Speaking of Arduinos, I've always wondered how these guys made an N64 bot with an Arduino Uno. The N64's controller protocol is much more complicated then the NES/SNES/Genesis', so I never really bothered trying myself.
I just recently completed this exact thing. As long as you use inline assembly, it's totally within reach. The protocol happens at 1 MHz (sub-bit are sent at 1us intervals), and the Uno has a 16 MHz processor, so you get 16 cycles between sub-bits.
Since most instructions take 1 cycle (except for some branching or larger arithmetic instructions), you have "plenty" of time to do any work you need. Most of the assembly consists of wait loops to make sure the timing is okay.
(Doing it with interrupts is out of the question, the latency from receiving an external interrupts to actually running code is too great.)
"genetic" algorithms? I think that is much too resource-intensive to be doable in the near and in the far future...
Er, I don't see how. Perhaps you mean it'll take too long? I agree with that if we're using pure genetic algorithms on input, which is why I said you'd have to introduce learning for this to be feasible.
But the genetic algorithms themselves aren't very resource intensive. Here's a video I uploaded a while ago but never made public (made public now):
Link to video
It's shows the result of an extremely basic genetic algorithm. The one on the left is the original input, the one on the right was the genetic result.
This was quite fun to make because I ran Mupen64 in my process space so I could manipulate it by changing its variables programmatically (which itself was a fun process of finding their addresses). I think I had 8 or 32 Mupen instances running concurrently for that video.
The way it works is that an input sequence is evaluated against some fitness function (in this case it was distance from honeycomb summed across all frames, ergo getting there fastest maximized fitness). The multiple Mupen instances allowed me to evaluated multiple sequences at once (this is just optimization, not necessary for correctness).
The N most fit sequences in the population are chosen to survive, and "mate" by mixing their input sequences using genetic crossover with random mutations. Then this new population is tested and so on. Over (infinite) time, this maximizes fitness. Obviously we don't have infinite time, so it's up to practical concerns to decide when it's finished.
partyboy1a wrote:
Also: It cannot be possible to find the "best" solutions without human interaction... The least what is required is to tell the program when you "win", and this alone is a difficult task, and may even be subjective (just think of the glitched version of Super Mario World, which shows "The End" while playing a normal level). Just defining the winning condition is most likely only possible by disassembling each game, every single line of code...
Right, the automation is only for the task of satisfying a goal, the goal is still specified by external input (human).
And yes, this does require you understand what values certain memory addresses take on, but this generally isn't too difficult to find out. And it only has to be figured out once. (For the example above I only needed to know the location of the honeycomb and the location of Banjo.)
But like I said, the genetic algorithm just doesn't converge fast enough without learning, and if I'm going to do learning I'd rather take a more formal approach.
I've been investigating the use of automatic theorem provers for solving TASes lately, coincidentally.
Don't want to to give out a bunch of information because people will inevitably get their hopes up for no reason, but I've been writing a new emulator as a hobby project where each system has a uniform simulation interface. This lets you perform symbolic execution on the system, and I'm trying to leverage constraint solvers (namely Z3) and automatic theorem proving to make conjectures about the game and prove them, ultimately conjectures of the form "this is the fastest way to satisfy these constraints: ...". (The "normal" approach of concolic testing isn't tractable for anything but small inputs.)
I originally intended to try to automate TASing with genetic algorithms, but this intrusive apporach is much better, and can (in theory) give provably optimal solutions. The genetic algorithm approach was fun though, because half the task was leveraging existing emulators to be evaluation functions. But it was extremely difficult to produce good results without also introducing learning, so I reapproached it from a formal learning side.
This is just a hobby project for now though. Good work on leveraging SAT!
I agree this is important, I was thinking about it just recently.
In the most "extreme" case, imagine someone (no not me) decides they want to compete with TasVideos as the repository for tool-assisting. Would it be okay for them to download and re-host every keypress file on this site? Should any license we come up with allow that?
I also agree there seems to be an implicit air of "this belongs to this site, but we all know it's okay to use the input however we want". Is there a point where "however we want" is not okay?
I don't think the judging process is as simple. It can't be efficiently automated. I think only the division to popular and unpopular speedruns can be automated, everything else (even Stars) should be manual. So I wouldn't start drawing flowcharts for the process, especially with symmetry in mind! :)
Actually I intended to leave the entire thing manual, I agree with adelikat that it's easiest that way, at least for a while.
