My prediction for DDP vs Chimaev

[Luffy]

Luffy you posted the code and good on you, proved you were sincere

But you haven't posted arguably the most important thing - the average result of this simulation.

The simulation actually gives many outcomes with probabilities attached to it. I've picked the one whose pattern appear the most, that is, the outcome that shows up more consistently in all the possible scenarios, which you can read in the OP. Ofc the OP was embellished with my own writing, it needs a level of creativity to imagine how each round would go, but I described the most consistent pattern run over by the simulations in the OP, which was posted... In the OP lulz

 

[FlowchartRog]

The simulation actually gives many outcomes with probabilities attached to it. I've picked the one whose pattern appear the most, that is, the outcome that shows up more consistently in all the possible scenarios, which you can read in the OP. Ofc the OP was embellished with my own writing, it needs a level of creativity to imagine how each round would go, but I described the most consistent pattern run over by the simulations in the OP, which was posted... In the OP lulz

I see. I didn't realise it was actually based on your simulation

Why were you banned btw? Obviously you're an odd guy, and you say whacky stuff and make loads of threads. However, evidently, you're not a troll, and you seem alright.

 

[Luffy]

I see. I didn't realise it was actually based on your simulation

Why were you banned btw? Obviously you're an odd guy, and you say whacky stuff and make loads of threads. However, evidently, you're not a troll, and you seem alright.

Dunno, many ppl started to ask "when will this dude be banned" so I think I just got overwhelmed and banned. I've made some troll comments that got me yellow card, like, I took Topuria's comment of "the Dagestani fighters hide in Dagestan to avoid PED tests, they are steroid machines" and posted it tho I was more like trolling (half trolling lol) and got a double yellow card. I even said to the mod on DM "well if that warrants a warn, then why isn't 80% of the forum with yellow cards? Am I the only one who can't say not proven shit to some fighters". The mod just sent me a and didn't even explain. Then when I posted a thread about Illia Topuria saying JJ was the goat and didn't need to fight anymore, I think the JJ hate cult group got all on their edge and yeah I think was banned off that. Heck, I had made a detailed, very detailed post about "JJ vs Stipe, a closer look", which was as serious as it could, it got moved to the troll section, ppl called be a troll... Even tho it took me 2 days to finish it as it was highly technical...

 

[Luffy]

You make DDP sound like Jason Voorhees ><

Not gonna lie, I still lol when I read this comment, didn't realize how it sounded graphic when I was thinking on the wording to describe the fight

 

[Luffy]

Well it might be cleaner than doing it manually, but is it faster? And it's no cleaner than actual vectors in C, which are clean and very fast. Arrays have boxing/unboxing. Compared to vectors which have direct data access, you're essentially doing 2 memory lookups.

For you this might be fine doing offline simulations etc, but for realtime applications speed is crucial.

For example there is a library in C# called Linq, that has some of the operators you've shown, but under the hood it's slow and generates a lot of garbage, so is far from ideal for real world use.

Have you tried benchmarking the a * b operator compared to manual loops/calculations?

Admittedly the code is certainly trim and clean with those kind of operators replacing loops, if it could compile to machine code that was just as efficient.

Untyped variables fucking suck though

I had missed this post. Those are fair point but a couple of things get lost...

... NumPy isn't boxed Python lists. Under the hood you're already in C/Fortran-style contiguous memory... a * b calls a vectorised when written in C that walks the buffer once. So you're not paying two Python-level look ups per element — Python just hands over the pointer, the loop happens in compiled code, and it hands a pointer back.

Showing a speed check (quick micro):

N = 10 000 000
a = np.random.rand(N)
b = np.random.rand(N)

t0 = time.perf_counter()
c2 = [a * b for I in range(N)] #
pure-Python loop
print("python loop:", time.perf_counter() - t0)

On my laptop I get ~0.07 s for NumPy vs ~3.8s for the naive loop. Hand-rolled C with -O3 will still win by a bit but the gap is small enough that I'd rather write one line than 30 and re-compile every tweak...


When it's still too slow in case, numba —> JIT your loop and keep Python syntax

cython/C++ —> drop to native like you said.

