ForumsGamesGemCraft CS: gem combining schemes competition

89 35599
daniil_sizov_98
offline
daniil_sizov_98
30 posts
Farmer

Some of you could have read my post on reddit (http://www.reddit.com/r/Gemcraft/comments/2385eg/sc_about_gem_combining_mechanics_and_why_we/) in which I explain how does gem combining works. But one of conclusions I've made â" "Try to avoid combining one color gems of different grades" â" is correct only in theory, since if you combine one color gems of different grades your gem's attack speed will suffer a lot. But in practice attack speed is capped by 30 shots per second.

So, in practice you can get some gain in specials (if your gem's attack speed is greater than 30 sps) by using special combining schemes, e.g. (((G20 + G20) + G20) + G20) instead of ((G20 + G20) + (G20 + G20)). By this post I want to start new funny competition â" for best combining scheme for Black and Orange! To set the new record you can post your G60 gem analogue (i.e. gem with the same cost as G60 gem), but since your scheme may not allow you to make G60 gem analogue, it will be better to post your ~G60 gem and a special value (let call it "growth speed&quot, which is:

log(result_gem_special / base_gem_special) / log(n) (n is number of identical base gems combined into resulted gem)

Let say you have base pure gem x and combining scheme g(x1, x2, ..., xn) (g(x1, x2, ..., xn) combines n input gems into one resulting gem). Then "growth value" will be (log(special(y) / special(x)) / log(n)), where y = f(x) = g(x, x, ..., x). You need to find such a scheme g(x1, ..., xn), which has maximum "growth value". Since G7+ gems have extra bonuses and gems lower than G7 have not, please, use only G7+ gems as your base gems to calculate "growth value".

Why I'm starting such a competition?

Because optimal combining schemes for Orange and Black is Holy Grail for all Gemcraft SC players. Growth values for standard combining are ~0.4647 for orange and ~0.1243 for black for total ~0.589 for Black/Orange. That means, that if you grade your B/O gem up, it's manaleech power will be 2 ^ 0.589 = 1.5042x, and after 80 grading ups it will be ~152'913'823'388'742x. But if we could find schemes, which will show even e.g. 0.51 and 0.14 growth values, if you grade your B/O gem up according to those schemes, it's manaleech power will be 2 ^ 0.65 = ~1.57x per grade, and after 80 grading ups it will be ~4'503'599'627'370'509x. So, G81 gem will leech 30x more mana and it will be 30x less time to get G100 gem and beat the endurance even without WoE and shrines

I will start:

Orange G61 analogue
http://i.imgur.com/qm19DGE.png
Orange growth value
http://i.imgur.com/h0fPVqZ.png

Black G61 analogue
http://i.imgur.com/3SOPW1A.png

Black growth value
http://i.imgur.com/sqWIZGV.png

Total groth: 0.62851
scheme_G81_manaleech / standard_G81_manaleech = 2 ^ (0.62851 * 80) / 2 ^ (0.589 * 80) = 8.94x

  • 89 Replies
Suuper
offline
Suuper
34 posts
Nomad

There are a couple things that are not quite right/misleading about the way you described it, though.
1) Saving the copy from one iteration to use (for the 13c) means that you can get slightly faster growth than .492. True, that's the growth over the initial grade 1 when using 1 iteration, but repeating it will raise that.
2) It can be repeated every 15.5x mana gain, you don't have to wait for a big 244x mana gain.

Anyway, I'm going to look for better methods. Hopefully I'll find some.

Suuper
offline
Suuper
34 posts
Nomad

Started with g1, used my method to g19. Growth from g7: 0.49225
Using a new method, same grades: 0.49255.
New method's cost increase is 15.26x. It's just a modified version of my previous one .
7 = 1 + 1 + 1 + 1 + 1 + 1 + 1
2 = 1 + 13A
Result = 7 + 2 + 2 + (2 + 13A + (13A + 13A))

Still looking for more...

12345ieee
offline
12345ieee
135 posts
Farmer

Tell me when you can beat this, at a growth of 0.5039062

Keep looking...

Astroshak
offline
Astroshak
268 posts
Peasant

LOL that's cruel :P

Suuper
offline
Suuper
34 posts
Nomad

I was going to use that combine method, but I would need 65 slots to hold gems, and scripting anything that used more than the 36 inventory slots is too much trouble. :/

12345ieee
offline
12345ieee
135 posts
Farmer

Get some smaller ones, we have tons of combine:
https://github.com/gemforce-team/gemforce/tree/master/results
If you can make a script that build a generic combine I'd be very interested

Suuper
offline
Suuper
34 posts
Nomad

Yeah, I can (did) make a bot to do my large combines for me. Hasn't been tested with lag yet and needs a few features/changes before I'll release it.