My only point is that if we're going to try to divide movies up a bit more, either we should have added only Vault and changed only Vault, or added all the categories these splits imply and changed them all. But we went halfway and watered things down.
This gives rise to this decision procedure, correct?
The symmetry of the axis graphic is reflected here, which is why I find this layout intuitive.
Sensibly, adelikat wants to give a place to optimal movies that aren't entertaining, and everyone agrees with this. I think the problem is only trying to add on the Vault category leads to imbalance, which is why people are suggesting additional categories to balance it out. So this is the desired layout from the original proposal (if I understand it correctly):
The problem is that people perceived this as ending up in practice as:
Because the categories seem diluted. I think this happens because Star is trying to take on some of the load Moon has from not having Playaround and Demo to help out. Really Star ought to have remained exactly the same subset of movies, and only movies that obsolete more entertaining runs (as well as previously Grued runs) would be reconsidered and moved to Vault. This would have only reduced the size of Standard movies and given Vault its place.
So that's why I think it may be worthwhile to tryin again but with all the organizational categories.
What I think adelikat wants to remind everyone though is that runs aren't being chopped off any more than they were in the past, and it's in fact the opposite. More runs will be accepted now because of Vault. Just compare the above decision procedures with what we used to have:
Nothing is being lost, except IMO the clarity of categories.
Thanks. My point is this should be ordered to guide the user to where they want to be, more easily visible from any page (interactive), and combinatorial. How can I join two of those categories together in my search? If I can, it's not immediately obvious how.
If I'm in the NES section and I want 2-player games, I either have to search for text on the page or go back to square one and start from the category choice, then search for NES. I'm very likely to just decide it's not worth the effort.
(Again I want to make it clear I understand I'm making suggestions without doing any of the work, and appreciate the people who actually do the work. I'd be willing to help, though currently all my coding freetime is gone for the next few weeks.)
Perhaps we should start a thread on improving site navigation to avoid overloading this thread.
Navigation-wise it might be useful (again with the check box idea) to restrict movie navigation by movie attribute. (By attribute I mean "Best Ending", "Heavy Glitch Abuse".)
This would be a column on the right-most half of the screen (different from the tiers). This space is already wasted anyway:
Just toggle the attributes to easily filter. I think the overall point is that navigation should be interactive and easily filterable. (Not that implementing this is trivial to do.)
The list should either be sorted by use-count or split-count. That is, most popular attributes to least popular, or most dividing to least dividing. I can't think of which one is better.
Also, today I moved every existing publication with a rating under 6.0 to the vault. I picked this number because the result set looked to be a good case of border-line vault material movies.
Just something to watch out for, but if the movie's final rating ever gets changed to be based on tier (for example, Vault movies get a final rating based more heavily on Technical Quality than Entertainment) then using the final rating to place by tier is divergent.
Assume for a moment we have the Vault category, where movies go if their final rating is under 6 (otherwise they stay where they are). If we do a scheme like final score depends on category, then when the movie is moved to Vault (because its final score is less than 6) its final score calculation will change to be more Technical based. This could be enough to move its final score back above 6, so it gets moved (if this is automated)/should be moved(for consistency) back out of Vault. But now its out of Vault so its Entertainment factor is more important, pushing the final score below 6, etc.
I like both the idea of moving between tiers automatically and changing final rating based on tier. The former just needs to use a score independent of that, like the pure Entertainment rating (which in both of our proposals is the only thing that determines if something goes in the "vault" category, as all other categories require entertainment).
You failed to explicitly mention what exactly you are trying to do. Issue (1) in my post refers to taking the number of ratings into account when converting the ratings to player points. Are you trying to change the way the average is calculated which incorporates the number of votes in such a way that the old player point formula can be used? Would the old player point formula still need adjustments?
I'm trying to change the way movie ratings are calculated to be more meaningfully determined and less ad hoc. This means factoring in sample size from the start, which in turn happens to affect your problem #1. So the answer is sure: the old formula could be used with the new movie rating system and your old formula suddenly automatically takes into account sample size. But I wasn't suggested it has to be either-or, you could improve both.
But factoring the sample size into both is unnecessary if it's already done in a previous step. You could (I don't know and would be willing to look) probably do the player points calculations on the intervals themselves, rank them by size, then use that ranking as an interpolation value from the lower bound to higher bound of the interval as the final score. Players with many higher-confidence scores would be rewarded.
