Also (not sure if this was commented on already), you probably forgot +1.25 sec min / +2.5 sec max conversion time granted by the Teuton team bonus which makes the graphs at 12:07 slightly confusing.
Oh, yes, didn't catch that when editing. Those graphs should be moved 1 second to the right, and should have an extra column on the end to account for the 2 second increase of the max time.
@@Ellanion that's not how computer lag works though, anyway with your sample size the difference cant be with 26% cant be dude to statistical inaccuracy , so more to investigate there
Around 15:00 I started typing a comment "what if a monk is converting something and swaps to a scout or other unit that resists conversion?" and you delivered shortly after. I've never seen such a complete explanation, picking everything, even including teutons and faith. Excelent work dude!
I didn't want to fail No Nut November, but then you had to upload this overly analytical and complex monk guide which I will never use, of a game I don't even play, I just couldn't hold it. Was worth it, 10/10 Monk Seconds. Make more of it.
@@devilpistons1269 In spite of what Monty Python would have you believe, the Spanish Inquisition didn't just barge in unexpectedly, they made their presence known when they came to town, and there was always a grace period during which people could voluntarily confess and repent.
Finally seeing what those experiments were about!! Great video!! So this means that, in the x9 tech mod... A Spanish player with 9 Inquisitions researched has minimum time of -5 monk seconds and conversions guaranteed after 1 second (i.e., guaranteed instant conversions). That sounds pretty cool xD
As soon as you broke down the conversion resistance values i was thinking "how the hell did they manage to increase it just by exactly 50% for Faith?" Turns out they just didn't 😂 But it's not like the AoE2 tooltip hasn't been wrong before right?
@@CrnaStrela and most of the percentages displayed in the civ bonuses were innacurate (like saying Japanese infantry attacked 25% faster when in reality it was 33%) ^^
This is an incredible video - really insightful and funny. I don't even want to think about how long it took to put together this video but I really appreciate it!
I am thouroghly impressed with this video, really awesome in depth explaination and such a high effort to always have something awesome to look at. Seriously that 5 second scene where the seasons change while the monk converts forever, must have taken hours to prepare and edit. Amazing video. You have got yourself a new fan
I just did a simulation test on your 26.6% percent find with 22200 samples, and got some interesting results. I used your described method to generate the random percentage. With RAND_MAX = 32767 we have a probability of 0.0222 (i.e. 2.22%) of finding conversion chance above 26.6% with 22200 samples. In some scientific circles a p-value of 0.05 is enough to refute this hypothesis. Running the same code with different RAND_MAX values gives some interesting results: with RAND_MAX = 255 (which is a random byte) we get probability 0.0749 (7.49%) with RAND_MAX = 127 (which is a random signed byte ignoring the sign) we get probability 0.4419 (44.19%) with RAND_MAX = 100 we get probability 0.0021 (0.21%) We are doing multiple tests here based on the same 26.6% data so to be statistically safe we should only propose a new hypothesis RAND_MAX = 127 instead of concluding that it must be so. To back up this argument, I want to say that if your 26.6% value would be precisely the conversion chance, then we would expect probability exactly 0.5 (50%). Because of statistical error the 26.6% will likely be slightly different from the conversion chance, but it should be close. Hence, RAND_MAX=127 being so close to the 50% does suggest it might be the true value. A seperate test would be needed to deny or confirm it.
I don't understand whats so great about this game. Monks are so hard to micro and use in general. The campaigns are sometimes very wrong historically. The early game takes way too long. I wish I had never played this beautiful gem! All the micro is so intense and I love eco management! Keep up the great work!
I don't think this does solve the problem. Given your sample size, we should be pretty surprised if you got a number of 26.6%, given the true number is 26%. A rough and ready calculation of the statistical error is sqrt(11k)/11k ~ 0.01. So I would expect an error on 26% of about 0.25. So this is a little over two standard deviations from the mean. That makes your result statistically significant. Two solutions-- go Bayesian (but what would our priors be?) or accumulate more data to really test this. I think you need to run a bigger experiment.
26.5 makes sense and .6 would be off by 1, or possibly correct. The reason is the 0 counts, not just the 1 thru 25, and also if using float random, it would add the .5% due to how it chooses to round (default or using a function). No need to use math, as he just did the simulation, but his isn't the only one who basically got 26% as he showed.
So I did some math, all values from 0 to 8519 lead to values between 0-25. 8519 is actually 25.99 so very close. So in theory, you should find a 26% success rate. I suspect the random number generator isn't the best. What strikes out as me is that a max of 2^15 isn't common, so it is possible they're taking the low bits of the libc 32 bit generator (which aren't always the best), or they made up their own with shitty values. You are also potentially running into patterns in the rng output because you have too many monks and you're basically running a significant portion of the period of the rng. The obvious solution is disassembling the game to find out what they use for rng.
