In almost any speedrun, there are many different strategies that can be used. Runners will often have to choose between spending their time practising a few harder levels or lots of easy levels, and this choice will often affect how they do in runs.
My question is: what would you think to people using computers and algorithms to help them choose what to practise? Would that be seen as fair or not?
What about using a program during to run to recommend which strats to go for on each level? (For example calculating if you can afford to go for a risky glitch or should take the slow but easy route).
An argument in favour of these sort of programs would be that they are not affecting how you play - even if they predict that you can beat the world record with these strats, you still need the skill to play well and consistently.
Am argument against them would be that they are a tool that is assisting you in your speedrun (and therefore banned) as you are using them to get better times in your runs.
What do you think?
No but as a concept, would it be allowed? (Whether or not it would be a good idea is a different question lol)
Unless the program has a way to automatically detect everything happening in the game (which would require quite the specific knowledge about each particular game), you'd need to input info into it somehow, which is going to lose you time over just processing and making the decision on your own.
The closest thing to this I can think of that works is something like the map layout viewer runners use for Digimon World 2. Dungeons have a set of static layouts it can choose from for each floor, but the chosen layout varies from run to run. The map viewer doesn't know which layout was chosen on your run, but by exploring a bit you can go ruling out other layouts until you narrow it down to one, at which point the map viewer then transposes into something akin to what the OP is mentioning.
Again, you do lose time inputting the info into the program (i.e. cycling through the different layouts in the program to find the one you need) when you could just memorize all the possible layouts and do all the guess-work mentally, so I personally do not see any issues with this idea.
I guess I was thinking of something like livesplit that keeps track of your time (very fast and easy to use) - and kind of like how it has a PB chance feature on livesplit it predicts how long the run would take using different strats
For example telling the runner if it's likely to be faster to do a risky strat or a safe strat based on their consistency.
Livesplit doesn't tell you if you should go for a risky strat or not. Livesplit just gives you information about your performance in the current run, and then you process it and make a decision as to whether or not to go for it.
In that sense, I don't see Livesplit giving you an unfair advantage, but rather just providing you with info you could have tracked yourself, so I'd see no problem.
PS: Just to clarify, we all obviously still use Livesplit because it's terribly convenient, I'm just saying that you could track that info yourself if you really wanted to, so it's still fair game.
Yeah that's true - I feel like I haven't explained what I'm thinking of very well but I see that people do often use programs to help them during their speedruns.
(I was thinking of a tool that would do what you were talking about in the top paragraph - collecting the data and then also saying if you should go for a strat or not.
Livesplit was just an example of how the data would be collected.)
I know it's a weird idea and probably not one that will be very useful lol, but you never know... it might turn out that computers are better than people at deciding what the best strats are
[quote]it might turn out that computers are better than people at deciding what the best strats are[/quote] usually the best strats are the fastest ones :D
Not always - the fastest strats are often much too hard / risky to go for if you aren't consistent enough
Are you then implying this algorithm you suggest would know about your consistency for each skip/glitch? How would you go about setting up the parameters for such a thing?
I'm not sure to be honest lol
It wouldn't be too hard to collect data on how long you took for each split (livesplit already does this), so as long as you told it what strat you did it could build up a lot of data about how long different strats take
From there it's just lots of calculations and computer stuff to predict how long it will take you next time, based on your past consistency
That certainly sounds like an interesting concept, though you'd need a sample of runs with enough size to avoid getting skewed results; otherwise one failed attempt would make the program never recommend that strat ever again.
In fact, in cases where you have a safe strat vs a risky strat and the risky strat ends up going lower than the safe strat, it'd effectively stagnate there, since the safe strat would consistently yield the same average results and always be faster than the sub-par risky strat average (and you'd never have a chance to improve the risky strat's time/speed average since you'd never do it).
Now, you could argue this is a good thing, but then you'd never get practice for the risky strat since well, it'd never get recommended and you'd never do it. Of course, you could practice it outside of runs, but then you run into the following two situations: A. The information from practice sessions is not fed to the program/algorithm. In this case, the program would be unaware of any improvements you've made during your practice sessions, meaning that inevitably you'd have to "disobey" the algorithm at some point to start using the risky strat again (and once you've improved upon your PBs for long enough without changing strats, I guarantee you'll want to do so at some point), at which point the program would've been made redundant in a sense. B. You manage to input this practice info into the program for it to take into account and change the strat's "score" to reflect any improvements. This poses the question: is the information from practice sessions really comparable to real runs? Would there be some sort of penalty multiplier to make it less impactful?
I think that the notion of a program that can perfectly predict a strat's speed based just on your past runs is a bit flawed in the sense that it's working on a dynamic object of study which is, well... you.
We as humans don't have a fixed "skill level", but rather develop through (mostly) trial and error, and unless the program can factor in all (or most of) the variables that come into that process it'll most likely come short one way or another.
Yeah true, chances are all the factors which affect how you do in a speedrun would make it impossible to model accurately - a person would be better at deciding which strats they think they would get.
That's also a good point that you would end up in a feedback loop - being inconsistent at a strat causes you to use it less, meaning you get less consistent and less of a chance to 'prove' that it can be useful.