Veril
Frame Savant
This project is intertwined with anything else the smashlab can hope to accomplish. Here we will compile "in-depth" and "simplified" frame and hitbox data, and the results of any analysis.
Goals:
I. Data for us:
1. All data that can be derived from PSA and Brawlbox, codified and organized.
2. The advantage permutations for all moves.
3. A reliable formula for hitstun.
4. A reliable formula for the point at which moves induce tumble (I'll dump enough examples that we should be able to extrapolate something, I already am really good at estimating these points but don't have an easy formula).
5. Quantifying the "human" element when testing frame perfect: we need to decide exactly how much SDI is reasonable, what is reasonable to expect from people in the way of DI and when, what should our baseline for "choice reaction time" be. This is not exactly what Bowyer or Adumbrodeus have been talking about, but it is necessary for each of their respective projects.
a. ex. I assume that people are able to input Quarter Circle SDI reliably on any move telegraphed by over their Choice Reaction time.
b. "" with all options that the players will both be aware of and able to execute the options available. Technical difficulty is only considered in very rare circumstances.
c. "" players aware of their options will always punish something with a -20 advantage or greater on reaction.
d. when a single dominant option is guaranteed to work, ie, it is a frame trap, players will overwhelmingly choose this option, and when they don't it falls into the nebulous realm of "mindgames" which the "option select" project will pursue.
II. Data for the masses:
1. Use a universal format that includes what players will want to know.
a. The hit-frames and first actionable frame.
b. Cancelable frames
c. Part Qualitiative hitbox data: ex. strong hitbubble, weak, etc
d. Damage by hitbubble
e. KO% by hitbubble with and without DI, center FD for any move which can KO under 250%.
f. The frame advantage "on-hit" and "on-block" for all moves
g. The % at which moves induce tumble, by character weight class (there is overlap so testing of all is unnecessary).
h. Hitbox/hurtbox images for all moves.
2. The best options characters possess to break strings and combos, based on positioning.
3. Frame traps pointed out, categorized and explained.
Sorry I was gone for so long, I'm gonna look over all these messages now...
Brawl Weight Modifiers
1.20 Bowser
1.16 Donkey Kong
1.13 Snake
1.12 King Dedede
1.10 Charizard
1.09 Ganondorf
1.08 Samus
1.07 Wario / Yoshi
1.06 R.O.B.
1.05 Ike
1.04 Captain Falcon / Link
1.02 Wolf
1.00 Ivysaur / Lucario
0.98 Mario
0.97 Luigi
0.95 Sonic
0.94 Lucas / Ness / Pit
0.93 Diddy Kong
0.92 Ice Climbers / Toon Link
0.90 Peach
0.87 Marth
0.85 Sheik / Zelda
0.82 Falco / Olimar
0.81 Zero Suit Samus
0.80 Fox
0.79 Metaknight / Pikachu
0.78 Kirby
0.75 Game & Watch / Squirtle
0.68 Jigglypuff
•the hitstun on all throws is roughly consistent across the cast at 0% (~same number of frames of stun). As would be expected, the hitstun growth rate varies by character.
•Throws with internal damage (ie, more than one hit) have a set amount of lag after the release point (I calculate the "lag" based on the frame the last hits damage is applied on... if you need clarification talk to me on AIM. This is the best and easiest way to find lag on throws).
•Throws without internally applied damage (single hit) have lag that varies (∂W) across the cast. Testing these is a nightmare and as such I'm only going to be testing a few characters single hit throws.
example of preliminary throw frame data test results from an earlier B+ set: kirby selected arbitrarily but I've done all of the characters at least once.
The most important thing you will ever learn about frame advance data mining:
moving that data around from my initial dump thread...
ICs visible hibubble .pac: http://smash.everalert.net/downloads/Fit...e_v0.1.pac
Goals:
I. Data for us:
1. All data that can be derived from PSA and Brawlbox, codified and organized.
2. The advantage permutations for all moves.
3. A reliable formula for hitstun.
4. A reliable formula for the point at which moves induce tumble (I'll dump enough examples that we should be able to extrapolate something, I already am really good at estimating these points but don't have an easy formula).
5. Quantifying the "human" element when testing frame perfect: we need to decide exactly how much SDI is reasonable, what is reasonable to expect from people in the way of DI and when, what should our baseline for "choice reaction time" be. This is not exactly what Bowyer or Adumbrodeus have been talking about, but it is necessary for each of their respective projects.
a. ex. I assume that people are able to input Quarter Circle SDI reliably on any move telegraphed by over their Choice Reaction time.
b. "" with all options that the players will both be aware of and able to execute the options available. Technical difficulty is only considered in very rare circumstances.
c. "" players aware of their options will always punish something with a -20 advantage or greater on reaction.
d. when a single dominant option is guaranteed to work, ie, it is a frame trap, players will overwhelmingly choose this option, and when they don't it falls into the nebulous realm of "mindgames" which the "option select" project will pursue.