Baxter wrote:
Also, the average is currently calculated I think by giving the entertainment rating a weight of 2/3 and the technical rating a weight of 1/3. Did you account for this? These ratings can be very different for certain movies; everyone agreeing that the movie is quite a technical achievement, but completely disagreeing on the entertainment value. This also makes you wonder how good the assumption of a normal distribution is.
I clearly specified I was considering only one column for exposition. You would calculate the confidence interval for both columns separately and then combine them in some fashion (i.e. could be the same as you just described), which I also said I was holding off from discussing because the teirs system is still being discussed and it would be a waste to bring it in.
(To be honest, this part becomes entirely subjective so I don't have much interest in it. It's the more objective rating calculations I think need improving.)
Baxter wrote:
I also don't know if this is a good assumption. I for instance would never watch a 4 hour rpg TAS of a game I haven't played, so I would never rate such a movie. If I had to, I would probably rate it extremely low. People watch and rate the movies they are in some way interested in, so the ratings that you do get are in no way a good representation of what the distribution of ratings if everyone were to vote, and finding out/approximating what everyone would vote (including people who have never player the game/have no interest in it) does not seem like a good goal then.
I understand what you're saying, but you're reading into what I said backwards. Or put another way, your problem solves itself. The point is that the movies on the site are ultimately capable of being voted on, and if you watch it and hate it and rate it 1.5, then that's an accurate sample of either "how a voter on the site liked it" or "how a watcher of this movie liked it". My model assumes movie ratings should attempt to be the mean of all voters, yours wants to be the mean of all watchers. But unless you force all voters to be watchers, our systems are one and the same.
(In other words, what my system calculates is "how all watchers liked it", which as the number of voters tends to the number of watchers becomes "how all voters liked it". This is no different than it is now. Your argument applies to the current system too, after all, since it's just an unadjusted mean.)
Baxter wrote:
Note that, as feos pointed out in the post above, discussing weight of entertainment/technical ratings or the way to calculate the final rating in general given the individual ratings is slightly off-topic.
Fair enough, but I figured if we're going to talk about redoing player points and discussing movie ratings, this was quite a suitable topic to bring it up.
Baxter wrote:
A average rating of a 6.3 currently gives about 100 player points. Just going by intuitions, how many player points would this 6.3 actually be worth if you know that it consisted of
There's no need for this approach, this is a solved problem in statistics, and I already outlined it. You're just ad hoc approximating the standard error.
EDIT: Actually Baxter, I mislead you and perhaps misunderstood where you were coming from, and was only giving you half of what you were asking for. I think that's where the confusion is. Sorry about that, let me recollect my thoughts and I'll post and update.
Thank you for that! Definitely a clearer perspective on what was jumbling around in my mind, including for me. I think this way simply makes more sense, as more natural division of movies.
And you're right that entertainment should still be a primary factor. If a SM64 A-less run wants to get submitted, if needs to have a wider audience than just the SM64 thread if it's going to represent the site.
If I understand what would get Star versus what would get Vault, that would mean, for example, the fastest 0-star SM64 run could be worth four times less than a slightly slower version that was more entertaining.
I understand the importance of entertainment, but that seems a bit wrong.
I think to do this at all we need to get some examples and see how the rankings change, and decide if that change is better or worse. For example, I agree with you conceptually, but maybe the number groupings should be smaller, like 1.2 for Star, 1.1 for Moon, 1.0 for Regular, and 0.9 for Vault.
(But we should probably wait until the tier's dust settles before decided how to do things. There are only three tiers now, for example, not four.)
You guys are right, I conflated "technical" with "speed". They technical axis was speed in my mind, I think I mixed the words up because I was thinking about the voting categories.
But I don't think the "speed/superplay" division is wrong: it's simply a fact that faster movies have been denied publication because of entertainment issues. I thought this tier proposal was a step to making the site not only a repository of entertaining movies (how it is now), but an archive of the fastest movies, even if they're not as fun to watch.
I agree that goals too esoteric should be denied ("press A only 37 times and Start twice, etc."). But I don't think people would accept an A-less SM64 run based on entertainment, even though all the most interested people, the SM64 community, would almost all assuredly want it. The answer to "I don't like it" for esoteric goals should be "then don't watch it"; only when nobody wants to watch it is it too esoteric. That seems contradictory to me to desire a category for the fastest runs but then say people have to find it entertaining anyway. (Whether or not they enjoyed watching how the goal was accomplished is different from whether or not that goal is well-defined enough to be on the site, IMO.)