@@meneldal What? F for not explaining properly. Where are you getting 8519 number? The reason is because in programming it isn't just 1 to 25, it's 0 to 25, 0 counts sometimes whereas in real life, typically if you said 'theres 25 spaces' you wouldn't think there's a 0 space (26 spaces). Since it's a float value with rounding, the .5 or .6 makes sense as it will round ontop of that, so thus 26.5% or 25.6% is valid. Not sure what 'math' you are doing.
@@BlackOps78321 I'm talking about the values from the random number generator. They are between 0 and 32767 as explained in the video. Among these values, all numbers up to 8519 (a total of 8520 possibilities) lead to a percentage below 25%. That turns out to be almost exactly 26% of the possible outputs from the rng.
for being about "some dumb stuff" most players wont ever give a damn about, this video has some really sweet editing and presentation. well done my dude
So correct me if I'm wrong, but monks won't reset even if they move to chase a target, as long as they aren't tasked to do something else or stop chasing because they lose the target. That means: 1) if in a fight a monk is chasing a running target, you can let it chase until another target is in range and swap only then, using the progress gained. 2) if in a large fight another unit of mine is providing LOS of a target which ran away from my monk further than the monks LOS, I can let the monk keep chasing and once he catches up (or someone else comes in range) use that charged up timer. It's probably not gonna come up often, but it seems like keeping LOS of a unit under conversion could make your monk more relevant in the middle of a fight. I always thought the moment they started walking to chase, progress was lost!
I'm not exactly sure if it's the LoS or a fixed distance from the Monk. I've seen units leave the Monks line of sight for a short time while the Monk keeps chasing them. I know that as long as the Monk keeps chasing the conversion continues, I don't know what exactly causes them to stop chasing.
This is insane. I can already see pros tasking monks to convert buildings for the instant conversions on units upon retargeting. This can easily become meta
@@martijnbouman8874 yep, it's already used in pro play, starting to convert a building and switching to an incoming knight or mangonel (if further tech is considered).
Units becoming potentially impossible to convert is a great example of why not to use hardcoded magic values. Setting the threshold to 1000 sounds like something one of my students would come up with...
yes. What they *should* have done is to have the algorithm immediately return "true" if the maximum conversion duration is reached without even calculating any further. That avoids the issue.
There seems to be about a 2-3% chance that you would get 26.6% or higher in 22000 tests if the true chance is 26%, so it's technically statistically significant but not too unreasonable that it was just a random event. On the other hand, there is about a 99.9999% chance that you wouldn't get 26.6% if the true chance was 25%
There were far more than 22000 samples. Of the 22000 units, 5940 were converted as soon as possible, 4306 took 2 attempts, 3242 took 3 attempts etc. That gives 5940 + 4306*2 + 3242*3 + 2272*4 + 1711*5 + 1217*6 + 3512*6 = 70295 samples. I'm too lazy to figure out how big the probability of finding 26.6% if the true value is 26%, but I'd expect it to be lower than 0.5%. Note: last conversion second was guaranteed to succeed and is thus not a sample, so the 3512 also counts for 6 samples each.
@@unknownusername9335 First of all, you'd need to calculate what percentage of the conversions actually went through, but assuming a geometric distribution that's a good point.
Fantastic work! I'm the maker of Medieval RPG map and I've got to experiment a lot with all those resources you mentioned (conversion resistance, min and max adjustments, etc) without really knowing what was being affected. Now I think I know what to change in order to get the results I expected. My thanks.
Not only did I learn about the strange world of monk math, I also learned that there is a copy/paste feature to the scenario editor hidden in the terrain menu.
Citation: T-West, K. (2020). How Monks Really Work: The Truth Behind the Wololo. V. 3. RUclips. 17:30 min. Some notes: 1. Last I checked (viewing the recording in Sony Vegas, in September), monk second in DE was exactly 1.25 sec. On release (and up until SotL video on the topic), it should have been close to ~1.1-1.13 sec (as -my own speculation- the interval of a monk-second was not modified to correspond to the "normal" game speed of DE). 2. 7:50 (targets running away from the Monk's range) - Did a small test in the Scenario Editor. It seems that Monk will continue to chase his "prey" unless the latter runs away so that the distance between the two becomes 18 tiles (not accurate; might be 17 or 19 tiles instead). If there is no LoS on the victim, Monk will stop in its tracks; however, if the target remains within the LoS (provided by something else), the Monk will (unless facing an obstacle) transition into sort of 'patrol' either resuming the conversion (and saving the timer) or stopping in its tracks (for whatever reason) when the original target re-appears within the Monk's range (9 tiles). 3. I did a much smaller sample size (100 targets * 50 test runs, only first monk-second counts) test back in the day (DE) and ended up with a 26.36% (1318 out of 5000) chance to convert. 0% accuracy resulted into several test runs with 1-3 conversions, and 99% accuracy always ended up with all the units being converted. Not sure if it can be deemed significant, but with 75% accuracy and 24 * 100 conversion attempts, the average chance to convert was 74.1%.
