|
|
Xarfax111
Bad-mannered
Supreme Hero
The last hero standing
|
posted July 10, 2007 12:39 AM |
|
|
Quote:
Very nice work u did here dimis, as usual! +QP applied of course.
What I find kinda interesting is the "relatively" low percentage for logistics (around 35%). It only takes 9th place, even behind pathfinding. So it isn't just an imagination I always have problems getting log with Crag or Gurni....
mmh..i think you come to the wrong conclusion. 9th place isnt that bad. Offense given and Wisdom + Magic Schools are in another rotation, plus Might Skills are prefered into the first up to 5 Slots, the probability of Logistics is very high. Plus Hack doenst have a wasted second skill.
____________
|
|
Xarfax111
Bad-mannered
Supreme Hero
The last hero standing
|
|
posted July 10, 2007 12:46 AM |
|
|
PLUS there is a way to get always Logistics with Hack! (Similar like gaining Earth all the time)
(but i wont tell ...ma precious ..its ma precious
Of course ANSA n stuff will get QPs, but will NEVER help you to find things like that. So go on ANSA ...hehe
____________
|
|
dimis
Responsible
Supreme Hero
Digitally signed by FoG
|
|
posted July 10, 2007 01:00 AM |
|
|
So, are you suggesting that there is another exception (apart from Wisdom and Magic)?
____________
The empty set
|
|
Binabik
Responsible
Legendary Hero
|
|
posted July 10, 2007 01:12 AM |
|
Edited by Binabik at 01:13, 10 Jul 2007.
|
I don't think he will tell you
I don't think there are any more special cases like wisdom and magic schools, at least not with individual skills. I suspect there is a way to select skills in a way that guarantees getting certain skills.
I still think there might be skill groups, with a certain number of skills from each group guaranteed. For example you might be guaranteed that 2 map skills are offered, 3 battle might skills, and 1 battle non-might skill (like tactics). I haven't found a pattern to it statistically, but from observation it seems like the AI attempts to make a well rounded hero with a combination of map, battle, magic, etc skills.
____________
|
|
dimis
Responsible
Supreme Hero
Digitally signed by FoG
|
|
posted July 10, 2007 01:28 AM |
|
Edited by dimis at 01:29, 10 Jul 2007.
|
Quote:
I don't think he will tell you
It's a yes / no answer. Nothing more.
You are not convincing me both.
First, I need indications. These come from tests (and observation).
So, I 'll use some results from my own tests.
Now, Crag-Hack on ANSA (btw "Always Right" was after all a better name) seems to be very close to what the model predicts regarding Logistics. In fact, if you look both of my tables here you will realize that Logistics is among the skills that boost our level of confidence regarding the model.
So, to my eyes, this is just another theory. No offense, but I have proofs that indicate that we have good oracle for Logistics, while I don't see an example to support the opposite.
However, if getting Logistics every time means more than 90-95% of the time for example, be patient and we'll talk about it soon. We'll work with more delicate user-defined strategies, in the worst case by the end of this week.
____________
The empty set
|
|
Binabik
Responsible
Legendary Hero
|
|
posted July 10, 2007 01:57 AM |
|
|
Maybe I should clarify. I don't know if I still have the data, but I did testing on the skill groups theory. When I said it was through observation, I meant I didn't compare expected results with actual results. But without going through the calculations, it just "seemed" like I was getting some groups of skills much more than would be expected.
For example there might be a group of 5-6 skills, and I got 1-2 skills from that group 24 out of 25 times. If I found a group that was 25 out of 25 times I would have posted it. The remarkable thing about these groups was that some of them had very low probability as individual skills, but as a group they had very high probability.
Another way of looking at it is consider 2 groups with 5 skills in each group (this is just an example). In one group the individual skill values add up to 10 (average of 2), and in the second group the skills add up to 20 (average of 4). But something from the first group was offered almost as many times as something from the second group *IF* nothing from that group had already been taken.
Anyway, I'm not trying to be pushy on this. The way you are doing it is fine. It needs to be done systematially rather than chasing every theory that comes along.
