I want to do a another optimization cycle. Last cycle best settings are temp 50, shift 15% and div 2. We need a new cipher that is about as difficult as jarlve_p20_5nulls_3skips but preferably something that we can run with 10,000 to 20,000 hc iterations to keep it manageable.
Test the following:
– Temp 40, 50 and 60
– Shift 12.5%, 15% and 17.5%
– Div 1.75, 2 and 2.25
– Plus other experimental changes that we want to try. You have suggested things here and there.
After the cycle we can back test the changes to jarlve_p20_5nulls_3skips and smokie_p20_5nulls_5skips. If favorable I like would to attack the 340 again up to 12 nulls & skips.
O.k., that sounds fine to me. Except I am not sure if trying Shift 12.5%, 15% and 17.5% will make that much of a difference.
Let’s say we have 10,000 hc, and eight nulls. 10,000 / 8 = roughly 1,250 iterations per null. The difference between 17.5% and 15% would only be 31 iterations per null, and those would be spread out throughout the 10,000 hc. It is supposed to be for fine tuning, and that doesn’t seem like much of a difference considering that if P = 20, there are 20 possible positions to choose from.
Some other possible approaches would be to run 85 continuous hc with random number = 1 to 340, then 15 continuous hc with random number = current position – Div to current position + Div. Maybe that way a more focused short period of fine tuning for all positions at once, in a cycle.
Or, 8,500 continuous hc and 1,500 continuous hc which I don’t think would work.
Try smokie P20 4/4 nulls(20,58,96,134) skips(219,249,279,309) if you want, it is 340 and I showed the nulls yellow and symbols to the left and right of the skip positions in blue. The idea was to find out if the skips had a difficult time pulling away from the nulls in the top half, or vice versa.
O.k., that sounds fine to me. Except I am not sure if trying Shift 12.5%, 15% and 17.5% will make that much of a difference.
10% and 20% then.
Let’s say we have 10,000 hc, and eight nulls. 10,000 / 8 = roughly 1,250 iterations per null. The difference between 17.5% and 15% would only be 31 iterations per null, and those would be spread out throughout the 10,000 hc. It is supposed to be for fine tuning, and that doesn’t seem like much of a difference considering that if P = 20, there are 20 possible positions to choose from.
Experimental change: a range from the start shift% value to the end shift% value that is spread out over the number of hc iterations.
Example: start shift% = 1, end shift% = 10, hc iterations = 20.
Hc iteration #1: 1% shift
Hc iteration #2: 1.5% shift
Hc iteration #3: 2% shift
Hc iteration #4: 2.5% shift
…
Hc iteration #20: 10% shift
Some other possible approaches would be to run 85 continuous hc with random number = 1 to 340, then 15 continuous hc with random number = current position – Div to current position + Div. Maybe that way a more focused short period of fine tuning for all positions at once, in a cycle.
Or, 8,500 continuous hc and 1,500 continuous hc which I don’t think would work.
Sorry. I do not grasp your idea.
Try smokie P20 4/4 nulls(20,58,96,134) skips(219,249,279,309) if you want, it is 340 and I showed the nulls yellow and symbols to the left and right of the skip positions in blue. The idea was to find out if the skips had a difficult time pulling away from the nulls in the top half, or vice versa.
Okay.
I mean, sometimes the program has a lot of the nulls and skips within 10 positions of their actual positions, but there is no solve. I thought just do the shift iterations all together in chunks. If the settings are 15 % and 2, then maybe 85 continuous iterations with no shift at all, then 15 iterations with all shift. Or, a different chunk size, like 425 continuous iterations with no shift at all, then 75 continuous iterations with all shift. Still the same ratio, but done in chunks. Similar to the sub restart idea, since the sub restarts often cause a solve.
I mean, sometimes the program has a lot of the nulls and skips within 10 positions of their actual positions, but there is no solve. I thought just do the shift iterations all together in chunks. If the settings are 15 % and 2, then maybe 85 continuous iterations with no shift at all, then 15 iterations with all shift. Or, a different chunk size, like 425 continuous iterations with no shift at all, then 75 continuous iterations with all shift. Still the same ratio, but done in chunks. Similar to the sub restart idea, since the sub restarts often cause a solve.
Sure, we could test that.
If
±=N
³=Y
Â=ZDoes any particular pattern become noticeable
in the cipher?
?
I made the entries for the new optimization cycle in the spreadsheet and will first determine solve rates for jarlve_p20_5nulls_3skips and smokie_p20_5nulls_5skips for back testing. That may take a while. Currently jarlve_p20_5nulls_3skips has a 14.36% solve rate with 20,000 hc iterations. Previously it was 11% with 40,000 hc iterations. It seems that all these small fixes have gone a long way!
After making an improvement to the substitution solver the solve rate for jarlve_p20_5nulls_3skips is currently at 22.3% while also running a little bit faster!
Wow that is great! Keep up the good work!
Wow that is great! Keep up the good work!
It has come down to 19,7% since but it still seems a good improvement.
Jarlve,
How is the testing going by the way?
Going slow. It was over 30 degrees in my apartment for a while and decided to not do any runs with that heat. It is better now and I resumed the run on smokie 5/5 a couple of days ago. Doing the last 800 restarts to make 2000. It is about half way. After that is done I want to find a cipher which has about a 15% solve rate with 10,000 HC. Then there are 3 test ciphers, easy/medium/hard. Changes could be tested much more quickly on the easy cipher, and if an improvement, validated on jarlve 5/3 and smokie 5/5.
I have been taking a break and reading some classical literature. Starting to get back to the idea of testing, and although I probably can’t make much of a contribution with the more difficult cryptograms, I may test with some easier ones so that I have a better understanding of what you are doing. There are still some ideas that can be explored over here with what I have.
What is the difference between increasing the general iterations and the general iterations factor?
Have you experimented with the general hill climber iterations factor and / or do we need that setting?
I like the over skip setting. I put the program to work on a 4/0 and it got a solve every few seconds. Felt good.
I live near Kansas City, right in the middle of the continent where the pioneers gathered in large numbers to make final preparations to travel by covered wagon to the west coast. It is also where anti slavery people from the north east settled, which was controversial leading up to the Civil War. It is very humid here in the spring and summer, and nice when the weather is cool but horrible when really hot. It generally rains a lot in the spring, and I have only had to water my lawn one time ever. Very lush green hilly terrain and lots of wildlife. I have never had to wash my car ha ha.
nice post smokie .. 
It generally rains a lot in the spring, and I have only had to water my lawn one time ever. Very lush green hilly terrain and lots of wildlife.
Sounds like a pretty great area and ideal conditions for mushroom picking if spring is warm enough (most people only look for mushrooms in autumn…a mistake). We have wild animals here too, but it is certainly not as widespread as in your area. We mainly have deer, stags, wild boars, foxes and hares.
The heat wave in Germany ended yesterday and now the temperatures have fallen by up to ten degrees. And it rained for the first time in weeks. Finally one can think clearly again and also go for a walk!
Translated with http://www.DeepL.com/Translator