Your sample size is n = 22,200. Using p = 0.26, your standard error is sqrt( p * (1-p)/n ) or about 0.029. The difference between 26% and 26.6% is thus right on the border of being statistically significant, using a two-sided test at the conventional threshhold.
have you tried adding a wait before the monk start converting, if the number generator is randomized from the time the game starts then the randomness would be compromised by the fact that in all examples the test has started at 0 game seconds
The random number generator's seed is determined by the system time at the start of the game. The 10 trials of 1,000 Monks each had different results, so there isn't an issue of the same seed and sequence of numbers being used for each trial. Coincidentally, I did use a 10 second wait at the beginning of these trials, but that was because not all of the information the CA team needed for their analysis was included at the very beginning of the rec.
At some point the AI just went idle and didn't continue converting. I guess that's just what happens when the AI script is "Immobile Units AI Gold". :)
Hello and thanks you for your amazing work ! Quick question, at 12:10 in the video: should the averages shown in green be in secondes instead of percent ?
1) Incredible work, I've learned more about monks ever after playing for 2 decades 2) It's not Two-tins, its Two-Tahns. Everyone says it correctly with the adjective "Teutonic". Just say Teutonic without the ic
two-tins does make sense in regards to the English language though, as it has a strong tendency to realize vowels in unstressed syllables as a "schwa" (the "i"-like sound in "tins"). So I don't mind the "two-tins" pronunciation. On the other hand, in "teutonic" the "ton" is the stressed syllable and as such, the vowel is fully realitzed as an "o".
I always wanted to know this. I had some suspicions of how it worked, but now I have finally a good explaination. I didn´t know that faith could make scout conversions take over 10 seconds. 18
I was so confused when you said there are no instant conversions, because I've definitely had them before. I guess I had accidentally targeted a building or a different unit and switched.
"more than you ever wanted to know about monk conversions" no sir...just enough...I have always been frustrated as to why my monks never seemed to convert quickly where my opponents working on multiple conversions seemed to get theirs faster. Now it makes sense. And using the code helped me as well as i enjoy working in python.
Wait now I have a question: if you delete a unit or building getting converted, do the minimum 4 monk secs still count if they were to change target or do they get reset?
0:40 Red Monk = Persian (Zorastrian or Islamic) Green Monk = American Orange Monk , South East Asian (Buddhist) Purple Monk = Middle East (Christian/Islamic/Jewish) Grey = European Monk [Germanic(Anglo-Saxon/Norse) or Christian or Islamic] Yellow Monk =East Asian (Chieneese/Korean/Japanese/Buddhist) Cyan Monk = Central Asian Monk (Tengrist) Blue Monk = North Indian (Hindu/Islamic/Zorastrian)
Always cool to see all the detailed mechanics like this! One thing I noted, in the logic you show for conversion % it explains up to 26.26% since the '100' value is very low likelihood of being selected it's basically the case that 26 options out of 99 succeed -> 26.26% conversion chance. Which is getting quite close to the 26.6 you measured. My best guess on the last .3% is a combo of sampling error and the rng having some imperfection which maybe messes with the draw rate on certain values?
If you add the values for 0 through 25 from the table at around 5:05, then divide that sum by RAND_MAX, you end up with a number very close to 26% (26.0... if I remember correctly). I don't know if the rand function would be "bad" enough to be likely to produce values that far away from 26% at such a large sample. It's implemented as a simple linear congruential generator, seeded by system time at the start of the game. I took a quick look at it, and nothing jumped out at me that would produce this big of a difference.
So can you use a group of monks with theocrathy on a building for more than 10 monk seconds, and switch them on a group of unit for multiple instant conversion ?
How precise is that pseudo-code? It wouldn't surprise me if the extra .6% in your conversion chance is explained by some kind of off-by-one error with conversion resistance.
It's what we came up with when I asked someone who is experienced at working with the exe file. We don't have the original source code, which unfortunately isn't open source. It wouldn't surprise me if there's something else that's missing.
Maybe this question is a little too arcane...but I have a bunch of fun playing with the 256x mod against Extreme AI. I've wondered exactly how certain technologies relating to conversion work when stacked. (256x is actually a really good magnifying glass to see how a tech can work without going into game files or anything!) I can see that Faith works more when it is stacked, but how does it play against a coded limit in the conversion time once you get it enough? Can you create a unit that basically can't be converted? On the opposite side, can Inquisition get you to an effective 0-second guaranteed conversion time?
Inquisition can bring the min and max times down to 0, but that's counteracted by Faith increasing the times. Faith can increase the conversion resistance up to 3*256 in that mod, but by that point increasing the min time probably has the greatest effect at limiting conversions.
