Jump to content


Photo

Zerack's DK Gear Optimizer v2.6.9 (03-MAR-2010)


  • This topic is locked This topic is locked
213 replies to this topic

#41 norg

norg

    King Hippo

  • Members
  • 589 posts

Posted 28 August 2009 - 01:44 PM

Yeah good point -- I hadn't considered that, while we're manually editing Hit to be worth less than Strength, the orange gem will always be valued higher than they yellow. I'll ignore the yellows until such a time as the hit/str issue gets resolved.

#42 dr_AllCOM3

dr_AllCOM3

    Great Tiger

  • Members
  • 985 posts

Posted 28 August 2009 - 04:30 PM

Nothing gets saved in 2.1.1 for me. Well, it does get saved. It's just not being loaded.

#43 Zerack

Zerack

    Piston Honda

  • Members
  • 219 posts

Posted 28 August 2009 - 04:37 PM

Nothing gets saved in 2.1.1 for me. Well, it does get saved. It's just not being loaded.


The only thing I can think is that it has something to do with the . vs , decimal point debate. Are you using ',' as your decimal point by any chance? I've no problem saving at all on my US English system.

#44 dr_AllCOM3

dr_AllCOM3

    Great Tiger

  • Members
  • 985 posts

Posted 28 August 2009 - 04:54 PM

The only thing I can think is that it has something to do with the . vs , decimal point debate. Are you using ',' as your decimal point by any chance? I've no problem saving at all on my US English system.


It's a fresh install and the only thing I've changed for testing is the race and a profession.
Edit: Seems to work, when I manually set region to 1.

#45 Zerack

Zerack

    Piston Honda

  • Members
  • 219 posts

Posted 28 August 2009 - 05:01 PM

It's a fresh install and the only thing I've changed for testing is the race and a profession.
Edit: Seems to work, when I manually set region to 1.


Tracked down and fixed. Should work fine in the next release.

Edit: It's tough for me to test the region stuff since I don't have access to a PC that is of a different persuasion :-P

#46 dr_AllCOM3

dr_AllCOM3

    Great Tiger

  • Members
  • 985 posts

Posted 28 August 2009 - 05:22 PM

Forcing Hit or Expertise caps will occasionally cause the program to crash.

It's more like always :).
Your optimizer should tend to go slightly above Exp/Hit cap instead of below.

#47 norg

norg

    King Hippo

  • Members
  • 589 posts

Posted 28 August 2009 - 06:39 PM

So going back the the ArPen issue: whose heads do we need to bang together here in order to get a better model for this stat? It used to be Methods doing the stat weights a while back -- is he still around? If not, who are the math guys around here who can properly get involved with this? Döc? Anyone else? It'd be good to try and get it sorted.

#48 Canadianloaf

Canadianloaf

    Glass Joe

  • Members
  • 18 posts

Posted 28 August 2009 - 10:10 PM

I mentioned this on the BiS thread, but I think it was lost in the midst of specific gear discussions:

Wouldn't crit rating have a different stat weight depending on the presence of 4 pc T9 bonus? Without the bonus, all crit would be valued less than it would with the bonus. I'm not sure that I'm smart enough to model it, but does anyone else share my view that a single weight for crit will not work?

Edit: I suppose this is dealt with by using the T9 bonus EP to represent the added value of crit, but then that bonus should have a sliding value dependent on your total crit rating, rather than a fixed EP number.

#49 Canadianloaf

Canadianloaf

    Glass Joe

  • Members
  • 18 posts

Posted 28 August 2009 - 10:15 PM

Zerack, I think there's an error in the stats for Shoulderplates of Trembling Rage. It should be 107 Str, not 115 (that's the Sta value).

#50 elf-boy

elf-boy

    Glass Joe

  • Members
  • 20 posts

Posted 29 August 2009 - 04:42 AM

I just ran the optimizer a few times. I used the Frost DW preset stats. I also ran using the stats from the first post on the revenge of the offhand thread. I found that unless I forced melee/spell hit it equipped me with ZERO +hit. As well as very little Expertise.

