks can be very confusing since I'd assume most of them don't even read the patch notes.

The type of play you'd expect would also greatly change, with much cheesier games being much more common. B2B battles resulting in cheese should not be a thing and it greatly lessens the enjoyability of such playstyles.

@primal stratus's suggestion, in my personal opinion, is a much better implementation, but this leads to a different problem of showing the counter. The garbage meter does a great job at displaying exactly how much garbage is in the queue, but only for complete integers. Assuming that some floating point bullshittery™️ is in play, it will be borderline impossible to show how much garbage is actually in the queue without explicitly showing it as a number.

In either scenario, the root problem is forcing the garbage to be more atomic.

Most games use integer amounts of damage for a reason; it is much easier to understand that 10 is 10, no more and no less. If the damage value is 8.822416284, the number starts feeling less concrete, and requires some rounding on the players' part to be understood, making the game significantly harder to play precisely. There is a reason every single RPG, FPS, and whatever game tend to use solely round numbers. When was the last time you saw an attack deal half a damage point? It shouldn't

Ok so osk wants to make the multipliers/modifiers etc. more atomic, and players want the garbage to be reverted. The solution? Change something else that is not the garbage; change literally anything else that isnt already discretized. If you really want to change the garbage, try to do it in such a way that doesn't result in massive rounding errors, such as by grouping excess fractional garbage to after the b2b chain.

ok so imma stop yapping now
im gonna be 100% honest and i may be wrong about a lot of things but thats my personal view on the current situation

also this isn't meant to be an attack against osk, just a (slightly biased) criticism on the implementation

also

TLDR; pls don't make decimal garbage/attack something to consider in any way, shape, or form if possible

[west quartz]

This might be too janky, but what if the decimal counter was reset to 0 on combo reset ? This would remove a lot of the randomness since when starting to attack the counter is always at 0.
E.g.

• you send single->tss->double : you're doing 0 + 2.5 + 1.5 which will always give 2 + 2 attack
• at B2Bx4 you send tsd->tsd : you're doing 6.408 + 6.478, you will always send 6 + 6 (cuz your counter stats at 0 at the beginning of your combo)

[primal stratus]

The solution that was made by baron and I does not really solve it, only remove RNG, similar to the first response. Garbage multiplier might be too strange when decimals are given notice. At least in truncating, it guarantees. I was think about how to display until I realised some of the decimals are very strange.

I also completely agree with unnecessary RNG tampering experiences in both casual and competitive play. Many games often have fixed values built in, especially in stats like damage, health, etc.

[primal stratus]

That may work, as that should fix all of the RNG without having to keep track of damage. Showing that is still a problem, and I have not thought of if it were not shown. It is important to notice that this does not fix the solo sending attacks like TSD, Quads, TSS, etc. Overall, I believe it is a more solid version of the old solution, sacrificing a bit of the original goal but making lines sent very consistent which is quite nice.

Still, I don’t believe there is a way to make all lines sent feel less linear because the values are small (0-10). Back to back t-spins and quads by themselves will most likely be sudden when garbage multiplier is increased due to being a small number.

Ex for solo sending:

4 x 1.2 = 4.8
4 x 1.25 = 5

Multiplicative increases are just hard to make not sudden in low values, especially when ignoring the decimals. Conversely, using those decimals for most methods lead to unexpected consequences and I and it seems that many believe that it is detrimental for Tetris. Finding a balance that both reducing sudden multipliers without adding consequences might be too difficult.

[broken shore]

i did actually offer this, but ultimately any solution similar to this basically tries to revert back to non decimal garbage as much as possible, so why not just do it completely the old way

[amber vessel]

What if we make garbage aminos that are like 0.4 high n shit

[primal stratus]

That is probably not possible.

[celest mango]

actually why not? cultris has smoothly scrolling garbage and people never seemed particularly pissed about that

[tiny mortar]

why is there decimals in the first place

[solid bluff]

mathematics

[past iron]

It's the way that values are calculated for the attack table. It's just that previously, the floor function was used to remove the decimal.

Currently, the number is rounded based on weighted probability. Examples:
4.9 value has 90% chance to be 5 attack (10% chance 4 attack)
1.3 value has 30% chance to be 2 attack (70% chance 1 attack)

[tiny mortar]

So the gambling update rly is a thing wow i was joking when i said that

[crisp flax]

i should test this but does singles have probability to send

[solid bluff]

doesn't seem like it

[hollow dawn]

you can no longer properly predict garbage cancelled/sent now

maybe you bank on a 90% but it fails

[solid bluff]

i think that's part of the plan, not allowing it to be so deterministic

[karmic valley]

how come this has been suggested like 3 times in this chat already

[proper geyser]

leftovers are equally rng honestly

nobodys gonna track that

it only works if its turned into an actual proper game mechanic

with ui and the such

[broken shore]

-# thats the same thing as this btw

no normal person will keep track of it, all it does is prevent someone from being exceptionally lucky and getting most of the 1-line bonuses

[west quartz]

True, and this can get quite substancial with 4W, where you could get a variability of about 12 to 16 lines garbage depending on luck

[celest mango]

https://tetr.io/about/patchnotes/#chlog_BETA_1_1_0 includes a toggle for this behavior

i noticed the TL preset (as well as default in general) actually defaults to DOWN...

[proper geyser]

if it was rng it wouldve been mentioned in the s2 changelist

[median tide]

So the change has been reverted for now?