That was awesome, I definetly LOVE your stuff ! BTW you just said that teh minimum building convert time was greater than the maximum unit convert time, but what is it ? How long of a timeframe do I have to switch my monk's target ? I would die laughing if I manage to pull that out in ranked.
Oh, I've never actually looked at that unit's stats, since he was replaced by the ranged Berserk in AoC. I'll have to dig into the AoK data at some point.
Also (not sure if this was commented on already), you probably forgot +1.25 sec min / +2.5 sec max conversion time granted by the Teuton team bonus which makes the graphs at 12:07 slightly confusing.
Oh, yes, didn't catch that when editing. Those graphs should be moved 1 second to the right, and should have an extra column on the end to account for the 2 second increase of the max time.
@@T-West , often times, you have a cataphract when you said that you're gonna lead us to something, are you the cataphract?
@@emilyding2496 No I just like Cataphracts.
@@T-West , okay, then I NEED TO ASK 2 THINGS.
1, which version do you like the best? HD one, or DE?
2. Are you one of those Byzantine fans?
@@T-West , SOTL likes Byzantines cus he loved AoE1. What about you.
I wish there was a Video where T West and SotL work together to solve a mystery about AoE
Better yet it’s about a mystery we don’t even know exists!
have you ever seen the two in the same room?
@@steinanderson I have the same theory
"Armed with 22,200 samples"...
Jesus dude.
I'm not sure he was among them, but he would probably have helped.
What? That's how science works
@@Ellanion that's not how computer lag works though, anyway with your sample size the difference cant be with 26% cant be dude to statistical inaccuracy , so more to investigate there
@@wansichen3743 lag takes no factor in the calculation since it is the % of them converted after 4 monk seconds, therefore it is not measuring time
@@EcclesiastesLiker-py5ts comment of the day 🤙
I didn't believe that scouts took longer to convert than Teutons.
But you talked to me for more than 10 seconds.
This took me way longer to get than it should have. :P
What a boring joke
*10 seconds later
Hahahaha u funny
XD
@@schmid1.079 maybe u have a min_get_it time 11
You mean 10 monk seconds*
This is the most holysome monk video on the internet
Hasaan
Good dad joke
So you're telling me that the good padre can get warmed up literally *talking to a wall* and then become the most suave orator in the realm? o.O
Around 15:00 I started typing a comment "what if a monk is converting something and swaps to a scout or other unit that resists conversion?" and you delivered shortly after. I've never seen such a complete explanation, picking everything, even including teutons and faith.
Excelent work dude!
Boring statistics and probabilities...
WOLOLOOO
...
...
...
...
Very insightful!
This comment is gold
I didn't want to fail No Nut November, but then you had to upload this overly analytical and complex monk guide which I will never use, of a game I don't even play, I just couldn't hold it. Was worth it, 10/10 Monk Seconds. Make more of it.
Couldn't have said it better myself, I just failed first day of september after watching this.
12:42 "If we have a scout with faith, then we have a laughably high 11 conversion resistance"
Well played
Wow.
...
Also 14:21 "The other technology that affects conversion time is *expectedly* the Spanish unique technology, Inquisition.
Is 11 the laughing taunt, I forgot been a while
@@arya.n.8252 1
@@devilpistons1269 In spite of what Monty Python would have you believe, the Spanish Inquisition didn't just barge in unexpectedly, they made their presence known when they came to town, and there was always a grace period during which people could voluntarily confess and repent.
The Daut on the monastery is pure art. Great video!
I came for AOE2 stuff
What I got was a Python case study
I'm not disappointed
Also, I saw what you did to that monastery 1111
5:54 "might not of used a large enough sample size"
LudaKRIS size map it is then...
Finally seeing what those experiments were about!! Great video!!
So this means that, in the x9 tech mod... A Spanish player with 9 Inquisitions researched has minimum time of -5 monk seconds and conversions guaranteed after 1 second (i.e., guaranteed instant conversions). That sounds pretty cool xD
Oh that's a good question about Inquisition, will need to experiment with that.
This do run the risk of negative number rollaround depending on the type used to handle the numbers
The video we needed, but didn't deserve, T-West continues to deliver!! Wolololo!
I've sometimes played with thought, that some day devs would tell us what really is happening behind the codes and mechanics, and save lots of time.
You know what I waited from HD?
Different monk skins for each region
I miss those from WololoKingdoms.
@@T-West I missed them too, those were awesome
I wish that were a thing, especially since we have a couple alternate monk skins for scenarios. It's not too late AoE2 dev team!
@@annaairahala9462 Yeah, I mean, the kings have their own different skin for each region
As soon as you broke down the conversion resistance values i was thinking "how the hell did they manage to increase it just by exactly 50% for Faith?"