I like your idea of turning demo into gruefood as long as my SM64 A-less example is publishable (again, example movie, not some hidden agenda). If it truly comes down to just entertaining or not, though, it does simplify things quite a bit. But I think at least we've made it clear there needs to be a fourth category for what I called Moon, for glitched runs, restricted goals, et. al.
Look, this is very solid. I like this, hopefully everyone else takes their time to read and appraise it.
Thanks!
"AnS wrote:
However, as for workbench voting, I think there's no need to make it any more difficult (one experiment has already failed in the past). Once the question is changed back to "Do you find this entertaining?", it will serve its purpose. Because technical merit and goal objectivity of a submission are more obvious to a judge, so he only needs to ask the audience about entertainment (well, sometimes he also needs to consider technical details, but such matter cannot be answered by a simple yes/meh/no, for this matter there is forum duscussion).
You're right, it does complicate it a bit too much. (Also, sorry for not beware aware of past experiments, I've been a bad lurker!) Whether it's primary or not is already built into the submission, voting would be redundant or contradictory.
But I still have a concern and I'll try to make it more concrete. Let's take for example a real potential secondary run: Super Mario 64 minimal A presses. This category is ranked by two factors: number of stars collected and the speed of their collection. Getting more stars than the existing run is an automatic win, and ties are broken by speed. Simple enough, definitely esoteric.
Now the point of these new tiers (I think) is to ensure these kinds of goals can be submitted to the site; they'll just be segmented off and clearly labeled "this requires you have a certain knack for this stuff" warning.
My worry is that if it's put through the current workbench system, where only the entertainment axis counts, enough people not interested in SM64 would vote no. If that happens, it's denied from what we're trying to make a valid thing. (And something similar has happened in the past for a primary run, where the 0-star record was beaten but the camera angles made it difficult to watch, so it's definitely possible for an esoteric goal.)
I think if the publication is clearly marked from the get-go as "This is Superplay/Superspeed/Demo", then we can vote on whether or not it belongs in that category. (And if no, the comments on the forums could say, "hey, no as a superspeed but yes as a demo, try resubmitting".)
<nitpick> "Quadrant" would refer to a four-way partition. If you are talking about a division into 8 parts in a three-axis system, the proper name would be "octant"... :P </nitpick>
Of all the proofreading I did that slipped by... :P (I'm sure among many others.)
This was quite a bit of fun. I ended up writing a C++11 program for demonstration, and source for it can be found on my blog (off-site to save space on this post). I wouldn't read it until you read this post, though. But as such, I omitted the math formulas from the post and instead provided links to the wiki pages describing it, as my code covers it all anyway (obviously).
Anyway, I'm mostly dealing with issue #1 in this post. I have ideas about #2 as well, but they are a bit dependent of the outcome of my other post from the tier proposal thread. So to avoid discussing things that may not turn out, I'll wait to post that stuff. Plus this is already quite lengthy.
So the problem is that a movie's ranking is, ultimately, an average of all the votes. (I'm going to keep referring to "a movie's ranking", but we all know there's actually two ratings: the entertainment rating and the tech quality rating. For now, just assume one; we can figure out how to combine the two into one later.)
As noted by Baxter, an average of 9 from a single vote is a lot less meaningful than an average of 9 from a million. Luckily this is a "solved" problem, and what we need are confidence intervals. What I'm going to describe is effectively an analog to the article "How Not To Sort By Average Rating", an idea made popular when Randall of xkcd pushed hard for it be to the sorting method used by comments on Reddit. We cannot use the algorithm directly because our votes are two scales from 0-100, rather than a single yes-or-no vote, but we can still use the confidence interval idea.
So to start, let's describe our overall goal. What we're really trying to calculate is the 'true' mean (average). That is, if every single possible viewer voted, we could take the mean and with 100% certainty say "this is the mean", because there couldn't possibly be another vote to throw it off. The problem is that not every possible voter casts a vote, so we have some uncertainty.
Warning: This post is about to get a bit mathy, skip to the text graphs to get a more intuitive idea if you don't care and just want pretty pictures.
We're going to calculate the 95% confidence interval of the mean vote. Basically, a 95% confidence interval says this: "there's a 5% chance that the 'true' mean lies outside of this calculated interval, but otherwise we're 95% sure it'll end up in here." I chose 95% because it's exceedingly common, and for our sample size (often not too large), asking for 99% or higher just ends up including most of the possible voting range, making the interval rather uninformative. The important part is this: as more votes are cast, the size of the confidence interval becomes smaller, because the confidence in the sample data's accuracy is higher. We'll see how to turn this nice feature into a final score later.