Is there anything I can PM/email you to help figure this out? Or was I doing something not quite right?

I also found the "Band of Draconic Guile" was not listed.

I also have a question about the weapon selections. It chose 2X bloodcrush cudgel over 2x Aledar's Battlestar. Is weapon speed wighting set up or enterable anywhere?

I am also curious if I can do anything to speed up the calculations? Is there a way to make the app multi-thread aware?

#51 radunio

radunio

    Glass Joe

  • Members
  • 1 posts

Posted 31 August 2009 - 12:20 PM

There is something realy wrong with trinkets calculations (tried with blood and DW frost wieghts).

Trinkets that I have atm are:
Mjolnir Runestone
Bandit' Insignia
Fury of Five Flights
DC:G
Death's Choice
Dark Matter

With gear I'm getting hit / exp capped which is OK but for trinkets im getting weird setups like:
- for blood - Fury of Five Flights and Bandit's Insignia
- for frost - Bandit's Insignia and Death's Choice

Meabe I'm wrong but shouldn't it be Death's Choice and DC:G for every build atm?

#52 EvilNuff

EvilNuff

    Glass Joe

  • Members
  • 24 posts

Posted 31 August 2009 - 11:01 PM

...I also have a question about the weapon selections. It chose 2X bloodcrush cudgel over 2x Aledar's Battlestar. Is weapon speed wighting set up or enterable anywhere?

I am also curious if I can do anything to speed up the calculations? Is there a way to make the app multi-thread aware?


If you look at the gear.txt file that accompanies the application that's where he appears to pull the item data. In there I cannot find any weapon speeds which leads me to believe that his app does not handle them. Weapon speed is important of course, so this leads me to question the applications other hidden logic?

Opening up the source would help answer some of these questions if the author is willing to do so.

#53 norg

norg

    King Hippo

  • Members
  • 589 posts

Posted 01 September 2009 - 02:46 PM

I don't think the program makes allowances for weapon speed. For the BiS lists I've just been including the slow weapons in my calculations and not bothering with the fast ones. Slow/slow is right for DW Frost, yeah?

#54 deathsythe

deathsythe

    Glass Joe

  • Members
  • 7 posts

Posted 01 September 2009 - 08:30 PM

To start out I love the program so thanks a lot for the hard work you've put into this. The question I have is when running the program after selecting a decent amount of items (both current and 10 man normal items from Ulduar and ToC) I continue to get a "Run Time Error '7' Out of Memory" message. When i watch the process in Task Manager the memory usage just goes threw the roof and then times out. Was wondering if you had any thoughts on how to fix this.

#55 Guest_advocate70_*

Guest_advocate70_*
  • Guests

Posted 01 September 2009 - 08:48 PM

Bug in latest version: The latest version lacks source info and ilevels for all of the trinkets, and it therefore gives an error that "all stat values and ilevels must be numeric" in order to continue.

#56 Orothar

Orothar

    Von Kaiser

  • Members
  • 46 posts

Posted 02 September 2009 - 01:17 AM

To start out I love the program so thanks a lot for the hard work you've put into this. The question I have is when running the program after selecting a decent amount of items (both current and 10 man normal items from Ulduar and ToC) I continue to get a "Run Time Error '7' Out of Memory" message. When i watch the process in Task Manager the memory usage just goes threw the roof and then times out. Was wondering if you had any thoughts on how to fix this.


I encountered the same problem when selecting a large amout of gear (5 pieces per slot), mem usage goes through the roof and the app crashes. When i select just 1 - 2 pieces per slot, it can get all possible gearsets pretty quickly, but then hangs at the "gemming gearsets"calculations. The mem usage at that point is fine, it just doesnt seem to do anything. I've tried selecting large amounts of possible gems and just gem per color, nothing seemed to help.

And btw, thnx for making a application like this.

#57 Anathem

Anathem

    Glass Joe

  • Members
  • 14 posts

Posted 02 September 2009 - 04:13 AM

Zerack, unless you plan on selling this software, I don't see any reason to keep it closed source. I personally would be interested in contributing to the project if you were to host it on Secure source code hosting and collaborative development - GitHub or a similar code collab site.

