WG - A simple well known C(19,6,3,3) - b=65 == 100%

Talk about anything Lotto Architect related which doesn't fit in the other forums
Post Reply
runes
Newbie
Newbie
Posts: 4
Joined: Sat Jul 30, 2016 9:18 pm

WG - A simple well known C(19,6,3,3) - b=65 == 100%

Post by runes » Sat Jul 30, 2016 10:08 pm

Hi everyone,
I was looking at the wg, but what striked me was that it failed to find a simple relatively well known covering of C(19,6,3,3) - b=65 == 100%, at least in few hours of pc processing time.
I had some parameters fiddled so it can avoid local min and I managed to make it optimise to 11 missed, ie %98.86+, buts still I am sure that this can even be achieved by a standard simulated annealing using the same pc time usage.
Also the problem is even more profound when I checked the Anastasios posted optimum C(45,5,3,5) - b=359 - L1 (100%) with WG. WG after couple hours just stays at around %97.1.
So my questions are:
Is WG able to find the 'best' coverings? If yes how?
Do we have to fiddle with the parameters to avoid local optimums and 'pray' till we get something?
Or we just use it as a tool to construct:
C(v1,k,t,m1,L,=b1)+ C(v2,k,t,m2,L,=b2)= C(v1+v2,k,t,m1+m2-1,L,=b1+b2), when possible?

Thanks

PS:
1. )BTW I like WG and GAT and this is not a negative feedback on the contrary, it is an admiration to the efford Anstasios has put to these apps.
2.) I have to admit that GAT is very vague to what it can do, I can c that u can not predict a number/draw 'cos is governed by God knows how many variable params and probably u can only observe their short term behaviour but how it does that, if it does, is still unkown to me, probably by some meean of random seed generators, who knows :)

User avatar
lottoarchitect
Site Admin
Posts: 1635
Joined: Tue Jan 15, 2008 5:03 pm
Location: Greece
Contact:

Re: WG - A simple well known C(19,6,3,3) - b=65 == 100%

Post by lottoarchitect » Sat Jul 30, 2016 10:51 pm

Hi runes, interesting questions. WG is very capable to produce many world records but it requires some guidance during the development stage. First, the covering c(19,6,3,3) has a current known record of 60 blocks, not 65. Either way, set search to extensive, bias expansion to -930 and bias tension to -930 and b to 64 and you'll get the covering in 64 blocks in a couple of minutes and in 63 blocks in five minutes. After that, you may try to create the current record of 60 blocks, by removing a couple of blocks at each step, or directly set b=60 with the same bias parameters and wait. You possibly need to tighten even more the search (<-930 for both expansion and tension) if no improvements occur after a while. It may be possible to even make it at 60 blocks within a few hours - haven't tested it however. Having said that, WG cannot construct each and every known world record. Especially it cannot approach records that are derived from covering theories that produce an optimal (in terms of blocks) output, typically this refers mostly to symmetric constructions (e.g. steiner). So, the reason you couldn't make this covering is "bad" utilization of the engine, the two most important parameters (bias sliders). In most other cases WG can equal or surpass the known records and even if it can't do that on its own within a logical time frame (refers mostly to huge coverings), if you feed it a recent record it has the ability to further improve it.
C(45,5,3,5) is a split wheel, produced with the equation you present. WG does not produce split wheels. It happens, a split wheel to always require fewer blocks compared to a non-split wheel. I would always prefer a non-split version of any given wheel really.