To calculate a confidence interval, we have to make an assumption about the distribution of votes. For those that don't know, a distribution specifies a probability to each possible outcome of a random experiment. Height is a very common example, and it can be modeled with a normal distribution. This means that there is an average height, where most people's height is, then it tailors off as you leave this average height (see picture on the normal distribution wiki page).
The normal distribution is very common across many measurements, and it's no different for votes. We can see individual votes as "guesses" to the true mean: there's going to be a single concentration of votes around an average, with the frequency of more deviant votes lowering as they become more extreme. Luckily for us, calculating the confidence interval from a normal distribution is easy, as is calculating the parameters for a normal distribution from sample data (votes).
Here's an example to make this concrete (all text graphs generated by the aforementioned C++11 program, with 1 million samples). These are the actual votes for Super Mario 64 (all votes are, for the duration of these examples, only from the publically-visible Entertainment column):
We can calculate the parameters for a normal distribution of this with just a few values. First, we need the mean. This is easy and already done on the site. Second, we need the variance and deviation (the deviation is just the square root of variance). The variance has more than one way of calculation, but I chose a fairly simple bias-corrected method.
Next comes the critical bit. We calculate the standard error. This is the deviation divided by the square root of the sample count (note: this is where sample count starts to come into play!). Luckily for us, because our voting population is fairly small, sometimes we get a fairly significant (>5%) sample size! Assuming a voting population of 100†, it only takes 5 votes to meet this criteria. This means it's worthwhile to factor in finite population correction (FPC). What this does is account for the fact that as the sample count (vote count) nears the total population size (number of voters), our confidence increases towards 100%. Once it reaches 100%, we no longer have a sample but a census, and the sample mean is the true mean. We just multiply our initial standard error by FPC to get the true error. (This is something we're lucky to be able to take advantage of. Consider a site like Reddit where only a tiny fraction of the entire user base will vote on a comment.)
Last, we need the z-score for the 97.5 percentile point of a normal distribution. (This is the number of standard deviations away from the mean that 95% of the values lie). It's not trivial to calculate, but it's constant and the value is approximately 1.959963984540.
So now we can calculate our 95% confidence interval. Take the mean vote score and subtract from it the error multiplied by the quantile to get the lower bound, and instead add this product to the mean to get the upper bound. Now since we know the interval can never go below 0 or above 10, clamp it if necessary. Ta-da! We have our interval. This interval has a 95% chance of containing the true value (though there is a 5% chance our sample mislead us!). Note this has the desired property of being dependent on the number of votes cast.
From the SM64 votes above, the resulting normal distribution is thus, where #'s indicate votes within the confidence interval and *'s indicate votes outside the interval:
Note the calculated numbers displayed under the graph. To demonstrate how the interval size changes depending on the sample size, here's the exact same calculation except only half of the SM64 votes are used:
As you can see, lower samples imply a larger range. The final step is to turn this interval into a single value. The linked articles on Reddit's comment ranking simply take the lower bound of the confidence interval, and this works extremely well. The reason is that fewer votes will bias the lower bound to a lower score, which solves our original problem: a few votes end up contributing less than many votes, even when the average is the same, because the confidence interval will be wider.
The justification for this is approach simple as well. The lower bound says: "I'm 95% certain your true mean won't be lower than this, but get more votes and I'll let you know! But until then, this is quite fair of a rank to get because I'm 5% sure I'm not overinflating your true mean; in the rare case it's wrong, you're welcome for the bonus." In practice low-voted movies won't get punished that much, but enough to knock near or equal means away from each other. (At most a single point: 7->6, for example.)
Also, remember that because our population is finite, as the number of votes reaches the total number of voters, the error value tends towards zero. At some point, with every vote accounted for, the error is zero and the lower bound and the mean coincide, giving the true mean.
So for SM64, the final score would be (roughly, as I don't know the private votes): 8.58737, which is a difference of -0.381055 from the simple mean. Note that every score on the site will go down slightly as the number of votes gets taken into account. This is okay: we only care about the relative ordering, and the number won't vary in practice that much at all. (Keep in mind that for ranking and calculations for problem #2 these need to store all the decimal places; truncated/rounding to one decimal for display is sensible.)
Usage-wise, just note that a single vote is not enough to calculate variance, and two votes can give a very meaningless answer unless the two votes happen to be close to each other. The site already requires three votes before it calculates a rating though, so that's good.