II. Data for the masses:
1. Use a universal format that includes what players will want to know.
a. The hit-frames and first actionable frame.
b. Cancelable frames
c. Part Qualitiative hitbox data: ex. strong hitbubble, weak, etc
d. Damage by hitbubble
e. KO% by hitbubble with and without DI, center FD for any move which can KO under 250%.
f. The frame advantage "on-hit" and "on-block" for all moves
g. The % at which moves induce tumble, by character weight class (there is overlap so testing of all is unnecessary).
h. Hitbox/hurtbox images for all moves.
2. The best options characters possess to break strings and combos, based on positioning.
3. Frame traps pointed out, categorized and explained.
Sorry I was gone for so long, I'm gonna look over all these messages now...
Late as balls, thanks to a combination of my computer breaking, my laptop hdd usb adapter thing taking forever to ship, and laziness, but it's finally here. The Solo DThrow Chaingrab Percentage List.
The laptop's in the shop, I can access my old files again; things are looking up! For now it's just an initial list of percentages and ways to escape, but over time I'll evolve this into something special, starting with the walking chaingrab percentages (<3 Hylian). Maybe even a general chaingrabbing guide, but I don't want to get too ambitious at the moment. xP
CHAINGRABBING
These figures assume you start the Chaingrab at 0% with DThrow fresh (not in the stale move queue at all). Obviously enough, the higher damage the character has when you start the Chaingrab fresh, the less you can Chaingrab them (higher start + fresh throw = more knockback = escaping lower).
The percentages look all out of whack, but it makes sense when you consider each character has a different weight, fall speed, size, shape and animation, among other attributes.
Bowser - 95% [30T]
Captain Falcon - 98% [31T]
Charizard - 92% [29T]
Diddy Kong - 62% [18T]
Donkey Kong - 65% [19T]
Falco - 87% [27T]
Fox - 87% [27T]
Ganondorf - 81% [25T]
Ice Climbers - 65% [19T]
Ike - 65% [19T]
Ivysaur - 62% [18T]
Jigglypuff - 16% [3T]
King Dedede - 103% [33T]
Kirby - 34% [8T]
Link - 70% [21T]
Lucario - 31% [7T]
Lucas - 76% [23T]
Luigi - 59% [17T]
Mario - 65% [19T]
Marth - 31% [7T]
Meta Knight - 65% [19T]
Mr. Game & Watch - 57% [16T]
Ness - 54% [15T]
Olimar - 37% [9T]
Peach - 24% [5T]
Pikachu - 70% [21T]
Pit - 59% [17T]
R.O.B. - 48% [13T]
Samus - 31% [7T]
Sheik - 90% [28T]
Snake - 84% [26T]
Sonic - 57% [16T]
Squirtle - 68% [20T]
Toon Link - 48% [13T]
Wario - 73% [22T]
Wolf - 98% [31T]
Yoshi - 81% [25T]
Zelda - 40% [10T]
Zero Suit Samus - 59% [17T]
I'm not sure about Mr. Game & Watch, not because he's tricky or anything, but because while compiling this list I didn't have my original document on hand. All I know is that he was one throw less than either Ice Climbers or Zero Suit Samus. Of course, I'll update with the correct value when I've had a chance to look over the old document.
ESCAPING
As a general rule, characters should DI Up-Away to escape the DThrow Chaingrab faster (though a few exceptions exist; can't remember who they are, don't have it on hand etc, but DIing as above is still good). The moves listed here also help escape MUCH earlier than simply jumping out, usually because of invincibility.
Some moves aren't as useful to escape as you'd think. Generally a move has to either have invincibility starting at frame 1 (preferably with a hitbox at the end, ala Fox's or Wolf's Shines), or move you completely out of the way really fast (within a few frames).
Fox - Shine (DSpecial)
Jigglypuff - Rest (DSpecial) or Double Jump, DJ preferred
Lucas - Double Jump DAir
Marth - Dolphin Slash (USpecial) or Double Jump, DJ preferred
Ness - Double Jump DAir
Peach - Float (Hold Jump)
Snake - Grenade (NSpecial), won't get you out immediately though.
Wolf - Shine (DSpecial)
Yoshi - Reverse Double Jump (Hold Away)
Zero Suit Samus - Flip Jump (DSpecial)
If a character is not listed here, all they can do to escape is DI well and jump out.
TODO
1a. Walking chaingrab percentages.
1b. Regrab knockback ranges.
1c. Complete frame data.
2a. Video supplement.
2b. In-depth data for each character.
TL;DR
- Yes, these are correct.
- Yes, I double-checked them.
- Yes, I accounted for DI.