[timid merlin]

Its down for rn

Which was used for season 1

[crisp flax]

is the table still different from ss1 with that change though

we will see when the formula is out

[tropic kindle]

Down

rng is new and only used in beta versions

[timid merlin]

I said down

and rn means right now xd

[violet jetty]

rather than rng it would make more sense for it to round down and save the decimal
for ex a 3.5, 6.2, 7.6 attack would send 3, 6, 8 and leave .3 in the decimals storage

[celest mango]

that's still effectively RNG though

who's keeping enough track on that

[violet jetty]

it would be consistent every time you did that setup

rather than being unpredictable

i still think strictly rounding down is better though

[primal stratus]

I agree

[karmic valley]

players on their way to suggest the exact same thing that 697 other people have suggested:

[violet jetty]

fair, but that's a bit much to backread

[wraith bone]

Personally I think fractional carryover and rounding down have their pros and cons, but both are way better than RNG. I've seen plenty of rounds where someone comes back from a board where 1 more row of garbage would've KO'ed them, that's a terrible thing to leave up to RNG.

Puyo Puyo does fractional carryover, although 1 extra Puyo is much less likely to swing a round compared to 1 line of garbage here.

[tribal adder]

How about fractional carryover that resets on combo?

[west quartz]

Pretty sure it was suggested somewhere already

[tribal adder]

Yes. I like that more than just fractional carryover

Though I’m not sure if I like getting 2 lines then 1 line then 2 lines of cheese when someone else 4w. It doesn’t feel intuitive

[celest mango]

yeah that's the primary issue, it's still effectively random

[cinder glade]

but its 100% deterministic which solves half of the problem

2 completely identical sequences of clears should always cancel out
edit: ok this is a tad bit wrong with some implications but my idea is still there (im just shit at getting it into words)

[broken shore]

thing is, if we wanna buff combos by leaving the decimal, why not just buff the combo table
(not advocating to buff the combo table)

[cinder glade]

weighted rng does not do that but carryover does

im ngl i think the combo buff is mostly unintended

[tribal adder]

What I don’t want is a 4 wide combo where lines sent is something like 0, 0, 0, 1, 0, 1, 1, 1, 1, 2, 1, 2, 2, 1, 2, …

Which the carryover has a small problem with

It just is not intuitive. Just round down

[celest mango]

weighted RNG was deterministic tho

[tribal adder]

It is because Math.random() is deterministic

[celest mango]

no, it was baked into the seed that decides where garbage columns are (you and your opponent get the same seed)

so "2 completely identical sequences of clears" did cancel out

[past iron]

philosophically everything can be deterministic

[celest mango]

literally weighted RNG was

[tribal adder]

How is the seed determined?

[celest mango]

server gives you a number, your client does magic with it

[broken shore]

aight lemme inspect my internet traffic or whatever
finds rng seed
ok sick now i know that my b2b x 4 quad will send 7

[celest mango]

obv you couldn't use that information in any meaningful capacity

[celest mango]

but implying that weighted RNG didn't do this is simply lying, as far as i know

[past iron]

you can also determine if opponents attack will round up or down lol

[karmic valley]

well carryover is also deterministic and more predictable than rng

[celest mango]

in the midst of battle? realllly? i guess people could hone in on the game sense but i don't really see it ever making sense to an outsider

[karmic valley]

more predictable as in it doesn't vary as much as rng

[west quartz]

Yeah pure RNG has the problem of higher variability from a spike perspective, especially on 4w

[violet jetty]

amount RNG seed thing is shared with garbage hole RNG,
honestly idk if that's good because it becomes even more unpredictable

is there a strong reason it was changed from round down to rng?

[crisp flax]

does knowing the seed even do anything since the rng is done serverside and your client still only know whatever the server says the rng comes out as

[crisp flax]

you dont know the server rng algo

¯_(ツ)_/¯

[celest mango]

well initially

this change allows us a lot more freedom and leeway with using modifiers, multipliers, etc as damage is less atomic
it's been removed now though lol

even from QP2

[abstract umbra]

fractional garbage that sticks around in your queue and can be canceled (but won't enter your board)

[proper geyser]

its still in qp2

[violet jetty]

not exactly. the RNG is also tied to the garbage you receive so it is not determinable at the start of a game given solely the attack you send
however it's not relevant because the brain can't process RNG
it doesn't matter that it's technically not random bc in all reality it's random to you

that would be useless imo
like how would that actually help you
it's not like you can change how much it will send

[celest mango]

oh i figured the replays breaking implied it had changed

but i guess "undefined" is rounding down

[proper geyser]

yea

[crisp flax]

that is queue rng

that sure

not garbage rng

[violet jetty]

garbage rng is the same seed

same formula for numbers

[gentle portal]

hm is it? that seems like it would technically lead to some weird divergences. like the number/ordering of garbage segments generated vs. fractional attacks added would change either a) the position of attacks, or b) the outcome of the fractional add

like how in GBA Fire Emblem games you can manipulate crit chance by snaking the move cursor so the A* takes different 50/50s, thus skipping random numbers used from the same source for other things

[violet jetty]

correct afaik that's how it works based on what I heard
I might be wrong tho, haven't actually looked at beta code
there's 1 queue RNG and 1 garbage RNG
doesn't matter tho cause it seems random
really missing the point of this thread

[gentle portal]

well, it seems like an additional thing to critique about the RNG implementation; the whole point of the shared seed for garbage is that the outcomes should be the same for both players. it seems to undermine that if the hole positions can differ for either player, or if one player can get that fractional converted into a +1 when the other would not, just due to subtle re-ordering of in essence the same actions

even if in practice players aren't reading the hole positions into midgame, it still seems kinda bankrupt if the one player can get a luckier/unluckier hole position because the other used something with a fractional value

[amber vessel]

what about no garbage whatsoever 🙂

[karmic valley]

who tf needs versus when we got score attack 🗣️