____________
|
|
dimis
Responsible
Supreme Hero
Digitally signed by FoG
|
|
posted July 10, 2007 02:56 AM |
|
Edited by dimis at 03:45, 10 Jul 2007.
|
I hate to post last, but I did a small test.
Again Crag, with ansaExtended. I created a group of 8 skills with skills having the smallest possible weights. So I filled first five slots with skills weighing 1 each, then the last 3 with skills weighing 2 each [of course I didn't choose Wisdom, since that would be cheating Instead I started filling the slots in a Left->Right manner]. The selections.txt file that I used was:
Quote:
DIPLOMACY NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL
FIRST_AID NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL
INTELLIGENCE NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL
SORCERY NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL
SCHOLAR NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL
EAGLE_EYE NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL
ESTATES NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL
NAVIGATION NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL NULL_SKILL
The outcome (which can be verified with an execution on everyone's pc):Quote:
Requested probability: 51.6269 %.
Doesn't this seem large enough for 8 skills weighing 11 in total (1.375 on average)? Moreover, we know for a fact that the random generator is somehow flawed or we don't know everything about the model. But generators used in computers are flawed by definition! Just check what is the result of Monte Carlo method in AlwaysLeft or AlwaysRight policy for any hero. What needs to be checked: Skills with equal weight that are neither a Magic School nor Wisdom, should have EXACTLY the same probability (Magic Schools with equal weight similar). However, things do not work that way, but that's the best we can do! I recognize though, that Heroes' generator is very absurd and has more "problems" than the slight discrepancies I describe above. On the other hand, these slight discrepancies can generate very large deviations on small tests.
Example: We know that most likely no more than 250 different skill sequences can result when following ANSA for Crag-Hack. Try to simulate Crag-Hack on ANSA with MonteCarlo for exactly 250 (or less) episodes. The results seem flawed. But they aren't! They try to capture something very-big with a very small number of iterations. It's a method that is almost doomed to fail by definition. And this is what we see the H-generator (Heroes-generator) does. I hope I make sense in this example.
Moreover, it would be nice if Tournament Edition was made through Reverse Engineering. This would reveal us the mechanism easily. However, T.E. is based on WoG and scripts allowed by WoG (at least this is what I 've understood without reading much about; as of WoG I am pretty much ignorant: I had never played up until T.E.).
Finally, speaking of Crag and T.E. I don't like that Crag starts with Basic Offense. This results in slower development from day-1 and lasts for many levels, since Crag should be 1 level ahead of its opponent in order to have the same amount of skills at expert level.
____________
The empty set
|
|
angelito
Honorable
Undefeatable Hero
proud father of a princess
|
|
posted July 10, 2007 11:08 AM |
|
|
Quote:
Quote:
Very nice work u did here dimis, as usual! +QP applied of course.
What I find kinda interesting is the "relatively" low percentage for logistics (around 35%). It only takes 9th place, even behind pathfinding. So it isn't just an imagination I always have problems getting log with Crag or Gurni....
mmh..i think you come to the wrong conclusion. 9th place isnt that bad. Offense given and Wisdom + Magic Schools are in another rotation, plus Might Skills are prefered into the first up to 5 Slots, the probability of Logistics is very high. Plus Hack doenst have a wasted second skill.
I was mostly refering to 4 skills which have a higher or nearly same possibility than logistics (Archery, Ballistics, Pathfinding, Scouting). I would have bet a barbarian has higher chance getting logistics than scouting. But seems I am wrong...
____________
Better judged by 12 than carried by 6.
|
|
dimis
Responsible
Supreme Hero
Digitally signed by FoG
|
|
posted July 10, 2007 10:54 PM |
|
|
I am placing a reminder here:
Archery and Logistics ought to have the same possibility on Crag when following AlwaysLeft. My first post with Crag_Hack on AlwaysLeft was just my first trial and nothing more. Perhaps 1,000,000 iterations is not large enough to assure convergence for 1 decimal digit (actually 3, since x%=x/100) on the percentages. (offtopic: Actually, I think it is on the verge of being accurate at that decimal point but as I said, I haven't studied the error which in turn can give a bound on the number of iterations needed by a good generator./offtopic) Anyway, I ran a longer experiment with Crag_Hack for 100,000,000 episodes this time. Here are the results and respective seed used for computations:
CRAG_HACK
~~~~~~~~~
AIR MAGIC : 67.2397 %.