As for GAT and how it does its analysis, I hope you understand I cannot really say much about it. You can be sure however that it does not have anything random within its process. No RND seeds involved. What it does is really unique, never attempted before.

runes
Newbie
Newbie
Posts: 4
Joined: Sat Jul 30, 2016 9:18 pm

Re: WG - A simple well known C(19,6,3,3) - b=65 == 100%

Post by runes » Sun Jul 31, 2016 12:07 am

Hi and thanks for the reply,

I had the feeling it had to do something with the biases..
It was not so clear to me, that I should select bias values so close to their lower limits and the actual 'mathematical' meaning of these values.
I assume from the documentation and my 'gut' feeling that this is how u avoid local optimums, but anyway, I can understand why you do not want to reveal more about yours algs and I fully repspect that.
About... Steiner's/Symmetrics, that's why mathematics r for.. to find these :)
C(45,5,3,5).. thanks for the info about the split wheel, this beast with 359 looked impossible to crack. And believe me I tried many opt algs from algebraic up to metaheuristic, not to mention the time wasted on reading papers about comb. opt. algs.
As for GAT ,that is why I put the smiley at the end, I am sure it has nothing to do with random or random seeds, I was just kidding. Although it looks like that 'God plays the dice with the universe' so why not any lottery system?
On the serious side of things, I have not really look at GAT in detail, but judging from the WG and the way you approach the whole mathematical aspect of coverings, I am positive that I will be pleasantly suprised.

Again thanks for your reply and keep the good work

runes
Newbie
Newbie
Posts: 4
Joined: Sat Jul 30, 2016 9:18 pm

Re: WG - A simple well known C(19,6,3,3) - b=65 == 100%

Post by runes » Sun Jul 31, 2016 7:36 am

Hi again,
I would like to confirm that C(19,6,3,3) b=65 L=1 was found fairly quickly once the biases were set to the values you suggested.
Furtheremore by just removing a line each time and re-otimizing I was able to optimize down to b=63 in a couple of minutes.
Using b=62 and after leaving it for an hour or so, it did not reach 100% but I am pretty confident that it will, once I change the biases to lower values or use the prog to a faster PC.
I am really impressed from WG and I have not even scratch its capabilities yet.

runes

User avatar
lottoarchitect
Site Admin
Posts: 1635
Joined: Tue Jan 15, 2008 5:03 pm
Location: Greece
Contact:

Re: WG - A simple well known C(19,6,3,3) - b=65 == 100%

Post by lottoarchitect » Sun Jul 31, 2016 9:04 am

I managed to make the record 19,6,3,3 in 60 blocks. After constructing it to 63 blocks, I changed the search to standard and bias to -878 (expansion) and -707 (tension), removed one block and waited for two hours. This made it to 62 blocks and after removing two more blocks it was again at 100% right away. So WG managed to produce the current record. The bias settings here are in fact more "loose" compared to the initial I suggested and this covering looks like it requires "bigger" jumps allowance to be constructed.

runes
Newbie
Newbie
Posts: 4
Joined: Sat Jul 30, 2016 9:18 pm

Re: WG - A simple well known C(19,6,3,3) - b=65 == 100%

Post by runes » Sun Jul 31, 2016 8:54 pm

Hi,
I have not yet managed to reach 62 and then 60 but I am using WG on a slow pc, so I am confident if I leave it run overnight, it will eventually find the covering.
I have to admit I have to look more closely to on how neighbors are selected specially in large search spaces. Fortunately there is so much literature around that it can be of help.
And now this brings me, to my other point of interest, my 'wishlist', which is not that much related to lottery.
First of all it would be nice if you could make the covering functionality of WG to work in parallel with other pcs, but I am not even sure if this is feasible with the algs u use.
Secondly and more importantly, it would be great to have an api exposed so the engine can be used to areas other than lottery. Ofcourse with commercial license attached to that etc..
There are many possibilities on how an api can be used, from automating engine tasks by scripting, or used as benchmark for other algs, to even using the engine to areas like cryptography, software testing etc
Talking about scripts it would be nice if you could add a plugin script engine to wg, for example, say LUA, but this is just the icing to the cake :)
Of course all the above are easy to say but hard to achieve and I am sure u have already thought of them as well.
Btw these suggestions/ideas are relevant for GAT as well.
Thanks
runes

Post Reply

Who is online

Users browsing this forum: No registered users and 14 guests