Hi,
i have a few questions about
AI analysis: The first 100 turns:
- how random is the AI's Decisions ? aka will i see the same game (following a promised debug thread)
- What shall be optimized
- Research efficiency
- growth
- military security (e.g. against rushes)
If i understand the posts correctly it is about research and building efficiency at the moment. Like if you built a settler make sure there is a proper escort.
So not reallz optimize it in terms of comparable to a human player but to prevent a complete waste of hammers like abondoning a almost finished granary for something else.
If you load up one of the saves using the version
here and play the game forward, it should in principle produce the exact same results. However, it won't in this case because I started and stopped automation a number of times to take screenshots and things like that along the way. If automation runs turns 0 - 10, then 10 - 20, there will be a different number of calls to the random number generator than if automation runs turns 0 - 20 without interruption. With the way "random" works in a computer, that means all the random dice rolls going forward will be different.
So, while the AI definitely makes random decisions, if two games start with the same configuration, same initial random seed, and same game code ... and the human player makes the same moves/decisions in the same order, then the games would play out exactly the same.
While many AI decisions have a random component, not all do and in some cases the non-random computed difference between two alternatives can be greater than the random add on. So, if you do start the save from 4000 BC in the other thread, you'll see certain definite patterns ... cities founded in the same places, things like that.
In terms of the goal of that thread, there are several:
- Looking for evidence of bugs from bad decisions
- Highlighting unnecessary inefficiency, like waiting settlers or idle workers
- Determining how to increase the coherence of the AIs early game strategy
When there are broken or blatantly inefficient things the AI is doing, those are the easiest to spot and the easiest to fix. Getting to a higher level of efficiency and coherence in strategy is a longer term goal, and usually the bugs have to be mostly out first.