Turns out they just didn't 😂
But it's not like the AoE2 tooltip hasn't been wrong before right?
Reminder Fervor didn't work the entire time during AoK and AoC
@@CrnaStrela and most of the percentages displayed in the civ bonuses were innacurate (like saying Japanese infantry attacked 25% faster when in reality it was 33%) ^^
@Emperor AlHasan that's why i used the past tense "were" rather than "are". Fervor is fixed too!
This is an incredible video - really insightful and funny. I don't even want to think about how long it took to put together this video but I really appreciate it!
You are a crazy man. Going to watch right now!
No regrets :D great information!
I am thouroghly impressed with this video, really awesome in depth explaination and such a high effort to always have something awesome to look at. Seriously that 5 second scene where the seasons change while the monk converts forever, must have taken hours to prepare and edit. Amazing video. You have got yourself a new fan
Spirit of the Law gonna be out of a job!
His time is over, embrace the new emperor.
Turns out a mathematician cannot compete with a programmer
Omg i thought this was spirit of the law. He sounds the same
@@satyakisil9711 Why not both?
@@Jokie155 There can only be one.
I just did a simulation test on your 26.6% percent find with 22200 samples, and got some interesting results. I used your described method to generate the random percentage. With RAND_MAX = 32767 we have a probability of 0.0222 (i.e. 2.22%) of finding conversion chance above 26.6% with 22200 samples. In some scientific circles a p-value of 0.05 is enough to refute this hypothesis.
Running the same code with different RAND_MAX values gives some interesting results:
with RAND_MAX = 255 (which is a random byte) we get probability 0.0749 (7.49%)
with RAND_MAX = 127 (which is a random signed byte ignoring the sign) we get probability 0.4419 (44.19%)
with RAND_MAX = 100 we get probability 0.0021 (0.21%)
We are doing multiple tests here based on the same 26.6% data so to be statistically safe we should only propose a new hypothesis RAND_MAX = 127 instead of concluding that it must be so. To back up this argument, I want to say that if your 26.6% value would be precisely the conversion chance, then we would expect probability exactly 0.5 (50%). Because of statistical error the 26.6% will likely be slightly different from the conversion chance, but it should be close. Hence, RAND_MAX=127 being so close to the 50% does suggest it might be the true value. A seperate test would be needed to deny or confirm it.
SOTL and TWest are secretly the same person.
I don't understand whats so great about this game. Monks are so hard to micro and use in general. The campaigns are sometimes very wrong historically. The early game takes way too long. I wish I had never played this beautiful gem! All the micro is so intense and I love eco management! Keep up the great work!
Awesome video with great analytics! Happy to see the parser I made was of use!
The video editing is incredible, dude. The code part was awesome and very well explained. Good work.
I don't think this does solve the problem. Given your sample size, we should be pretty surprised if you got a number of 26.6%, given the true number is 26%. A rough and ready calculation of the statistical error is sqrt(11k)/11k ~ 0.01. So I would expect an error on 26% of about 0.25. So this is a little over two standard deviations from the mean. That makes your result statistically significant.
Two solutions-- go Bayesian (but what would our priors be?) or accumulate more data to really test this. I think you need to run a bigger experiment.
26.5 makes sense and .6 would be off by 1, or possibly correct. The reason is the 0 counts, not just the 1 thru 25, and also if using float random, it would add the .5% due to how it chooses to round (default or using a function). No need to use math, as he just did the simulation, but his isn't the only one who basically got 26% as he showed.
So I did some math, all values from 0 to 8519 lead to values between 0-25. 8519 is actually 25.99 so very close. So in theory, you should find a 26% success rate.
I suspect the random number generator isn't the best. What strikes out as me is that a max of 2^15 isn't common, so it is possible they're taking the low bits of the libc 32 bit generator (which aren't always the best), or they made up their own with shitty values.
You are also potentially running into patterns in the rng output because you have too many monks and you're basically running a significant portion of the period of the rng.
The obvious solution is disassembling the game to find out what they use for rng.
@@meneldal What? F for not explaining properly.
Where are you getting 8519 number?
The reason is because in programming it isn't just 1 to 25, it's 0 to 25, 0 counts sometimes whereas in real life, typically if you said 'theres 25 spaces' you wouldn't think there's a 0 space (26 spaces). Since it's a float value with rounding, the .5 or .6 makes sense as it will round ontop of that, so thus 26.5% or 25.6% is valid.
Not sure what 'math' you are doing.
@@BlackOps78321 I'm talking about the values from the random number generator. They are between 0 and 32767 as explained in the video. Among these values, all numbers up to 8519 (a total of 8520 possibilities) lead to a percentage below 25%. That turns out to be almost exactly 26% of the possible outputs from the rng.
or a third one: ask Matt Pritchard about it
for being about "some dumb stuff" most players wont ever give a damn about, this video has some really sweet editing and presentation. well done my dude
So correct me if I'm wrong, but monks won't reset even if they move to chase a target, as long as they aren't tasked to do something else or stop chasing because they lose the target. That means:
1) if in a fight a monk is chasing a running target, you can let it chase until another target is in range and swap only then, using the progress gained.