About GPU, CuPy or PyTorch tensor ops. Same idea, prototype in Python and hot-path in whatever meets the SLA.

Linq chains keep deferred iterations in manager memory, so yeah you can create garbage. NumPy doesn't because a * b allocates once. Apples and pineapples haha

And the untyped vars, add import numpy as np; a:np.ndarray and run mypy. Static analysis, auto complete, the works. Optional typing means dev speed first, guardrails when you need them.

For a Monte Carlo that chews through 10-20 M samples offline, NumPy's plenty... If I ever need sub-millisedond live inference, I'd shove the kernel into C/Numba and just keep the rest in Python glue..... Best of both words.

 

[Luffy]

The simulation actually gives many outcomes with probabilities attached to it. I've picked the one whose pattern appear the most, that is, the outcome that shows up more consistently in all the possible scenarios, which you can read in the OP. Ofc the OP was embellished with my own writing, it needs a level of creativity to imagine how each round would go, but I described the most consistent pattern run over by the simulations in the OP, which was posted... In the OP lulz

Think of it as two layers that I stack on top of each other:

Hard numbers from the sim. The Monte-Carlo tells me who is most likely to win each round and how (control-heavy vs strike-heavy, early sub windows, cardio drop-off etc). From that I get a "skeleton" like:

R1 —> 79 % Khamzat (control > damage)
R2-R5 —> 71-85 % DDP (damage > control)



And then I add the narrative skin creative writing with the "flesh" that skeleton out with a plausible sequence of events that matches those probabilities and matches what the fighters actually do on tape. For instance, Khamzat’s R1 dive —> lines up with his real Holland / Usman entries. DDP’s body work leg kicks —> lines up with Whittaker / Strickland II footage. Cardio swing after five minutes —> documented versus Burns/Usman. The round by round writing is “creative” in the sense that I pick specific strikes and scrambles to illustrate the trend, but the beat leans on something that the sim says is most common and what the fighters have already shown in real fights.

If the numbers ever flipped, let's say the sim suddenly gave Khamzat a 60 % edge in the third round because of a new gas-tank metric, then I’d rewrite the narrative to fit that new structure. The writing isn’t a totally random fanfic, it’s an illustrative storyboard of the modal path that the model spits out with ofc a level of creativity for immersion that I'd add since the mod isn't going to spit everything out...

 

[FlowchartRog]

I had missed this post. Those are fair point but a couple of things get lost...

... NumPy isn't boxed Python lists. Under the hood you're already in C/Fortran-style contiguous memory... a * b calls a vectorised when written in C that walks the buffer once. So you're not paying two Python-level look ups per element — Python just hands over the pointer, the loop happens in compiled code, and it hands a pointer back.

Showing a speed check (quick micro):

N = 10 000 000
a = np.random.rand(N)
b = np.random.rand(N)

t0 = time.perf_counter()
c2 = [a * b for I in range(N)] #
pure-Python loop
print("python loop:", time.perf_counter() - t0)

On my laptop I get ~0.07 s for NumPy vs ~3.8s for the naive loop. Hand-rolled C with -O3 will still win by a bit but the gap is small enough that I'd rather write one line than 30 and re-compile every tweak...


When it's still too slow in case, numba —> JIT your loop and keep Python syntax

cython/C++ —> drop to native like you said.

About GPU, CuPy or PyTorch tensor ops. Same idea, prototype in Python and hot-path in whatever meets the SLA.

Linq chains keep deferred iterations in manager memory, so yeah you can create garbage. NumPy doesn't because a * b allocates once. Apples and pineapples haha

And the untyped vars, add import numpy as np; a:np.ndarray and run mypy. Static analysis, auto complete, the works. Optional typing means dev speed first, guardrails when you need them.

For a Monte Carlo that chews through 10-20 M samples offline, NumPy's plenty... If I ever need sub-millisedond live inference, I'd shove the kernel into C/Numba and just keep the rest in Python glue..... Best of both words.

Impressive, NumPty is legit

Couple of questions for you Luffy.

Did you learn Python/Statistics as part of your chemical engineering, or is this a side interest of yours? Also, why Python?

I ask partly as I'm not acquainted with Pythion, so don't really know its target uses.

