The 6 thread test entries have been created but please do no yet start work on them yet.
It is looking like I am not getting as many solves over 22222 as you. Any ideas why?
The solution is to normalize the temperature of the hill climber by the amount of threads. I am testing this change and will get back with the results within a few days. For now the initial results suggest that this temperature normalization is exactly the right fix.
It is looking like I am not getting as many solves over 22222 as you. Any ideas why?
When I downloaded the latest versions of the program, the ngram and other folders were not included, so I used ones from a previous version. But it is still on performance mode 5 grams. Or perhaps I have a setting incorrect. Or perhaps test a little while longer.
Yes, only the program file was included with that download. It surely is 100% related to the number of threads we are using. You do have nulls & skips on 8 right? Lowering or increasing the temperature could fix this problem for you. I could run multiple instances of AZdecrypt with 6 threads to simulate your computer. I will make some new entries for this in the spreadsheet. Sorry that the optimization is taking so long!
I have the same settings as several posts above, except for minor changes due to what trial I am working on. I have it on 8 nulls and skips. I confess I am not using the Broadwell version, apprehensive and thought it only affected speed in a minor way. I don’t understand why speed and temperature could affect the number of solves over 22222 with the same number of iterations.
Optimization taking a while no problem for me. I apologize if I have caused any delay.
I have the same settings as several posts above, except for minor changes due to what trial I am working on. I have it on 8 nulls and skips. I confess I am not using the Broadwell version, apprehensive and thought it only affected speed in a minor way. I don’t understand why speed and temperature could affect the number of solves over 22222 with the same number of iterations.
Broadwell or not, the version does not matter. Nor the speed of our computers. The problem is (hopefully) that I did not normalize the temperature over the number of CPU threads, and that is were we differ. Working on it.
Optimization taking a while no problem for me. I apologize if I have caused any delay.
You have not caused any delay.
The problem is (hopefully) that I did not normalize the temperature over the number of CPU threads, and that is were we differ.
That was exactly the problem and it is fixed now. Will upload a new version in about one week and by then the optimization cycle should be finished also.
O.k., Jarlve: I will look forward to it and to further optimization and eventual attack. Thanks.
O.k., Jarlve: I will look forward to it and to further optimization and eventual attack. Thanks.
A cipher with 8 nulls did not solve with 500 restarts. The hill climber selects either a null or a skip with a 50/50% chance/bias. The cipher solved after introducing a random selection bias in the range from 0/100% to 100/0% per restart. I am becoming quite confident with the nulls & skips solver. Pretty sure that we can go all the way up to 12 nulls & skips. We should continue on the 340, what do you think smokie?
HI Jarlve: I will look at the spreadsheet tomorrow, as I just walked in the door. Is there a new version of the solver? Should we do more testing, like with one or more of 12/12 first? I have plenty of patience, and could easily take the time to work with more test ciphers if you could. When moving forward with the 340, I say start with P15 because 12/12 would be within range. Thanks.
I feel confident enough with the hill climber to continue on the 340 if you are okay with that. And we may want to revisit 5 to 8 nulls & skips because a cipher with 8 nulls was unable to solve with versions prior to 5/20/2018.
5/20/2018: https://drive.google.com/open?id=1ZWBlP … hMx5gRTFC8
O.k., I will make a spreadsheet, similar to the first one, except without so many columns with Xs. Then you can customize, etc. and we can attack. I have some stuff to get caught up on around the house and then will work on it today.
O.k., I will make a spreadsheet, similar to the first one, except without so many columns with Xs. Then you can customize, etc. and we can attack. I have some stuff to get caught up on around the house and then will work on it today.
Okay, great.
I cleaned up the original 340 spreadsheet ready for any customization you may want to make. I saved your notes at the top, but they may need to be slightly updated to make sense. We can always add new trial numbers if you want to redo some of the work already done except with the new version. I will download and begin work probably early tomorrow morning.
I cleaned up the original 340 spreadsheet ready for any customization you may want to make. I saved your notes at the top, but they may need to be slightly updated to make sense. We can always add new trial numbers if you want to redo some of the work already done except with the new version. I will download and begin work probably early tomorrow morning.
Okay. It is good. I am adding extra restarts from 5 to 8 nulls & skips first before starting the heavy work.
I have been experimenting with the program and the message that I made on page 96 of this thread, which has 17 nulls. Set to 100k null skip restarts. I can see what you mean about how there is a difference between nulls and skips. If the program tries a skip, it inserts a null, which increases multiplicity, like you said, and increases the score. The highest scores are with mostly if not all skips. Are you still tinkering with the program? I noticed that you removed the null skip bias option. How are things going over there? Do we need changes to the spreadsheet for the options?
I have been experimenting with the program and the message that I made on page 96 of this thread, which has 17 nulls. Set to 100k null skip restarts. I can see what you mean about how there is a difference between nulls and skips. If the program tries a skip, it inserts a null, which increases multiplicity, like you said, and increases the score. The highest scores are with mostly if not all skips. Are you still tinkering with the program? I noticed that you removed the null skip bias option. How are things going over there? Do we need changes to the spreadsheet for the options?
The hill climber now uses a random bias in the range 0 to 100. This bias is randomly rolled with every restart. Say it rolls a bias of 50, then the hill climber has a 50% chance to select either a null or a skip. If rolled at 10, the hill climber has a 10% chance to select a null and 90% chance to select a skip. If rolled at 97, the hill climber has a 97% chance to select a null and a 3% chance to select a skip.
In the optimization spreadsheet there is an entry with a 0% solve rate percentage. That is from a cipher with 8 nulls which proved unsolvable with the standard bias of 50 (50%/50% chance null or skip). When I changed the bias to be random like explained in the above paragraph this cipher improved to a 9% solve rate percentage.
I am currently not tinkering with the hill climber. I removed the bias option because the hill climber now determines it randomly. The spreadsheets are fine to me.