2) if in a large fight another unit of mine is providing LOS of a target which ran away from my monk further than the monks LOS, I can let the monk keep chasing and once he catches up (or someone else comes in range) use that charged up timer.
It's probably not gonna come up often, but it seems like keeping LOS of a unit under conversion could make your monk more relevant in the middle of a fight. I always thought the moment they started walking to chase, progress was lost!
I'm not exactly sure if it's the LoS or a fixed distance from the Monk. I've seen units leave the Monks line of sight for a short time while the Monk keeps chasing them. I know that as long as the Monk keeps chasing the conversion continues, I don't know what exactly causes them to stop chasing.
@@T-West ah fair enough. Either way it's kind of a big thing to know that I'm better off letting them chase. Thanks!
If I had the power to grant you a PhD for this video, I certainly would. You're now Dr. T-West in my heart.
At least I have the power to grant this comment a heart!
I just woke up from a weird ass dream where my monks all got converted and i rage quit myself awake just to have this in my 2nd recommendation slot..
great job man, hope you can include more SOTL style videos!
Building charging just changed my life
You kept teasing us with thousands of monks on twitter and I’m not disappointed with the final results.
3:16 The image and sound i feel before waking up from a nightmare.
This is insane. I can already see pros tasking monks to convert buildings for the instant conversions on units upon retargeting. This can easily become meta
They already knew about this particular mechanic.
@@martijnbouman8874 yep, it's already used in pro play, starting to convert a building and switching to an incoming knight or mangonel (if further tech is considered).
"Say a prayer to Lord Daut" - Wololo, please dear Lord, teach us the ways of the castle :)
Units becoming potentially impossible to convert is a great example of why not to use hardcoded magic values. Setting the threshold to 1000 sounds like something one of my students would come up with...
yes. What they *should* have done is to have the algorithm immediately return "true" if the maximum conversion duration is reached without even calculating any further. That avoids the issue.
Last time I was this early my Thanksgiving turkey was still in the oven...wait.
Don't forget about it, you don't want 100 Food going to waste!
11 likes for both comments 11
13:45 YO IS THAT DAUT ON THE SIDE OF THE MONASTERY!?
Indeed. All hail Lord DauT!
@@T-West I ALMOST SHIT MYSELF LAUGHING SO HARD lol If I wasn't watching on my big monitor I never would have noticed. nicely done!
@@TomsLife9 It's a mod :)
oh wow! i would never have noticed had you not pointed that out. im getting that mod now
I am now a monk expert, thanks Travis
The moment you whipped out a Python library is the moment I knew this was not going to be amateur hour. You've won a new convert!
There seems to be about a 2-3% chance that you would get 26.6% or higher in 22000 tests if the true chance is 26%, so it's technically statistically significant but not too unreasonable that it was just a random event. On the other hand, there is about a 99.9999% chance that you wouldn't get 26.6% if the true chance was 25%
There were far more than 22000 samples. Of the 22000 units, 5940 were converted as soon as possible, 4306 took 2 attempts, 3242 took 3 attempts etc. That gives 5940 + 4306*2 + 3242*3 + 2272*4 + 1711*5 + 1217*6 + 3512*6 = 70295 samples.
I'm too lazy to figure out how big the probability of finding 26.6% if the true value is 26%, but I'd expect it to be lower than 0.5%.
Note: last conversion second was guaranteed to succeed and is thus not a sample, so the 3512 also counts for 6 samples each.
@@unknownusername9335 First of all, you'd need to calculate what percentage of the conversions actually went through, but assuming a geometric distribution that's a good point.
@@unknownusername9335 Yes, at 4:09 that's how I arrive at 26.6%, using all of the "conversion attempts" during seconds 4 through 9.
@@T-West I assumed your 26.6% is correct, my reply was to @Iwer Sonsch's 22000 samples.
You and Spirit of the law should team up for the ultimate AoE explained channel...
Fantastic work! I'm the maker of Medieval RPG map and I've got to experiment a lot with all those resources you mentioned (conversion resistance, min and max adjustments, etc) without really knowing what was being affected. Now I think I know what to change in order to get the results I expected. My thanks.
Not only did I learn about the strange world of monk math, I also learned that there is a copy/paste feature to the scenario editor hidden in the terrain menu.
"We all love monks."
Persian mains: REEEEEEE
damn, you're analysing monks, I'm about to lure my first boar
Spirit of the conversions :0
Hey @SpiritOfTheLaw... A new challenger approaches...
