its valid but its just another way of saying dont play when tired, avoid late hours, avoid randoming, etc
I see Concede's point, but probability rules usually apply to a relatively big pool in order to make predictions. I consider 2k games a significant pool in this case, while i still sceptically ask if the idea is valid.
Well you can try it, but the best way by far is to find a hero that you are really good with and do all games the same thing, for me the best hero ( though i think i am really bad with it ) is templar, i dont know but everytime i play templar is ranked its usually a free win.
I play tinker in ranked too but i do it for fun, i am REALLY bad with tinker so ....
Again, the point here isnt really me trying to improve MMR at all costs, its only about the viability of the idea itself, if matematics can be applied here in this case, like I stated.
Assume that you randomed in all the 2k games, with all wr's at 50%, and considering that different matches have different players and that I play consistently at this particular level, it is valid to use statistics and probability to enhance MMR?
I am not saying or implyng that i am magically locked in 50% wr. It's only about the viability of the idea. If both unranked and ranked matches have a 50% at 2k games at a particular point, can you use probability calculus to, say, try to loose more games in unranked matches and win more in ranked in the immediate future?
that only works if they both are based on modifying the same rating, which they don't, so no.
losing a game in ranked makes it easier for you at your current level to win a subsequent game in ranked since your rating gets lower. this does not apply to unranked. 50% is simply a byproduct of this system.
Speaking only from the point of the probability theory, you can go by the same principle as betting on the "red" and the "black". Each consecutive loss has statistically lower chance of appearing. So, if you do something like "I'll only go ranked after 5+ loss streak", assuming it won't affect you mentally, the proportion of your wins in ranked most likely to increase.
However, keep in mind, that this theory is extremely vague and only reliably works when used on big numbers, and by big numbers I mean at least millions.
Probability of having five consecutive loses and then having one win is exactly the same as having six consecutive loses.
let's say you have 50% win chance (which you don't)
then you can say losing 2 games in a row has 25% chance (0.5 x 0.5)
BUT after your first loss the next game has obviously 50% chance of winning again
what you said is like claiming after getting 10 heads in a row you have 99% chance of getting tails
i'm pretty sure the last line of your post wasn't there before the edit, but maybe I just misread it.
I only edited the grammar.
If you don't get what I'm talking about:
-Probability of getting "heads" is 50%;
-Average amount of "heads" among 10 tosses is 5;
-If you will toss the coin in a group of 10 times and will only bet on "tails" after getting at least 5 "heads" in a row for the first 5 times, statistically your chances of winning will increase.
i don't think it will lol
they are mutually exclusive events (i don't know if this is how you call it in english)
dude if you play 5 heads your 6th toss won't have more than 50% chance of being tails, this makes no sense at all.
maybe you're thinking about the monty hall problem? (which is a different issue)
I aced statistics, and im pretty sure it doesnt work that way... you seem to be committing the gamblers fallacy...
And that's why I say it's not about what you think. It won't make sense if you try to apply it here and now. It's a principle of BIG numbers.
For each toss, the probability of getting either "heads" or "tails" is 50%, that's why in said group average amount of "heads" will be five. This doesn't means you can't throw 6, 7 or 10 "heads" in a row, but after a BIG number of iterations, groups with 6 or more "heads" will have less appearances => this means, the chance of getting it is lower => if you will only bet in a situation where you can ensure the actual chance of you losing is lowered - you'll win more frequent.
I kinda get it now, why they don't teach it in school.
which is just another fancy way of saying "git gud scrub"
you're better off increasing your chances of winning (improving) than playing at the roulette with 50% chance each game
it doesn't work that way, any stats major will tell you that. .. it is fallacious to reduce the law of large numbers to a limited sample. what you have done is textbook gambler's fallacy through a reduction of the law of large numbers.
I fell like a engineering student like me should know all this math stuff, but i have no fucking idea what you guys are talking about.
I showed my stats prof this and he laughed
what do you not understand about how the law of large numbers not being applicable to individual samples
LMAO did u show him the cesspool that is dotabuff forums, or did u just c+p the individual posts without telling him it's from a dota forum
What do you not understand about me stating it "not being applicable to individual samples" in the very first post and repeating it one more time in the latest?
I'm imagining Zenoth showing his professor this and the professor goes like "dude you play Dota? add me my nick is ^^DragonFist^^"
Your first post was correct. But
-If you will toss the coin in a group of 10 times and will only bet on "tails" after getting at least 5 "heads" in a row for the first 5 times, statistically your chances of winning will increase.
this part is fundamentally wrong
just think of it, for all sets of 10 which start with 5 heads, the average heads in that set will simply be 7.5. You have not increased your chances at all
somehow you think its equivalent to your first post
I think like epsi-kun. No matter how insignificant it is, the chances of you getting a heads after 5 tails really increase. That is taught in school, at least for me it was.