#58 Zerack

Zerack

    Piston Honda

  • Members
  • 219 posts

Posted 02 September 2009 - 03:19 PM

I'll be posting a new version later today with some updates, but let me try and address some of the issues that have come up:

@dr_AllCOM3 - As discussed, there was a problem where the region wasn't being set to 1 after it did the region check. I think this should work in the new version without trouble, but my region checking is a hack at the moment that I'm not happy with. Let me know in the future if it continues to cause trouble. Regarding the optimizer tending to go over the hitcaps rather than under - that's not the nature of an optimization like this. Tendency would be to end up slightly under, which is technically optimal from a numeric point of view. I'm slowly getting around to fixing the hitcap / expcap forcing options, so that should suffice when they are available.

@elf-boy - If you are still having the issue where it generates sets with no +hit, please upload and PM me a text dump of the results of a calculation. That contains information I can use to determine what exactly is going on. Regarding speeding up the calculations, I'm working on it. I'm talking with someone about helping me make the program multi-thread aware which should result in good time savings, but it's not ready yet. Hang tight.

@radunio - The trinket issue was fixed. It is likely you are using the release that briefly contained a trinket error. Download again and reinstall and you should have no problems. Let me know if the issue persists.

@EvilNuff - The current version does not support weapon speeds, but I've added it in the next version. This should assuage some of your fears that the program is doing something mystical and / or arcane to derive set values. I assure you it is not.

@advocate70 - I'm unable to reproduce this error either on my development machine or on my two personal machines running the most recent code. Is anyone else having this problem?

@ deathsythe / orathar - I'll look into dealing with this. Can you give me an idea of how large the program is growing before crashing? So far I've optimized primarily for speed, but there are ways that I can make the program use less memory. In fact if speed wasn't an issue I could keep memory usage ridiculously low, but I don't think anyone would like those calculation times.

@Anathem - I'll consider making the source public as I've no plans to attempt to profit from this. My hesitation is that I do not by any means know what I am doing, and I'm fairly certain I'd be the laughing stock of the EJ community if anyone halfway competent looked at the code :) Regardless if there's something specific you are interested in helping with please let me know and perhaps we can get something set up.

I think that covers most of what was said in the thread since I last checked in. There's been some discussion about weights so I'm going to double check that they are all up to date with the respective threads before posting the next version, but hopefully that won't take long and I can get up a new version this evening. With apologies, Norg, this version will support weapon speed so you may have to do some recalculating. Please don't hurt me.

#59 Anathem

Anathem

    Glass Joe

  • Members
  • 14 posts

Posted 02 September 2009 - 04:12 PM

The bulk of the calculation time seems to be spent searching for an optimal gem set. I want to take a look at that code and see if I can rewrite it better, for myself only, if not for public release.

A few ideas:
1. You're probably finding several top gear/gem sets and displaying the single "best" one -- why not save the top n sets and allow the user to pick a slightly non-optimal solution at his preference?

2. Settling for near-optimal solutions allows massively reduced search time in large-search-space optimization problems.

3. Threading support. Gemming multiple gearsets doesn't seem to be a necessarily iterative process, we might be able to speed it up significantly on multi-core processors -- of course I don't know how you've implemented this already.

These are just a few off the top of my head, it's hard to say what I could do without seeing what's been done. I really would like to at least browse the code if you're willing. I'm PMing you my email address.

#60 dr_AllCOM3

dr_AllCOM3

    Great Tiger

  • Members
  • 985 posts

Posted 02 September 2009 - 05:04 PM

I'll consider making the source public as I've no plans to attempt to profit from this. My hesitation is that I do not by any means know what I am doing, and I'm fairly certain I'd be the laughing stock of the EJ community if anyone halfway competent looked at the code :) Regardless if there's something specific you are interested in helping with please let me know and perhaps we can get something set up.


You won't be laughed at ;). You can be glad, if there are five people who at least understand the code. If helps a lot at bugfixing though, judging from my experiences from the sim. Afabar and me always find stuff to correct.




0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users