Impressive TWest... Bravo for the analysis and the effort! 👏
Of ALL THE INFORMATION IN THE UNIVERSE I find myself learning about this right now.
Finally, you've been teasing about this video for ages man!
From the Dark Age to the Imperial Age!
Citation: T-West, K. (2020). How Monks Really Work: The Truth Behind the Wololo. V. 3. RUclips. 17:30 min.
Some notes:
1. Last I checked (viewing the recording in Sony Vegas, in September), monk second in DE was exactly 1.25 sec. On release (and up until SotL video on the topic), it should have been close to ~1.1-1.13 sec (as -my own speculation- the interval of a monk-second was not modified to correspond to the "normal" game speed of DE).
2. 7:50 (targets running away from the Monk's range) - Did a small test in the Scenario Editor. It seems that Monk will continue to chase his "prey" unless the latter runs away so that the distance between the two becomes 18 tiles (not accurate; might be 17 or 19 tiles instead). If there is no LoS on the victim, Monk will stop in its tracks; however, if the target remains within the LoS (provided by something else), the Monk will (unless facing an obstacle) transition into sort of 'patrol' either resuming the conversion (and saving the timer) or stopping in its tracks (for whatever reason) when the original target re-appears within the Monk's range (9 tiles).
3. I did a much smaller sample size (100 targets * 50 test runs, only first monk-second counts) test back in the day (DE) and ended up with a 26.36% (1318 out of 5000) chance to convert. 0% accuracy resulted into several test runs with 1-3 conversions, and 99% accuracy always ended up with all the units being converted. Not sure if it can be deemed significant, but with 75% accuracy and 24 * 100 conversion attempts, the average chance to convert was 74.1%.
Your sample size is n = 22,200. Using p = 0.26, your standard error is sqrt( p * (1-p)/n ) or about 0.029.
The difference between 26% and 26.6% is thus right on the border of being statistically significant, using a two-sided test at the conventional threshhold.
Honestly really well thought out research and a great range of scenarios. Very informative
the trick of monks converting buildings and insta gib units on a switcheroo... wow, i didnt know thaat
This video clearly needs an update, I hope T-West does it very soon
have you tried adding a wait before the monk start converting, if the number generator is randomized from the time the game starts then the randomness would be compromised by the fact that in all examples the test has started at 0 game seconds
The random number generator's seed is determined by the system time at the start of the game. The 10 trials of 1,000 Monks each had different results, so there isn't an issue of the same seed and sequence of numbers being used for each trial.
Coincidentally, I did use a 10 second wait at the beginning of these trials, but that was because not all of the information the CA team needed for their analysis was included at the very beginning of the rec.
I thought Spirit of the law was the most insane, i've just found out there is a tough competition ! Good job =)
wait a minute, how did my jacket become blue all of a sudden?!
I was most interested in seeing how that monk free-for-all at the end went! Which team won!?
At some point the AI just went idle and didn't continue converting. I guess that's just what happens when the AI script is "Immobile Units AI Gold". :)
Hello and thanks you for your amazing work ! Quick question, at 12:10 in the video: should the averages shown in green be in secondes instead of percent ?
Oh, yes, those are seconds, not percents.
AoE2 content in glorious DE & 60fps form, what a time to be alive in
1) Incredible work, I've learned more about monks ever after playing for 2 decades
2) It's not Two-tins, its Two-Tahns. Everyone says it correctly with the adjective "Teutonic". Just say Teutonic without the ic
two-tins does make sense in regards to the English language though, as it has a strong tendency to realize vowels in unstressed syllables as a "schwa" (the "i"-like sound in "tins"). So I don't mind the "two-tins" pronunciation.
On the other hand, in "teutonic" the "ton" is the stressed syllable and as such, the vowel is fully realitzed as an "o".
Going for 22.200 sample size is just the type of commitment we need.
w o l o l o
yeeesh I've missed your videos, this must have taken a while to record xD, Thanks for the holy extraordinary truth behind the wololo!!
I always wanted to know this. I had some suspicions of how it worked, but now I have finally a good explaination. I didn´t know that faith could make scout conversions take over 10 seconds.
18
Prof. Dr. of A.o.E. T-West
Before this video:
-How Monks Really Work?
-They're saying Wololo and converting
After this video:
-How Monks Really Work?
-That's a long story...
So that's how that one Hussar solo'd 4 out of my 6 monks before being converted....
when there is a bug that the developer hope no one noticed until now
I was so confused when you said there are no instant conversions, because I've definitely had them before. I guess I had accidentally targeted a building or a different unit and switched.
The building tip sounds awesome... will I remember/ be able to use it in game? No probably not
"more than you ever wanted to know about monk conversions" no sir...just enough...I have always been frustrated as to why my monks never seemed to convert quickly where my opponents working on multiple conversions seemed to get theirs faster. Now it makes sense. And using the code helped me as well as i enjoy working in python.