I do agree with your overall point though. The work I do is realtime and performance focused, however you sometimes have to weigh up time/legibility/performance otherwise we'd all be writing Assembly

Most CPUs are fast in the modern era, what I think is equally important as optimised calculations is minimising GC. At runtime, I make sure pretty much all calculations are stack rather than heap, try to eliminate adding to the heap at runtime, and use as much object pooling as possible etc.

 

[contrail]

Too bad we never got to see Chimaev against top talent before covid nuked his body. This is a competitive fight but DDP is a tall order for someone who is prone to fading. We'll see.

 

[Cowboy Kurt Angle]

I did this with Max and Dustin I did a tale study on the last 5 striker vs striker matchups that they did

I determined that Max was probably gonna win 49-46 or even get a late round 4 stoppage, Max just looked way faster then Dustin and Dustin had definitely slowed down in the fights that I watched

 

[Dacha]

I just hope Shara Bullet prediction doesn't come through.

Thats all.

 

[Wicked_1]

The problem is that guy's like Till and Brunson have had dominant positions on DDP, relatively easily . They are not even close to Chimaev in mma wrestling. Now Chimaev is training totally different than when he had issues with cardio and has a top guy coaching him on cardio now. Everyone has this idea that because DDP lasted to get into the deep waters against no one that tried to sub him instead of striking on the ground. Chimaev will dominate DDP on the ground .
Can DDP stop the takedown? No Can he get up..he even says he won't try to but either way he can't so how's his defense against top level subs? Well everyone will find out this weekend. And The New Undisputed Middleweight Champion of the world winner by submission at 3:35 of the 1st round..yeah it's not going to be close .
I like DDP and know what he brings in toughness and digging deep. He has been so tired and still found the way to finish or keep throwing those looping punches. I also know that he gets hit often and has been taken down by people who shouldn't be able to take him down. They weren't Burns or Usman . And now at 185 Chimaev will have better cardio and strength.
DDP is made to beat dominant strikers with his odd jerky style and iron chin DDPs coach called Chimaev a 1 dimensional offensive wrestler . That's about as far off as any idiot could believe. Wrestlers aren't using subs. Or many of the takedowns he uses. Wrestlers will tell you that training with wrestlers as DDP is doing in camp, does not prepare you for his mma wrestling .
He held his own against a Olympic wrestler that just got done at Olympics before training with Chimaev. That's an incredible thing for an mma wrestler as the 2 tiers are nowhere near each other in levels. Olympic wrestling is top notch . He was handling the guy and that just shouldn't be possible so easily DDP is not remotely close to Olympic level . Trying to defend against that style of wrestling is not an effective technique against what Chimaev does . I know everyone thinks DDP will survive and Chimaev will gas and he will take advantage. It won't happen

 

[Luffy]

The problem is that guy's like Till and Brunson have had dominant positions on DDP, relatively easily . They are not even close to Chimaev in mma wrestling. Now Chimaev is training totally different than when he had issues with cardio and has a top guy coaching him on cardio now. Everyone has this idea that because DDP lasted to get into the deep waters against no one that tried to sub him instead of striking on the ground. Chimaev will dominate DDP on the ground .
Can DDP stop the takedown? No Can he get up..he even says he won't try to but either way he can't so how's his defense against top level subs? Well everyone will find out this weekend. And The New Undisputed Middleweight Champion of the world winner by submission at 3:35 of the 1st round..yeah it's not going to be close .
I like DDP and know what he brings in toughness and digging deep. He has been so tired and still found the way to finish or keep throwing those looping punches. I also know that he gets hit often and has been taken down by people who shouldn't be able to take him down. They weren't Burns or Usman . And now at 185 Chimaev will have better cardio and strength.
DDP is made to beat dominant strikers with his odd jerky style and iron chin DDPs coach called Chimaev a 1 dimensional offensive wrestler . That's about as far off as any idiot could believe. Wrestlers aren't using subs. Or many of the takedowns he uses. Wrestlers will tell you that training with wrestlers as DDP is doing in camp, does not prepare you for his mma wrestling .
He held his own against a Olympic wrestler that just got done at Olympics before training with Chimaev. That's an incredible thing for an mma wrestler as the 2 tiers are nowhere near each other in levels. Olympic wrestling is top notch . He was handling the guy and that just shouldn't be possible so easily DDP is not remotely close to Olympic level . Trying to defend against that style of wrestling is not an effective technique against what Chimaev does . I know everyone thinks DDP will survive and Chimaev will gas and he will take advantage. It won't happen