Turning two scores into one will need to be discussed after the tiers thing settles down. I think a movie in an "entertainment-based" category should get most of its score from the Entertainment rating, while a movie in a "technical-based" category should get most of its score from the Tech Quality rating. It might be worthwhile to investigate interval arithmetic (something I'm not as familiar with) for this task.
And that's it. What essentially comes down to a few multiplications and a couple square roots gives us a very meaningful and theoretically justified score (not just ad hoc tweaking). Here are some more example plots and data:
The program contains additional tests with manually constructed data.
Thanks for reading.
-----
†There could be some argument about what this finite voter population should be. It can either be the number of people who have cast at least one vote (are voters), or the number of people that could cast a vote (could be voters). Right now the most rated movie has 89 votes, so for my tests I assumed (using the former metric) that the total voting population is 100. If we go with the former, to my knowledge that means the number of registered forum members is the count, nearer to 5030. With this metric, it takes 230 votes to reach 5%, and considering the number of lurkers there are this seems useless to me. Either way though, a new calculation must be done for each movie every time this population count increases, and I assume the former increases less often (thinking about server load now). Another implementation strategy is to have "gates". These are just multiples of 50 (for example), and each time a gate is reached the population count increases by 50 and the gate is increased by 50. This allows the FPC to near 1 on highly-rated movies, but avoids constant server load whenever the population increases due to happenstance.
(Huge final note: there are other and potentially better ways to do everything I described: maybe a credible interval would work better, or another distribution better fits the data. I think, though, that the chances that a normal distribution being insufficient are extremely small and not worth the computational effort required to move to a hypothetically better model. What I've presented is fairly easy to implement.)
I think a lot of this confusion has come about because we've split the categorization of movies into three different axes, but have yet to explicitly name the resulting quadrants before we talk about them. So we're talking about ideas not even made concrete.
To me, the splits are as follows: Entertainment vs. Technical (speed) vs. Goal. Games can have high or low entertainment values, high or low technical merit, and popular or esoteric goals.
If you excuse the poor Visio 3D diagram, I've made visualizations of the resulting categories. Note: I made my own names for the categories that make sense to me, and I'll detail them below. Most important, I've reevaluated what I think a "Moon" movie should be, so don't let that throw you off.Please don't read the rest of this post until you've read the above-linked visualization page.
-----
Now that that's out of the way, we can talk about which categories belong on the site. Obviously Grue isn't published at all.
Star and Moon are two halves of the same coin (highly entertaining, higly technical), just with different goals, obviously both are published. I think Superplay and Superspeed belong as well: Superplay being effectively what voting was prior to this entire proposal, and Superspeed being originally proposed as "Vault". We all agree that people want entertainment and speed; just not all agree to what degree of each. This solves that problem.
As the discussion went on it was clear people would still like to have technical glitch runs (not belonging on the primary goal side of things), which is where Demo comes in.
The only one unsure (to me) is Freerun. People make freeruns all the time of various games (especially SM64, for example), and I don't think they belong on the site. Not that they're bad like Grue is, but without an obvious goal a freerun is purely entertainment, unlike its counterpart Superplay which, while intended for entertainment, is still working towards a fairly popular goal. In other words, if technical merit (speed) isn't an issue, then any run could effectively fall back to "Freerun" status, including the dreaded Monopoly runs. (That is, the only thing that keeps Demo in but Freerun out is that it has the highly objective goal of technical excellence. This provides a meaningful-enough ranking.)
To answer the problem of "but I don't like Superspeed movies" or "but I don't like Superplay movies", we start by agreeing to show Star movies on the front page. Everyone already agrees these movies are truly the goal of the site. (And maybe fewer Moon movies mixed in.) But then if someone asks to see more by clicking Movies or otherwise navigating to the "main" movie storage, they need to be able to select, independently from their navigation through the site, what categories they would want to see. I envision something like this:
Where the top options are independently toggleable, and all selected by default. If I know I don't like Superspeed, I just untoggle it and those movies aren't displayed to me as long as the cookie survives.
---
I think that voting is a separate issue. If we were to have an ideal voting system (regardless of feasability), I would change the current "Yes/No" approach to publishing and instead make it a ranked vote. That is, display the following rankable options:
- Superplay
- Superspeed
- Demo
- Grue
So for example I could say "(1) Superspeed (2) Demo (3) Superplay" as my preference to how the movie is published. That is, I really consider the movie a Superplay, but if not enough people agree then I think it should be published as a Demo, and if that doesn't pass then a Superplay. I don't rank Grue because I think the movie should be published regardless, and I don't want to give weight to it being denied. (Though obviously that doesn't stop others from voting Grue.)
