AI Aggression Randomization

[Foocaux]

PlayerID() by itself doesn't seem to provide too much randomness either...

I'm now trying to combine empire name & galaxy seed to the Seed() call to see if rand0 gets simultaneously consistent & a bit more random!

Patch incoming if it does!

[Geoff the Medio]

PlayerID() by itself doesn't seem to provide too much randomness either...

What do you mean by "doesn't seem"?

[Foocaux]

In my system AI number 1 will always be at max aggression, regardless of galaxy seed, or actual empire name, etc, etc...

[Foocaux]

Ok, I've managed to make some progress: I got AI aggressiveness to be random in windows, and dependant on both AI empire name & AI playerID. I've started playing around with it a bit, bu RL is calling, so here's my patch, warts & all...

EDIT: in my first version of this post I somehow sent a patch that confused the AI name with its empire name. Not anymore! I don't think there are any major warts left & I'm pretty happy with it as it is now. 'Course, there could be warnings from gcc and/or clang lurking in there...

[Geoff the Medio]

As noted above, using the player ID for random seeding is probably a bad idea as it prevent reproducibility of setup results if the mapping between player name and id and empire id changes between game starts.

Also, the use of the custom name hashing function is potentially problematic. I suggest using the boost hashing method posted above.

Also, comments explaining some of what's going on with the range tests would be useful... it's very unclear what some of the code changes are doing.

[Foocaux]

Is this what you had in mind?

[Geoff the Medio]

This doesn't make sense:

Code: Select all

boost::lexical_cast<unsigned int>(empName)

Empire names are not (generally) string representations of integers.

Re:

Code: Select all

// if it's in the top 25% then decrease aggression.

Why?

[Foocaux]

That cast (as a fallback in case hashing fails) was there to still provide some variety based on the Chars comprising the empire name.

I simply followed the original example for galaxy seeds: they seem to be mostly alphabetical, but that very cast is used as a fallback: given that you told me to do the same for empire names as what was done for galaxy seeds, I imagined you meant for me to do everything the same way. Apologies if I misunderstood.


The 25% of AI with lower aggression is already implemented in trunk, for osx & linux.

I've just followed what was there already, implemented by Dilvish, I believe.

The 8% of AI having even less aggression was totally my idea, though.
I did pluck that second number - and lower aggression level - out of thin air, it kind of 'felt right', somehow.

[Foocaux]

Ok, last iteration on mySeed generation. Yesterday evening I misread FO's original code, so the fallback mySeed generator was pretty useless.

Now with a working fallback generator, that can also be commented out... tbh I'm not clear why there's a need to bracket boost::hash inside a try/catch...

[Dilvish]

hey guys, I'm back. The combo of galaxy seed and empire name sounds fine/good to me, enabling reproducibility without creating excessive predictability.

As for aggression distribution, I had originally started with a broader distribution, but after a fair bit of feedback we settled on the current distribution simply having 25% to be a level below max. I think that's been working out fairly well, and I think that at least for 0.4.4 we should not change that.

We could certainly have another discussion about it, and perhaps we should wind up adding a new table for it into universe_tables.txt so that it could be more readily user-customizable.

[Foocaux]

Hello Dilvish, do pick what's good from my patch, discard the rest!

Cheers.

[Vezzra]

Yep, and please merge the fix into the release branch too

[Dilvish]

Yep, and please merge the fix into the release branch too

Sounds fine, though it may be another day or two until I get to that. Yesterday morning and this morning I took my daughters freediving for abalone (we got 4, hurrah)-- first time for them, and it's been over a dozen years since I've gone, so I am rather a bit exhausted just now.

[Vezzra]

Take your time, FO is a hobby we do for fun, no need to rush anything

Oh, btw, concerning merging commits between branches/trunk (because you mentioned it in one of the other threads): The method you described of course should work, but there is a simpler solution: svn merge is your friend!

[adrian_broher]

Oh, btw, concerning merging commits between branches/trunk (because you mentioned it in one of the other threads): The method you described of course should work, but there is a simpler solution: svn merge is your friend!

NO! svn merge without proper parameters will merge everything. Read http://svnbook.red-bean.com/en/1.7/svn. ... rrypicking before merging.