Also, I'm still working on finding ways to quickly calculate optimal gem combines. I've found that 2^x combine methods are not the best. (E.g the best combine using 500 or less is a 383c.)
(How long did it take you to calculate that combine? My current methods would take weeks, if not months.)

And, I realized that how I was calculating the number of slots required isn't really accurate. Other combining orders can do better, but I still don't think that that huge leech combine will work in 36 slots.

Astroshak
offline
Astroshak
268 posts
Peasant

Suuper .. personally, I've found that the best way to handle a large combine is with a spreadsheet.

By having a spreadsheet handy, it is possible to easily do a 1024 combine upgrade. And once you've figured out how to do it by hand, if you're into that sort of thing, you can script it relatively easily.

The 1024 managem upgrade plan I've got would not require all 36 gem box slots, because it makes use of the Duplicate key, and I assemble a lot of gems by hand then add them to the gem I'm building. The same thing with the 1024 killgem upgrade - this one is even easier, as I can make most of the sub-subgems and dupe those to create the subgems to add to the gem I'm building.

Suuper
offline
Suuper
34 posts
Nomad

A 1024 combine can be done with much fewer slots than 36, and I've used my bot for 7,000+ combines also with much fewer slots.

I'm not looking for the most practical solution that has already been found; I'm looking for even better. I enjoy a challenge.

(And as I've said, I already have a 'script' for combining. Paste combine pattern -> press a couple buttons -> watch it make quick work.)

Btw, I'm assuming that the combines that ieee just linked to are all up-to-date best known methods. Correct?

psorek
offline
psorek
447 posts
Jester

Yes, they are We are also pretty sure that they can't be done better, that means you won't invent better 1024-combine than ours becouse we just checked all possiibilities of combining 1024 same gems into one and chose best one.

Suuper
offline
Suuper
34 posts
Nomad

You couldn't have possibly checked ALL for the higher combines, but I believe you about the pure orange being best possible. Multi-colored I haven't looked into thoroughly, but I can see a way that the same search methods might fail to find the truly ideal method. (I'll post back on this once I get around to my testing/searching.)

12345ieee
offline
12345ieee
135 posts
Farmer

Orange gem combines are EXACT, unless rounding errors you CANNOT do better, no matter what algorithm you use, we REALLY checked EVERY combination till 1 million base gems.

Composite gem combines are slightly approximated (or we couldn't go over 512c), but I can guarantee that you cannot do better than 1/1000 over ours (if needed I can get exact combines till 512, if you need to check against yours).

As for 2^n combines not being the better ones it is a well known point, we use 2^n because they are easier to compare against each other.

I have all the optimal combines for orange up to 1M and managems and killgems up to 262144, if you need them to compare against yours just ask (you can find them in the gemforce-tables repo, but you need to use the gemforce programs to extract them).

I don't think you should reinvent the wheel, though, I took 2 months to build over psorek's algorithms to improve and extend them, you'll take much longer to achieve better/bigger results than gemforce ones (and, to be honest, I don't think you will succeed unless you are at least a CS graduate).

If you can find better combines than the ones we have I'll be happy to hear, if instead you are interested in building over our project we could talk about it.

Lastly, any chance to see your auto combining script?

Happy gemcrafting
12345ieee

Suuper
offline
Suuper
34 posts
Nomad

No you didn't, there are millions of millions of millions of ways to combine 1 million gems. You may have found a way to find a proven best still, but calculating through them all isn't possible.

I'm writing in c#, do you want code or the compiled program? (Windows)
(Either way I haven't finished it yet.)

12345ieee
offline
12345ieee
135 posts
Farmer

Maybe I explained it badly.
We don't compute all the 1,000,000! permutations of base gems, we keep reusing the results computed in the previous steps to compute only the needed gems (this was the great idea of psorek) and only compute the ones that can actually be useful, but the algorithm is still 100% exact and runs in O(n^2*log(n)) (the leech one, the composite ones are a bit slower, but can be easily made exact).

You can see it yourself, it's released open source at https://github.com/gemforce-team/gemforce

As for yours, I'd be very interested in it anyway, but I'd really like to have the source code so I can understand it and tinker with it (I know some C#, not like C but well enough).

UgAhgItHurts
offline
UgAhgItHurts
259 posts
Peasant

Suuper, a single gem type is a much easier problem than multiple types, your posited trillion possibilities (actually much larger, but a trillion is no big deal to a comp anyway) have tons of degenerate solutions that give identical results.

You can further limit the solutions of interest with the combining formulas, grades (n+1)+n being the biggest bump, you just have to optimize for getting those best combinations as many times as possible (again easier for pure orange since the only thing of interest is the special).

Of course there's some contention in the starting assumptions for how to construct the most effective killing/mana gain platform (which has some variability), things vary. For example if your goal is to finish endurance vs maximize xp, you'll build very different things to do that, at least there isn't a gem spec/combine that'll do all that yet

Showing 61-75 of 89