If instead I think it should either be Demo or nothing, I'd vote "(1) Demo (2) Grue".
You can use Schulze Method to determine the final result. But again, this would require some special coding, I'm unsure how feasible that is. Regardless, I don't think leaving the voting as "Yes/No" makes sense with the new scheme, because whether or not I think it should be published depends on how the movie is trying to present itself.
Star and Moon are not options because they are just "ascended" publications, hand selected. Stars being highly-entertaining superspeed runs (or equivalently, highly-technical superplays), and Moons being highly-entertaining demos (or equivalently, highly-technical freeruns).
Anyway, those are my thoughts on the matter. Thanks for reading.
(Note that I think the overall rating of movies and players can be improved as well, as is being discussed; see this post.)
I think you have misunderstood my point. I don't mean that you actually need to write down a number for it to "count". I consider formal expression just as valid.
Let's consider my view of the "size" (let's go with s(n)) of a number. Let's have a look at a big number and express it in a number of ways:
- 2147483647
- 8th Mersenne prime
- Int.MAXVALUE
- 2^31 - 1
The point is, no matter what formal expression we choose, we still need some amount of characters to express it. Now I consider the size of a number to be the least amount of "units" with which I can express it.
Let's also assume, for the sake of argument, a system in which we can express our numbers. Obviously, our current system is shit as it uses a crapload of pixels (or ink or whatever) to even express something as simple as 0 or 1. We could obviously do better by letting "absolutely nothing" represent 0 and we'll have a single Higgs Boson be 1. Then we'll need to define quite a few primes so we'll define the function m(n) (the nth Mersenne prime) and we'll have it be represented by a neutrino. Now we can get our number by having a neutrino followed by eight Higgs Bosons!
Obviously we can do better than this, so we'll just assume our system S is the optimal system. That is, in our system the size (as defined before, units being elementary particles in this case) of any given number is the minimum of all possible systems.
I've tried, I've honestly tried. I have the best system possible. But I still end up with an infinite amount of numbers that are all non-zero size. I also seem to have an upper limit (u) to my size: the amount of elementary particles in the universe. But as there's infinite numbers, there exists at least one number c for which s(c) > u. Crap, I can't express that number.
I don't think this is different than what I said at all, but then obviously that means I'm misunderstanding you. :)
In my head, you're saying (as Warp is) that we have a system S that can in principle write any number. "In principle" meaning given infinite resources and time, every number can be expressed. For the sake of simplicity I want to use the sub-Peano system I defined above, because I think it's safe to say improvements on this (like your particle example) are just scaling factors.
Start with 0. This is the first natural number. Now put an S in front of it: S 0. That's the second. Now put an S in front of that, ad infinitum. For any natural number, this process will eventually write it out, by definition.
What we all agree on is that we, in reality, cannot actually do this process.
Is that correct? When I was reading your response I think I had a glimpse of what you were trying to imply, but I lost it as a thought about it. Could pick out
the wrong part about from above? Sorry for being dense!
Scepheo wrote:
I'm not claiming any system to be better than any other system. In fact, I don't see anyone doing this. Where did you get that idea? But still, my point is that no matter how hard you try, there will always be numbers you can't operate on.
I meant the mathematicians that argue about the things I linked to (it happens). My bad.
At the moment, data like that is not easily available. Unless Ilari wips something up.
You can see the individual ratings by clicking the number of votes. Some people have chosen not to show their votes publicly, so you won't be able to get the complete data you'd like probably.
Yeah, I just ended up grabbing data manually (the public data anyway) from a select few games.
To make my understanding clear, we have (in utmost generality):
A set of movies with a set of votes in [0, 100] (divided by ten for display) for a set of different categories, with a set of authors? (In the current case, categories are "Entertainment" and "Tech Quality".) And there's potentially going to be a tier.
We want to, for each author across all movies, generate some score based off all this information?
I'll post something later tonight with what I have so far. It's more sweeping than just changing the score calculation, though, and would also effect the calculation of a movie's final score as well.
The current and suggested methods do seem a bit ad hoc, I'm interested in sampling some data to run some tests, is there any easily available?