- Yes, I checked for moves to escape with. (I actually did a LOT of impromptu frame data for other characters because of this, gotta love having no immediate access to the internet.)
- Yes, sometimes you have to be dang-near frame perfect to pull off a regrab on some characters at higher percentages.
... I think that's everything. :/
Brawl Weight Modifiers
1.20 Bowser
1.16 Donkey Kong
1.13 Snake
1.12 King Dedede
1.10 Charizard
1.09 Ganondorf
1.08 Samus
1.07 Wario / Yoshi
1.06 R.O.B.
1.05 Ike
1.04 Captain Falcon / Link
1.02 Wolf
1.00 Ivysaur / Lucario
0.98 Mario
0.97 Luigi
0.95 Sonic
0.94 Lucas / Ness / Pit
0.93 Diddy Kong
0.92 Ice Climbers / Toon Link
0.90 Peach
0.87 Marth
0.85 Sheik / Zelda
0.82 Falco / Olimar
0.81 Zero Suit Samus
0.80 Fox
0.79 Metaknight / Pikachu
0.78 Kirby
0.75 Game & Watch / Squirtle
0.68 Jigglypuff
Some notes from me on throw data:Magus gave me my first research project and the rest is history: Throw information said:
You should probably find the release frames and 1st actionable frames for weight dependant throws against Ivysaur/Lucario as they have a 1x multiplier (weight of 100 in PSA). Then knowing that you should be able to freely calculate it being used on any of the other characters if you wanted.
This is my thread that includes throw frame data for Melee:
http://www.smashboards.com/showthread.php?t=206469
The actual slowdown/speedup effect on the animation's progression is a result of: 100/Weight
So with a weight value of 120 (Bowser), the animation will progress by 0.8333... for every 1 in-game frame that passes. If an event occurs on 10.000 of the animation, it would instead happen on frame 12 in the game.
The animation would progress like so:
0.833, 1.667. 2.500, 3.333, 4.167, 5.000, 5.833, 6.667, 7.500, 8.333, 9.167, 10.000 (event)
You can shortcut the entire process by simply taking the frame the event takes place on and multiplying it by their Weight/100 (Bowser would be a 1.200 multiplier), and then rounding UP to the next frame.
So something taking place on 14.000 of the animation would happen on frame 17 for Bowser
(14 x 1.2 -> 16.8 -> 17)
Example:
***A throw releases on frame 14.000 of the animation and you can shield starting on frame 32.000 (Marth's Melee f-throw).***
When used on Lucario the animation would progress like so:
1.000, 2.000, 3.000, 4.000, ...[snip]... 13.000, 14.000 (release), 15.000, 16.000, ...[snip]... 31.000, 32.000 (1st actionable frame)
Resulting in 18 frames of lag (frames 14 through 31 inclusive)
Which could be easily calculated with:
(32 x 1.00 -> 32) - (14 x 1.00 -> 14) = 18
32 - 14 = 18
When that same throw is used on Bowser it would go:
0.833, 1.667. 2.500, 3.333, 4.167, ...[snip]... 12.500, 13.333, 14.167 (release), 15.000, 15.833, ...[snip]... 30.833, 31.667, 32.500 (1st actionable frame)
Resulting in 22 frames of lag
Calculation:
(32 x 1.20 -> 38.4 -> 39) - (14 x 1.20 -> 16.8 -> 17) = 22
39 - 17 = 22
•the hitstun on all throws is roughly consistent across the cast at 0% (~same number of frames of stun). As would be expected, the hitstun growth rate varies by character.
•Throws with internal damage (ie, more than one hit) have a set amount of lag after the release point (I calculate the "lag" based on the frame the last hits damage is applied on... if you need clarification talk to me on AIM. This is the best and easiest way to find lag on throws).
•Throws without internally applied damage (single hit) have lag that varies (∂W) across the cast. Testing these is a nightmare and as such I'm only going to be testing a few characters single hit throws.
example of preliminary throw frame data test results from an earlier B+ set: kirby selected arbitrarily but I've done all of the characters at least once.