ARCHERY : 30.8536 %.
ARMORER : 27.0397 %.
ARTILLERY : 34.4772 %.
BALLISTICS : 34.4801 %.
DIPLOMACY : 5.00611 %.
EAGLE EYE : 9.80839 %.
EARTH MAGIC : 67.2636 %.
ESTATES : 9.80656 %.
FIRE MAGIC : 55.842 %.
FIRST AID : 5.00015 %.
INTELLIGENCE: 5.00468 %.
LEADERSHIP : 23.0224 %.
LEARNING : 18.8251 %.
LOGISTICS : 30.8636 %.
LUCK : 14.4136 %.
MYSTICISM : 14.4175 %.
NAVIGATION : 9.80777 %.
NECROMANCY : 0 %.
OFFENSE : 100 %.
PATHFINDING : 34.4773 %.
RESISTANCE : 27.0398 %.
SCHOLAR : 5.00217 %.
SCOUTING : 34.4759 %.
SORCERY : 5.00341 %.
TACTICS : 34.4728 %.
WATER MAGIC : 0 %.
WISDOM : 96.3568 %.
Duration : 2356.3 seconds.
Seed : 1184060638
Notice that now Logistics has higher chance than Archery. But as I said, these two have to be the same. And they are trying. Eventually, they will converge to the same number. Notice also that Luck and Mysticism are closer to their true number (and in between), etc...
____________
The empty set
|
|
dimis
Responsible
Supreme Hero
Digitally signed by FoG
|
posted July 11, 2007 09:01 PM |
|
Edited by angelito at 20:27, 11 Nov 2009.
|
internals_mc: Evaluation of user's policy with Monte Carlo methods
Current Version: 2.0
Source Code and win32 Executable
Source code can be obtained here.
(updated: April 9, 2008 / Reason: incorrect version previously uploaded due to an error in the makefile) win32 executable can be obtained here.@dimi: Check reminder on home log below.
Supported Algorithms
ANSA or AR: These two identifiers indicate the "Always Right" strategy, i.e. always choose the Right offer when advancing a hero. In cases where there is only one option, then obviously that one option is selected.
AL: This is the "Always Left" policy, i.e. the symmetric policy than that defined above.
ALTP: This is the "Always Left-Then Preference" policy, i.e. whenever there is a skill for upgrade, the hero upgrades that skill, but if two new skills are offered, then the hero selects the one that prefers the most. The filename that is read in order to determine user's preference is called altp_prefs.txt.
Note also that heroes that start with 2 skills always choose the left offer until both of them are at expert level (i.e. in this case the preferences don't count). If you feel that this should change, just let me know with a post of yours here on thread.
SPOU: This is the "Seek Preference-Otherwise Upgrade" policy. User's preferences are read from file spou_prefs.txt. The idea is to get ANY of the top 8 skills whenever these are offered, otherwise upgrade any existing skills. When two skills are presented for upgrade the one with higher preference is picked first.
However, there are cases where the hero can be really unfortunate and be offered two skills for upgrade and none is on top 8 list. In this case the one with the higher preference is selected. However, now the hero has virtually one "unwanted" skill, hence next time he has to decide for getting a skill, he will search ONLY in top 7 list.
Similarly, with 2 unwanted skills, a new skill is selected if and only if this appears in top 6 list, and so on ... (important: Please also read the log in the end of post to get some more info on policy.)
Printing Methods
DEPRECATED, NEAT, FANCY
By default FANCY method is used.
Command Line Arguments
On command line the program expects 4 or 5 arguments.
1st: The hero class.
2nd: The hero name.
3rd: The policy followed by user.
4th: The number of episodes.
5th (Optional): The output method for the results. If not specified, FANCY is used by default.
You can get a detailed help on command line by giving:
Quote:
internals_mc -help
Sample Execution
Assume you want to experiment with Crag-Hack in "altp" policy for 1,000,000 episodes. Then it suffices to execute the following:
Quote:
internals_mc barbarian crag_hack altp 1000000
Of course, do not forget to have the "altp_prefs.txt" file in the same directory. In case altp_prefs.txt has structure as in this file, the results should be similar to these:Quote:
CRAG_HACK
~~~~~~~~~
AIR MAGIC |====================================> 73.33 %
ARCHERY |============> 26.58 %
ARMORER |=======================> 47.59 %
ARTILLERY |================> 34.63 %
BALLISTICS |===========> 24.09 %
DIPLOMACY |=> 3.08 %
EAGLE EYE |==> 5.66 %
EARTH MAGIC |=====================================> 76.99 %
ESTATES |==> 5.15 %
FIRE MAGIC |=======================> 47.91 %
FIRST AID |> 2.48 %
INTELLIGENCE |> 2.35 %
LEADERSHIP |=================> 35.11 %
LEARNING |===> 7.47 %
LOGISTICS |==============================> 62.77 %
LUCK |==========> 21.62 %
MYSTICISM |=> 4.52 %
NAVIGATION |> 2.54 %
NECROMANCY | 0.00 %
OFFENSE |=================================================> 100.00 %
PATHFINDING |===========================> 55.13 %
RESISTANCE |====================> 42.67 %
SCHOLAR |> 1.14 %
SCOUTING |> 1.01 %
SORCERY | 0.00 %
TACTICS |===============================> 63.15 %
WATER MAGIC | 0.00 %
WISDOM |==========================> 53.00 %
Duration : 13.02 seconds.
Seed : 1184918425
Similarly, a
spou_prefs.txt file with same contents as the
altp_prefs.txt file above will produce an output similar to the following:Quote:
> internals_mc barbarian crag_hack spou 1000000
CRAG_HACK
~~~~~~~~~
AIR MAGIC |=============================================> 91.98 %
ARCHERY |======> 14.01 %
ARMORER |==================================> 69.83 %
ARTILLERY |========> 18.82 %
BALLISTICS |=====> 12.47 %
DIPLOMACY |> 1.50 %
EAGLE EYE |> 2.79 %
EARTH MAGIC |==============================================> 93.28 %
ESTATES |> 2.53 %
FIRE MAGIC |=============> 28.58 %
FIRST AID |> 1.20 %
INTELLIGENCE |> 1.16 %
LEADERSHIP |=========> 19.48 %
LEARNING |=> 3.69 %
LOGISTICS |=======================================> 80.54 %
LUCK |=====> 11.56 %
MYSTICISM |> 2.22 %
NAVIGATION |> 1.22 %
NECROMANCY | 0.00 %
OFFENSE |=================================================> 100.00 %
PATHFINDING |=====================================> 75.16 %
RESISTANCE |==========================> 53.16 %
SCHOLAR |> 0.54 %
SCOUTING |> 0.49 %
SORCERY | 0.00 %
TACTICS |========================================> 82.60 %
WATER MAGIC | 0.00 %
WISDOM |===============> 31.18 %
Duration : 13.78 seconds.
Seed : 1184918804
User Preference Files: Structure and Samples
The structure is very simple actually. All you need to do is place on separate lines different skills (any combination of lower case and upper case letters is allowed). Each line should begin with a letter (i.e. the first letter of the skill that is represented in this line) and should finish as soon as the name of the skill finishes.
The higher you put a skill in your file, the higher preference it gets. The parser reads the 28 first lines from the preference files in order to induce ordering. Valid names for skills:
AIR_MAGIC, ARCHERY, ARMORER, ARTILLERY, BALLISTICS, DIPLOMACY, EAGLE_EYE, EARTH MAGIC, ESTATES, FIRE_MAGIC, FIRST_AID, INTELLIGENCE, LEADERSHIP, LEARNING, LOGISTICS, LUCK, MYSTICISM, NAVIGATION, NECROMANCY, OFFENSE, PATHFINDING, RESISTANCE, SCHOLAR, SCOUTING, SORCERY, TACTICS, WATER_MAGIC, WISDOM
Sample files (same contents on both):
altp_prefs.txt
spou_prefs.txt
Smoothing occurrences:
In the cases of AR (ANSA) and AL, the computed empirical probabilities are smoothed in order to minimize the error more. This is probably going to be useful in non-deterministic policies in the future (i.e. you toss a coin to decide whether you pick skill A or skill B; e.g. choose between Eagle Eye and Navigation in a random generated map )
Other command-line options
You can also execute the program with the following options:
-version Presents current version. Sample execution: Quote:
internals_mc.exe -version
-about Presents an "about" message. Sample execution: Quote:
internals_mc.exe -about
Anyway,
Have FUN!