That is, is there a dump of data containing movie votes (note: not the final average score, but the votes themselves; I think using the average might not be the best strategy for this), movie author count, and resulting score?
You're both making valid points but I think you're both talking past each other.
Warp et. al. are saying that we can create a definition natural numbers such that any natural number can be expressed. So if we have, for example (purposefully incomplete):
1) 0 is a natural number.
2) For any natural number n, S n is a natural number.
(For those not familiar with this notion, S stands for Successor.)
Then, by definition, any natural number is expressible as S S ... S 0, because if it weren't then it wouldn't be a natural number per the axioms. (And we could further define some rules that turn S S 0 into 2, and vice versa.)
Scepheo et. al. are saying that it is impossible to actually completely write an arbitrary natural number. In other words, the ellipses I wrote above can contain arbitrarily many S's, yet at some point it is impossible for me to physically write down that many S's.
This "paradox" comes from the fact we can claim a function has a certain kind (also called sort or type in other formal contexts) of value, yet not calculate that value. For example, Graham's number. We can say this is a natural number without actually writing it down: S S S ...(many many!) 0. Whether or not we should be allowed to make these kinds of claims is (to some) up for debate.
The idea that numbers should not be considered extant if they cannot actually be written down is called "Ultrafinitism" or "actualism". So while you can express an algorithm for calculating Graham's number, you cannot say the result of this algorithm is a natural number until you write it down.
I personally find this point of argument meaningless, from both sides. Mathematics is a tool. If assuming I can operate on results I cannot write down as if I could is useful to me, then I will make that assumption. If it turns out this assumption introduces problems that make my system no longer suitable to me, I won't. Either way, to claim a certain mathematical system is the "right" one is just nonsense.
I looked around and I couldn't find this thing in specific.
I was wondering if there is a TAS of someone doing the warp in meteo in cockpit view. I managed to do three rings as a kid, but I eventually gave up on doing it because it's really hard. It's one of those "Now I can die happy" kinda things lol.
Three rings? Do you mean Training Mode instead of Meteo Warp?
No, what is meant are the warp rings right before the boss of Meteo that take you to Katina/Katarina if you go through all seven of them. Try going into first person mode as soon as you begin rotating, and then try to go through all seven of them. You'll see it's extremely difficult.
Oh duh, thanks. Doing the warp rings, not the warp level. That would be interesting to see.
I looked around and I couldn't find this thing in specific.
I was wondering if there is a TAS of someone doing the warp in meteo in cockpit view. I managed to do three rings as a kid, but I eventually gave up on doing it because it's really hard. It's one of those "Now I can die happy" kinda things lol.
Three rings? Do you mean Training Mode instead of Meteo Warp?
Do you mean in principle or practically?
There's no reason in principle why lag cannot be emulated properly, it's done through cycle-accurate simulation.
The symmetry of the axis graphic is reflected here, which is why I find this layout intuitive.
Sensibly, adelikat wants to give a place to optimal movies that aren't entertaining, and everyone agrees with this. I think the problem is only trying to add on the Vault category leads to imbalance, which is why people are suggesting additional categories to balance it out. So this is the desired layout from the original proposal (if I understand it correctly):
The problem is that people perceived this as ending up in practice as:
Because the categories seem diluted. I think this happens because Star is trying to take on some of the load Moon has from not having Playaround and Demo to help out. Really Star ought to have remained exactly the same subset of movies, and only movies that obsolete more entertaining runs (as well as previously Grued runs) would be reconsidered and moved to Vault. This would have only reduced the size of Standard movies and given Vault its place.
So that's why I think it may be worthwhile to tryin again but with all the organizational categories.
What I think adelikat wants to remind everyone though is that runs aren't being chopped off any more than they were in the past, and it's in fact the opposite. More runs will be accepted now because of Vault. Just compare the above decision procedures with what we used to have:
Nothing is being lost, except IMO the clarity of categories.
Just toggle the attributes to easily filter. I think the overall point is that navigation should be interactive and easily filterable. (Not that implementing this is trivial to do.)
The list should either be sorted by use-count or split-count. That is, most popular attributes to least popular, or most dividing to least dividing. I can't think of which one is better.
Where the top options are independently toggleable, and all selected by default. If I know I don't like Superspeed, I just untoggle it and those movies aren't displayed to me as long as the cookie survives.
---
I think that voting is a separate issue. If we were to have an ideal voting system (regardless of feasability), I would change the current "Yes/No" approach to publishing and instead make it a ranked vote. That is, display the following rankable options: