Why is Elo decreasing so much?

I rembemver, a few weeks ago I was 10th with an elo of 2213 and now I’m 8th with an elo of 1969 and the 10th guy doesn’t even have over 1900 anymore. How has it come to this? Is it because of kkroep and ryan d? Because they seem pretty much unaffected.

There are just at lot more algos in the top at the moment, which causes deflation. You do not see it on the leaderboard because the same users have many algos high up right now (and remove them as well).

Overall there will always be a normal distribution as explained here.

1 Like

I tend to agree with 8’s (and others’) analysis of the ELO mechanisms. Everything seems to be more or less reasonable.

That being said, I would love to know how ryan and kkroep got ahold of such an insurmountable lead, and held it while everyone else plunged down to 1900’s. I’ve never seen anything like that gap in all my months of playing terminal. Well done.

1 Like

I am partially to blame for the large shift in elo. I’ve been experimenting quite a bit, and specifically the drops in elo I observed recently match my upload schedule. So several of my algorithms have been dragging themselves up to the other top 10 algos. By then making new iterations and deleting old versions with high elo I have been draining quite a bit of elo from the top 10 I think. (that was not the purpose though, I am genuinely trying to develop a good algo).

Unfortunately, my most recent algos have a distinct weakness that is mercilessly exposed by @Ryan_Draves algos… :confused: So I’ll probably keep experimenting a bit more

kkroep already said something similar, but people who are uploading high performing “top 10 killing” algos would explain the drop in elo across the board. My old track algos (Track-3.4 and Track-3.9) have both experienced around the same rise and drop of elo at the same time, but have always settled back around 5th place after a while. The current state of the leaderboard shouldn’t be taken very seriously when trying to find out the end state after the week of settlement.

Yes, of course… what’s that weakness again?

Anyways, I noticed something similar with the deflating elo, but I didn’t think my uploads would be enough to cause it. Perhaps the top several players climbing and taking down top algos through many iterations has been enough.

I imagine there are lots of new uploads and submissions from many different people that is contributing to it. I know personally I have uploaded a ton of new algos in this last week.

I’ll tell you the weaknesses of my algo somewhere in januari. Is that okay? :stuck_out_tongue:

Yeah this is true, I see a lot of iterations on the ladder. And because it is mainly the top 10 who are iterating before the deadline, a lot of high elo algos are deleted and the elo is flooded out of the system.

I notice some impressive improvements on the ladder. I guess everybody is trying to give it their best shot before the finals. Good luck everybody!

2 Likes

Looking at recent development, you can see that ranks 3 to 10 are extremely volatile, playing many matches while only one of rank 1 and 2 played any match.

You do not even need the graphs, you can just look at their creation date and match count and will notice that they barely play. 71 matches in one week is not much at all. Your algo played way over 100 matches in less time (way over because even the first matches that are still visible are against high elo algos).

Both of the algos got up there and now will not change their ranks for the moment because they are not assigned any matches.

1 Like

This is exactly right, I was making a new forum post about this when I saw this.
This is a huge problem. GA isn’t the best algo of ryanD and demux_1.9 certainly isnt my best algo currently. But all the other algos are so low that my better algos cannot climb high enough to overtake my worse performing algo.

Now I am in a situation where I am considering to take down my top, but old algo, because I don’t want that algo to win over my others.

The second problem is that I believe there are plenty of algorithms that where developed very recently, that could challenge the current top 2, but they never play so we won’t find out. For demux_1.9 I know this to be true, because I have other iterations that loose to which demux_1.9 would have little resistance. Because of the volatility the top algos are stuck on an island. Especially the same match-up never happens twice, as soon as the top algo has played against all matches within range (which is a small amount), it just sits idle. GA has a shockingly low match count in the last 4 days.

This is a flaw in matchmaking that wasn’t exposed before because the elo gap was never this large. I hope this can be looked at before the grand finale… but I’m worried about this

[edit] I’ll make a separate forum post about this because we need to get this to the attention of the terminal team urgently. Please help that discussion to bring it to the attention of the developers

1 Like

Another frustration I have with the frantic iteration this last week (which is entirely expected and understandable) is that the leaderboard is in such a state of flux that it is difficult to discern any meaningful information about it in terms of where you might expect your algo to be at the end of the settling period. I’ve seen all places on the leaderboard in 3rd through 12th be scrambled around on almost a daily basis, and the losses of my own algos are like a revolving door as opponents are put up and taken down.

I’m not really sure what I would want different though, considering that this IS the last week and of course everyone wants to improve. But it sure does make me nervous realizing that (possibly for the first time since I started playing) an algo being in, for example 3rd, doesn’t provide the slightest assurance that you would even be in the top 10 next week. With lots of algos reaching the top on a sort of a “sugar high”, with successful opening games but many opponents yet unmet, I’m finding the results of the final top 10 to be, for the most part, largely unpredictable.

Yes, I agree with this, yesterday I went to bed as 3rd and woke up as just about 10th. The positions just change so quickly that I’m kind of afraid of not being in the top ten when the competition ends