Home Log
Version 1.0
Home Generated: July 11, 2007.
Version 1.1
Released: July 20, 2007.
* You can get a description of the log file on command line with the following command:
Quote:
internals_mc.exe -log
* Fixed bugs when reading preference files. Beats me why it was problematic in version 1.0 but anyway!
* Some optimization enhancements here and there.
* Generation of a mathematical toolbox that allows:
- Smoothing values in AR and AL policies on skills with same weight.
- Different printing methods of the results (this is the optional 5th argument).
* Fixed bug in SPOU policy behavior: Version 1.0 had a weird problem illustrated by the following example. Assume you wanted to compute spou-probabilities for Crag-Hack and on your preference list you placed offense in 8th slot. Then, in the event that you got 1 unwanted skill along leveling-up, the machine was looking on the first 7 slots to see whether it could accept the new skill or not. However, since the hero always possesses offense (starting skill) plus an unwanted skill (i.e. position on file > 8), then it is obvious that the hero can NOT get all 7 skills on the first 7 slots that it is looking at.  Function qualifies_on_top on policies.cpp file performs this qualification for each skill based on the temporary knowledge of acquired skills so far.
Version 2.0
Released : April 7, 2008.
* POSIX threads are implemented in order to speed up computations.
> A beta version not supporting Windows was released on April 3, 2008.
* The executable was created with MS Visual C++ 2005 Express Edition (now 2008 is available).
However, Service Pack 1, as well as SDK
were also installed on the system. Therefore, the actual compiler that created the executable is of version 9.0.
* Conditional compilation works under all platforms.Source code (as well as executable) can be obtained from the links on top.
The previous version (1.1) # for a reason that I don't remember atm # did not compile under Unix(-like) operating systems, here is the same version (1.1) which actually does!
Modifications and thoughts posted on beta version: here.
Finally, the terminal output on the experiments described above can be obtained here.
reminder for me: Fix output of -log AND -about commands. -log has incorrect spacing, and -about obviously doesn't read global constant for version, neither takes time and date of compilation time (-log does so though!).
COMMENTS:
1] Ideas for better names on policies (even if these have already been implemented) are more than welcome.
2] Finally, any other idea for something to be implemented is again very welcome. Example: When reading the file with preferences, if it does not contain in top 8 list one of hero's initial skills, you are directed to edit the file and place the skill in the appropriate position in top 8 list. Do you have any propositions for automatic fix-up on file so that you don't need to edit it (i.e. place the skill in appropriate position within the file)? I didn't find some time to think about this problem so far.
____________
The empty set
|
|
Xarfax111
Bad-mannered
Supreme Hero
The last hero standing
|
|
posted July 14, 2007 05:54 PM |
|
|
...awww...comon NO!? interest in that Always-Logistic-Thing???
Not important? No efforts?
____________
|
|
dimis
Responsible
Supreme Hero
Digitally signed by FoG
|
posted July 14, 2007 06:32 PM |
|
|
Quote:
...awww...comon NO!? interest in that Always-Logistic-Thing???
Of course!
Quote:
Not important? No efforts?
Important, but no efforts at the moment. Many other things to do (both Heroes-related or not). Btw., version 1.0 above of the internals_mc doesn't work correctly for many reasons (does not read correctly preference files, spou is problematic {I 'll explain with the next release of internals_mc}). Actually, this is what I am working on right now regarding Heroes.
____________
The empty set
|
|
EliteLeader
Adventuring Hero
|
|
posted July 19, 2007 01:31 AM |
|
|
Quote:
...awww...comon NO!? interest in that Always-Logistic-Thing???
Not important? No efforts?