Code:
Kirby f-throw [62-45=17]* b-throw [46-33=13]* up-throw [80-51=29]* d-throw [90-60=30]
HS@0 Lag Adv @0 HS@0 Lag Adv @0 HS@0 Lag Adv @0 HS@0 Lag Adv @0
DK 32 19 13 42 15 27 41 33 8 41 30 11
Snake 32 20 12 42 14 28 41 33 8 41 30 11
Bowser 32 19 13 42 15 27 41 32 9 41 30 11
DDD 32 19 13 42 15 27 41 32 9 41 30 11
Zard 32 19 13 42 14 28 41 31 10 41 30 11
Ganon 32 18 14 42 15 27 41 32 9 41 30 11
Samus 32 18 14 42 14 28 41 31 10 41 30 11
Wario 32 18 14 42 14 28 41 31 10 41 30 11
Yoshi 32 18 14 42 14 28 41 31 10 41 30 11
Rob 32 18 14 42 14 28 41 30 11 41 30 11
Ike 32 18 14 42 14 28 41 30 11 41 30 11
CF 32 18 14 42 13 29 41 30 11 41 30 11
Link 32 18 14 42 13 29 41 30 11 41 30 11
Wolf 32 18 14 42 13 29 41 29 12 41 30 11
Ivy 32 17 15 42 13 29 41 29 12 41 30 11
Lucario 32 17 15 42 13 29 41 29 12 41 30 11
Mario 32 16 16 42 13 29 41 29 12 41 30 11
Luigi 32 17 15 42 12 30 41 28 13 41 30 11
sonic 32 16 16 42 12 30 41 27 14 41 30 11
Lucas 32 16 16 42 12 30 41 28 13 41 30 11
Ness 32 16 16 42 12 30 41 28 13 41 30 11
Pit 32 16 16 42 12 30 41 28 13 41 30 11
dK 32 16 16 42 12 30 41 27 14 41 30 11
IC 32 16 16 42 12 30 41 27 14 41 30 11
TL 32 16 16 42 12 30 41 27 14 41 30 11
Peach 32 15 17 42 12 30 41 26 15 41 30 11
Marth 32 14 18 42 12 30 41 25 16 41 30 11
Sheik 32 14 18 42 11 31 41 24 17 41 30 11
Zelda 32 14 18 42 11 31 41 24 17 41 30 11
Falco 32 14 18 42 10 32 41 24 17 41 30 11
Olimar 32 14 18 42 10 32 41 24 17 41 30 11
ZSS 32 14 18 42 11 31 41 23 18 41 30 11
Fox 32 14 18 42 10 32 41 23 18 41 30 11
MK 32 13 19 42 10 32 41 23 18 41 30 11
Pikachu 32 13 19 42 10 32 41 23 18 41 30 11
Kirby 32 13 19 42 10 32 41 23 18 41 30 11
GW 32 13 19 42 10 32 41 21 20 41 30 11
Squirt 32 13 19 42 10 32 41 21 20 41 30 11
JP 32 12 20 42 9 33 41 20 21 41 30 11
The most important thing you will ever learn about frame advance data mining:
Veril said:OK, time to share something kinda boring but really useful.
Counting frames is incredibly tedious and a waste of time. If you're collecting data that requires frame-advance rather than PSA (say you're finding shield or hitstun data) its a lot easier if you give a frame value to the clock timer. What I like to do to make things easiest for me is to have the initial input frames for whatever I'm testing occur on the x:00 mark.
So I made this little chart for easy conversions:
Ex. lets say I'm finding throw data.Code:Time Frame x:00 0 98 1 48 31 96 2 46 32 95 3 45 33 93 4 43 34 91 5 41 35 90 6 40 36 88 7 38 37 86 8 36 38 85 9 35 39 83 10 33 40 81 11 31 41 80 12 30 42 78 13 28 43 76 14 26 44 75 15 25 45 73 16 23 46 71 17 21 47 70 18 20 48 68 19 18 49 66 20 16 50 65 21 15 51 63 22 13 52 61 23 11 53 60 24 10 54 58 25 8 55 56 26 6 56 55 27 5 57 53 28 3 58 51 29 1 59 50 30 x+1:00 60
I'll grab Lucario (with his weight mod of 0) at the desired % with the character in question. Let the timer run down to the xx:00 mark (seconds:milliseconds), and input the throw on that frame. I than have a point of reference and can really easily gather the rest of the data by buffering shield to determine the IASA, alternating jump and tap-jump to determine the escape frame, watching the damage marker to determine release frame... and I can play with DI or teching etc etc etc... you see how this is way easier than counting frames when you're doing close to a thousand tests minimum.
How to find stuff in frame advance
Grounded IASA: buffer shield
Aerial IASA or escape frame: alternate jump and tap jump
With finding the hitstun escape frame its best to make sure the opponent is thrown offstage, otherwise you'll find the "first techable frame" which you may also want depending on the throw.
If you are gathering hitstun data:
-hitstun = Escape frame - Release frame
-keep in mind its % and character dependent
-most characters suffer equal hitstun from a move at 0%.
-moves with a high bkb to KBG ratio will have more consistent hitstun values across the cast and %.
If you are gathering ONLY advantage data on a throw:
-Advantage on a non-teched throw= Escape frame (thrown character) - First actionable frame.
-keep in mind that weight dependent throws will yield different FAFs based on character thrown.
obviously Advantage also = (hitstun + release frame) - First actionable frame.
moving that data around from my initial dump thread...
ICs visible hibubble .pac: http://smash.everalert.net/downloads/Fit...e_v0.1.pac