Ironworks Gaming Forum - View Single Post

Ziggurat · 08-19-2003, 04:26 PM

Skip this long one if don't care to hear more about math, etc...

I agree a model is crucial, although many models can be tested.

What would need to be done to simplify this process is make a party of identical characters. Say 5 samurai, same race (and sex?) with same armor and weapons. Add a bishop perhaps to help keep them alive. Then track the number of hits against identical enemies, along with all relevant stats including all those mentioned above in previous posts. This could be alot of parameters to track!! But write them all down for each battle. Keep it simple. Total crits and TS may be enough to count, but you will need the other params, too.

There will be differences in how many hits each sam gets based on the randomness and the inevitable changing enemies in the same battle, so there is an inherent fuzziness to the data. Some of this may be averaged out over the long run. Position in the party must be rotated evenly. And these strikes really don't emerge regularly til mid game. This project would be time-consuming.

On leveling, make each develop differently, one trait at a time. Say, keep piety or intelligence the same for all three. (That is why I choose 5 instead of 4 sams.) Each one pushes one trait, with the other trait always being piety (or intelligence). As some traits lead to better fighters, or more attacks, etc., the number of total attack attempts and penetrations should probably be tracked as well. You perhaps see how this is getting ridulous...

Then you have five equations in five unknowns and need to find a correlation between these 5 factors and the quantity which is the outcome, critical strike average or tigerstrike average. These will have small correlative "slopes" which give how fast the CS or TS skill rises with each of the five abilites (assuming they are linear equations! If not, this would show up on a graphic display of each). It may be assumed something like piety or intelligence would have a low correlation coefficient (statisticaly saying no cause or effect of the quantity on CS or TS) although the description of intelligence in the game says it effects close combat (maybe just the rate of increase in leveling it up during combat?). If so, it makes the equation even easier to solve if there are less parameters. And it would give you a double check. This might also be useful to help reduce the inevitable noise since there are not a large number of battles.

This is one way to do it, multiple models can be tried with the same data to determine the best fit (highest correlation). Another technique called the simplex method can zero in on the solution so you don't need a computer if you know the technique or have a good linear algebra book. Non-linear equations probably aren't used, but any of these parameters are multiplied instead of added it adds more difficulty and the model would have to be changed.

Once it is discovered how these four or five character traits effect the strikes, then you can start your real work!

You can start looking at how enemy armor class, weapon damage, initiative or number of attacks, to penetrate, close combat, martial arts, carrying capacity load, bow skills, range skills, special bonuses from weapons, and the critical strike skill itself effect the CS and TS rate. (I probably left a few out...)

You also may want to use CS or TS per swing instead of the total per battle (but only count the first swing if it is a Tiger Strike).

For the data to be meaningful, it would better to cut out as much extraneous noise as possible. This is my first thoughts on determining the equations. As I said, it wouldn't be easy.

I wouldn't even attempt to do something on this scale with a group effort using different parties and characters, too many variables. A good math program COULD be used to solve this, say twenty equations with twenty unknowns, but the data would have to be pretty good for it to be meaningful. And a large data set that would be! And how would you be sure if a factor was 0.02 (relating, say, CS probabilty of occurence to critical strike skill) when it was really 0.023? Remember, these factors are tweaked during game testing and may even differ between races. My two cents...Zig

[ 08-19-2003, 04:46 PM: Message edited by: Ziggurat ]

08-19-2003, 04:26 PM	#11
Ziggurat Symbol of Cyric Join Date: November 4, 2001 Location: Baltimore, Md Age: 71 Posts: 1,106	Skip this long one if don't care to hear more about math, etc... I agree a model is crucial, although many models can be tested. What would need to be done to simplify this process is make a party of identical characters. Say 5 samurai, same race (and sex?) with same armor and weapons. Add a bishop perhaps to help keep them alive. Then track the number of hits against identical enemies, along with all relevant stats including all those mentioned above in previous posts. This could be alot of parameters to track!! But write them all down for each battle. Keep it simple. Total crits and TS may be enough to count, but you will need the other params, too. There will be differences in how many hits each sam gets based on the randomness and the inevitable changing enemies in the same battle, so there is an inherent fuzziness to the data. Some of this may be averaged out over the long run. Position in the party must be rotated evenly. And these strikes really don't emerge regularly til mid game. This project would be time-consuming. On leveling, make each develop differently, one trait at a time. Say, keep piety or intelligence the same for all three. (That is why I choose 5 instead of 4 sams.) Each one pushes one trait, with the other trait always being piety (or intelligence). As some traits lead to better fighters, or more attacks, etc., the number of total attack attempts and penetrations should probably be tracked as well. You perhaps see how this is getting ridulous... Then you have five equations in five unknowns and need to find a correlation between these 5 factors and the quantity which is the outcome, critical strike average or tigerstrike average. These will have small correlative "slopes" which give how fast the CS or TS skill rises with each of the five abilites (assuming they are linear equations! If not, this would show up on a graphic display of each). It may be assumed something like piety or intelligence would have a low correlation coefficient (statisticaly saying no cause or effect of the quantity on CS or TS) although the description of intelligence in the game says it effects close combat (maybe just the rate of increase in leveling it up during combat?). If so, it makes the equation even easier to solve if there are less parameters. And it would give you a double check. This might also be useful to help reduce the inevitable noise since there are not a large number of battles. This is one way to do it, multiple models can be tried with the same data to determine the best fit (highest correlation). Another technique called the simplex method can zero in on the solution so you don't need a computer if you know the technique or have a good linear algebra book. Non-linear equations probably aren't used, but any of these parameters are multiplied instead of added it adds more difficulty and the model would have to be changed. Once it is discovered how these four or five character traits effect the strikes, then you can start your real work! You can start looking at how enemy armor class, weapon damage, initiative or number of attacks, to penetrate, close combat, martial arts, carrying capacity load, bow skills, range skills, special bonuses from weapons, and the critical strike skill itself effect the CS and TS rate. (I probably left a few out...) You also may want to use CS or TS per swing instead of the total per battle (but only count the first swing if it is a Tiger Strike). For the data to be meaningful, it would better to cut out as much extraneous noise as possible. This is my first thoughts on determining the equations. As I said, it wouldn't be easy. I wouldn't even attempt to do something on this scale with a group effort using different parties and characters, too many variables. A good math program COULD be used to solve this, say twenty equations with twenty unknowns, but the data would have to be pretty good for it to be meaningful. And a large data set that would be! And how would you be sure if a factor was 0.02 (relating, say, CS probabilty of occurence to critical strike skill) when it was really 0.023? Remember, these factors are tweaked during game testing and may even differ between races. My two cents...Zig [ 08-19-2003, 04:46 PM: Message edited by: Ziggurat ] __________________ How are you today, Citizen?