The particular tool i wanted to start from was something like conditional probability:
http://en.wikipedia.org/wiki/Probability#Conditional_probability
However, in the case of dota is not that linear. There are a great deals of factors that should be taken into account. Your mates, your opponents, picks..
The question arised when i felt, and saw many of my mates saying the this: A great number of times, it seems like i am loosing more because of some1 doing stuff i consider wrong in my team, or they have an exceptional players, etc. A equally great number of times it seems like the opposite and i win. It is rare a game where i actually play really bad. Even after watching replays and discussing it with some friends.
Now this is not me giving excuses. But i think we can all agree, dota is a team game, and as such, winning or loosing doesnt depend only of ourselves, but the others. It depends a lot from external things.
Now considering that the chances are even, couldnt we possibly use some form of "advanced" conditional probability to counter the chance factor and increase MMR?
I think we can, just like epsik-kun said in his coin example. We should be able to do this to some extend. However, this seems highly controversial..
About the increased skill of your opponents, i initially didnt though it should invalidate the results, since the skill of your allies will also increase. However Zenoth has a pretty solid point: Using this, ranked matches' difficulty will change, but unranked should stay all the same. So, does this spoil the conditional probability? Or can we assume that since our allies skill also increases or decreases along that of the opponents, we can consider it all the same?
even given multiple group tosses what I just said its true
by looking only at the set of all groups starting with five heads, you have already skewed the probability such that the average number of heads is no longer 5. In this set it would be 7.5, and your statement would not hold.
conditional probability here is only applicable when you are observing multiple independent events
what you are looking at here is "There is no principle that a small number of observations will coincide with the expected value or that a streak of one value will immediately be "balanced" by the others. See the Gambler's fallacy."
This makes sense, but also given the fact that initial idea was to "have groups with 5+ heads", making it "5 in a row" was just to simplify it. You can go for any combination given you've got these heads, even if it's only for the last toss. But I do agree with you.
I would just like to say that if Relentless was here this whole thing would already be soved.
I think Zenoth nailed it.. my idea really was constantly checking the last 20 games to calculate the conditional probability for the next ones. The likelyhood would chance according to the sample size i considered, which would make any number insignificant.
If i considered the 5 throwns as losses, it chance of winning would be high. But if I increased the sample, and hte previous 10 were wins, the likelyhood suddently would be me loosing.
But that is an argument against the sample size... which means i must consider everything i have... and the greater the sample size the closer it would get to 50%. I guess it would reach a point were i couldnt really guess anything, since i would be so close to that rate. Although it seems it can be fundamentally right, if i assumed a constant chance of 50%. However, with the minimum flutuations dota implies it probably isnt possible to reach any significant profitable result i guess...
Any additional thoughs are welcome, but i would already like to specially thank epsik-kun and zenoth for the paciance and actual arguments in this discussion, with logical proof.
guys im pro at statistics too and it might seem that it wont work BUT it does... just read what i posted a bit mroe carefully! consider this case: you find out you always play tide after winning twice with void but that tide game always ends up in a loss! lets say we can attribute a 90% chance to lose with that tide! so based on game history he could decide to not play tide but play some other shit (any other hero) where his chance to win is 60%. it makes a difference and statistically finding defects in your that lead to certain results and avoiding them will lead to a 50% win (in case of no data) or whatever history says the result will be! though abusing this will skewer the result towers your real skill!!!!!!! BUT your real skill by use of this method will potentially be higher depending on how you use the system.
lets say i have 10 sites where i drill for oil! fuck i drilled 5 times in each place but i only got result in 4 of them! (ofc not each time) they all had initially the same chance for containing oil but once i discovered that some of them had and gae enough searches on the others, it would be a waste to keep wasting millions on searching for oil on the 6 ones that didnt give me result WHEN im looking for FAST results and INSTANT money
blabla
Моля, впишете се, за да публикувате коментар.
Hi
Recently I had this idea, and I need help to figure out if this makes sence.
Let's say, i have 2k games, and my wr is currently 50%. My wr never went above 52% or below 48%, which makes me believe my results are consistent.
What I want, is to use this statistics in order to raise my RMM. By considering my last 20 game results, i will play in ranked matches when my 20game-wr is relatively low and play unranked matches when it is relatively high, in order to increase my rmm.
Can this be done? Is this valid? What do you think?
Ty in advance