Amazing video, beside from the obvious use in mp, this video really helps in scenario creation. Thank you
the 500 pop 4v4 lag problem? never heard of it. the infamous 7k wololo battle is where its at
Wait now I have a question: if you delete a unit or building getting converted, do the minimum 4 monk secs still count if they were to change target or do they get reset?
They get reset. Although I'm not sure if you can still switch the Monk's target during the time while the delete animation is still ongoing.
This guy is like super Spirit of the Law, and that's insane!
I like the idea of using pseudocode to explain technical detail
Excellent video, TWest! TY.
0:40
Red Monk = Persian (Zorastrian or Islamic)
Green Monk = American
Orange Monk , South East Asian (Buddhist)
Purple Monk = Middle East (Christian/Islamic/Jewish)
Grey = European Monk [Germanic(Anglo-Saxon/Norse) or Christian or Islamic]
Yellow Monk =East Asian (Chieneese/Korean/Japanese/Buddhist)
Cyan Monk = Central Asian Monk (Tengrist)
Blue Monk = North Indian (Hindu/Islamic/Zorastrian)
Always cool to see all the detailed mechanics like this! One thing I noted, in the logic you show for conversion % it explains up to 26.26% since the '100' value is very low likelihood of being selected it's basically the case that 26 options out of 99 succeed -> 26.26% conversion chance. Which is getting quite close to the 26.6 you measured. My best guess on the last .3% is a combo of sampling error and the rng having some imperfection which maybe messes with the draw rate on certain values?
If you add the values for 0 through 25 from the table at around 5:05, then divide that sum by RAND_MAX, you end up with a number very close to 26% (26.0... if I remember correctly).
I don't know if the rand function would be "bad" enough to be likely to produce values that far away from 26% at such a large sample. It's implemented as a simple linear congruential generator, seeded by system time at the start of the game. I took a quick look at it, and nothing jumped out at me that would produce this big of a difference.
Just came here to say, we need conversion ratios of a sicilian scout after first crusade tech with teuton ally
0:08 oh no not this cursed scenario again *shivers*
I loved the seasons and world changing during, "forever, and ever, and ever...".
ok T-West, at 3:46 i admit u r the best youtuber for aoe community. i do still love sotl and t90 😂 but for u mad respect 😎👍
So can you use a group of monks with theocrathy on a building for more than 10 monk seconds, and switch them on a group of unit for multiple instant conversion ?
Yes!
How precise is that pseudo-code? It wouldn't surprise me if the extra .6% in your conversion chance is explained by some kind of off-by-one error with conversion resistance.
It's what we came up with when I asked someone who is experienced at working with the exe file. We don't have the original source code, which unfortunately isn't open source. It wouldn't surprise me if there's something else that's missing.
You and spirit* of the law, you guys are crazy.. keep it up!
Maybe this question is a little too arcane...but I have a bunch of fun playing with the 256x mod against Extreme AI. I've wondered exactly how certain technologies relating to conversion work when stacked. (256x is actually a really good magnifying glass to see how a tech can work without going into game files or anything!) I can see that Faith works more when it is stacked, but how does it play against a coded limit in the conversion time once you get it enough? Can you create a unit that basically can't be converted? On the opposite side, can Inquisition get you to an effective 0-second guaranteed conversion time?
Inquisition can bring the min and max times down to 0, but that's counteracted by Faith increasing the times. Faith can increase the conversion resistance up to 3*256 in that mod, but by that point increasing the min time probably has the greatest effect at limiting conversions.
Omg this is so important!! I've been playing monks wrong my whole life 😅
I want to know what goes through your head before you decide to make video, idk what it is but I love it
idk what it is either...
This is truly the Lord's work
There really needs to be a little bit of monk AI. At least in skirmishes and campaigns.
AI opponents magically and perfectly micro them.
My head hurts . By far ur best content .
That was awesome, I definetly LOVE your stuff ! BTW you just said that teh minimum building convert time was greater than the maximum unit convert time, but what is it ? How long of a timeframe do I have to switch my monk's target ? I would die laughing if I manage to pull that out in ranked.
The building convert time is 15-25 seconds, see the values of the resources in the lower left around 15:05
@@T-West Oh thank you, I thought it was the ressource ID and not its value my bad 11... So I have 5s monk-time... I'll make it work 11
So how fast does Harold Haardraade from the original AOE2 convert? He converts much faster than other monks.
Oh, I've never actually looked at that unit's stats, since he was replaced by the ranged Berserk in AoC. I'll have to dig into the AoK data at some point.
I'm wondering years later with the Sicilians first crusade and the devotion tech, how does it take now?
"Spirit of the Law? Impossible! We saw you fall!"
"I am the Spirit of T-West, and I come to you now at the turn of the tide!"
Spirit of the West
So.... Where is the confidence interval of the probability at 4:07 ?