The problem though is that DDP's recover has a high % and is repeatable, not just isolated... He breaks wrist control quickly, turns head position, builds base instead of giving up guard... It's not a matter of "can DDP be taken down?" as much as it is "can he be held and threatened a finish before Khamzat can no longer sustain the same pressure?"... That's way tougher imo

DDP has literally said the opposite in multiple camps "I'll wrestle Khamzat". And it's shown that he does stand, wall walks, control hip posture and usually punishes on breaks... His best thing isn't his TDD but denying rides and finishes... Maybe the cardio has gotten better, but new cardio coach is not a claim that can be proven until we see how Khamzat deals with round 4/5 while having to wrestle heavily. Chimaev has never won a five rounds in UFC... Whenever he went to the third round, it shows the consistent R1 big pressure and spike, R2-3 fading more. I can't give him a 25 minute pressure cardio until it's proven. I'm not saying his wrestling is not amazing, but he finishes often in rides + back control instead of a heavy ground and pound pressure, for example... And if you look at DDP while on bottom position, he has odd stances , frames and he repeatedly builds wall which makes that finishing style less of a sure thing than you think, specially in a round...

Yeah DDP wasn't that serious with the one dimensional wrestler, no one thinks Khamzat is one dimensional still tho, he's far more proven as an offensive grappler than a five round control fighter who keeps digging and winning minutes... Plus DDP constantly targets body and legs to tax opponents' cardio... if he lands those, which he does generally, a bunch of body and legs strikes, as the fight goes on to round 2 and 3 Khamzat's wrestling attempts drop in speed and quality, and DDP starts imo to take over, as he is proven in deep waters and extended fights.

 

[AMAZINGUFC]

My prediction for DDP vs Chimaev

There is some bad blood between the two, but they still decide to touch gloves before referee Herb Dean.

Chimaev reaches out with a front kick, and he backs Du Plessis up against the cage.

Chimaev walks down and loads up on a bomb of right hand.
That shot right down the middle detonates on the side of DDP head, sending the weird guy down and out.

As HD is sprinting to stop the fight, “Borz” gets off a few more shots that are completely unnecessary.

Wow.
What a wild first round knockout!

 

[jeskola]

I'd definitely have chose another username nowadays. Luffy was a random pick in 2017 since I couldn't think of whatever username as I'm terrible with coming up with new usernames lol. But since I'm known as Luffy by everyone here no way I'm changing that now even tho I wish haha

Yeah, wouldn't want to ruin your stellar reputation of shit posting you horrible, long-winded takes in new threads that could easily be replies in existing threads

 

[Luffy]

Yeah, wouldn't want to ruin your stellar reputation of shit posting you horrible, long-winded takes in new threads that could easily be replies in existing threads

Oh someone has me in their head. I'm sorry if I hurt you bro, I don't even remember talking to you nor your username. I said I wouldn't change precisely because it's already how I'm known here. Go get some fresh air and vent some frustration you have to a psychiatrist idk... I don't remember hurting you, I don't even know you

 

[Sensee]

Here we go again, another Luffy gem. What on earth is he gonna tell us this time

I humbly donate this to you as a reward:

 

[Luffy]

I humbly donate this to you as a reward:

What are those cat stuff? @Luthien also posts Luffy :cat emoticon:

I don't get it lol

 

[Luthien]

What are those cat stuff? @Luthien also posts Luffy :cat emoticon:

I don't get it lol

The Shercat party never stops

 

[Luffy]

The Shercat party never stops

What does it mean to be a cat? Lol

 

[Luthien]

What does it mean to be a cat? Lol

Being a cat means spending 80% of life sleeping, 15% plotting the downfall of the dog, and 5% sprinting around the house at 3 a.m. for reasons we will never explain