Share dem precious with usssssss 
____________
"One fine day in the middle of the night, Two dead men got up to fight
Back to back they faced each other, Drew their swords and shot each other"
|
|
dimis
Responsible
Supreme Hero
Digitally signed by FoG
|
posted July 20, 2007 11:06 AM |
|
|
internals_mc: New Release
Version 1.1 has been released!
Check out changes and bugfixes in log file at end of home-post.
____________
The empty set
|
|
dimis
Responsible
Supreme Hero
Digitally signed by FoG
|
|
posted July 20, 2007 11:38 AM |
|
Edited by dimis at 11:41, 20 Jul 2007.
|
Logistics related
Why everyone (including me  ) forgets Ecoris' post Searching for skill groups v. 2?
If there is something like always-logistics thing, then either:
a) It falls into an exception given by the conclusions of Ecoris in his post,
or
b) The limited amount of different sequences that are actually generated by Heroes generator allow an exploitation by user when picking skills so that he gets Logistics every time or nearly always.
However, nearly always is just not good enough. For example, the "seek preference otherwise upgrade" policy and the example given on home-page-post of internals_mc, indicates that if Logistics is your primary concern then, you are expected to get it on 80% of the cases (keeping in mind the rest of the selections) with Crag-Hack. Of course, the policy has other problems that do not reflect actual game-playing but that's another issue ...
Btw, in the past we thought that if someone gets Logistics, then the upgrade on the subsequent level-ups is Logistics. This was refuted.
Another thing that was (partially) refuted was that if two sequences are identical for the first 4 level-ups, then, they are identical for the rest level-ups.
Concluding, I parallelize Xarfax with Fermat. He has some very good observations based on actual game-playing and interesting tests for skill offers (not pure nonsense-theory like I do ), but sometimes these do not say the whole truth. More than likely he is not saying the whole truth once again ...
____________
The empty set
|
|
Xarfax111
Bad-mannered
Supreme Hero
The last hero standing
|
|
posted July 20, 2007 10:59 PM |
|
|
..well Dimis...what can i say
First, someone copied a complete skill table from a Heroes site, that should be unnamed into "Valley of the Notionless". He got a cheap QP from angelito simply for copying it. I said this skill table is nonsense (as it is). Everyone told me to shut up, but in the end i proofed them wrong.
Then i said that you can always get earth with Hack if you want. Everyone said this is just blah, but i proofed them wrong.
Now you little ANSA lover tell me that im wrong again. (j/k) Altough you are wrong i wont tell you .
"its ma precious..."
____________
|
|
dimis
Responsible
Supreme Hero
Digitally signed by FoG
|
posted July 21, 2007 01:23 AM |
|
|
That's what I am saying!
The forum is tooo small to get the proof!
Anyway,
I 'll leave for 3-4 days. I would be glad if I am going to be surprised when I come back!
cya
____________
The empty set
|
|
dimis
Responsible
Supreme Hero
Digitally signed by FoG
|
|
posted July 23, 2007 09:43 PM |
|
|
As expected, nothing happened in this thread these days!
Of course there were some great matches underway, but I don't think this was the reason.
As for the rest, the skill table is not nonsense. I don't get why you still say something like that. Moreover, I didn't say you are wrong. I said that it is more likely for you to be wrong than not.
What ever you think allows you to get Logistics every time, must fall in the conclusions of Ecoris or provide a way that guarantees a very high chance of getting Logistics (say 95%+) so that the problematic generator of Heroes makes it happen in all cases. But this latter won't be an exception like wisdom and magic schools. It is because the generator creates a very limited amount of different skill sequences. I hope you get what I am saying.
____________
The empty set
|
|
dimis
Responsible
Supreme Hero
Digitally signed by FoG
|
|
posted July 24, 2007 01:36 PM |
|
Edited by dimis at 13:37, 24 Jul 2007.
|
AlwaysLeft
Added AlwaysLeft results here (last post in page).
angelito add a filler if possible. Just in case ...
____________
The empty set
|
|
angelito
Honorable
Undefeatable Hero
proud father of a princess
|
|
posted July 24, 2007 01:40 PM |
|
|
Keep the good work coming dimis...and don't get off the way by jealous comments..
____________
Better judged by 12 than carried by 6.
|
| |
|
|
This thread is pages long: 
