Plenty of abilities are unusable by the AI. I wouldn't count this.
650 AI Game 4UC Stats / Analysis
- Thread starter L. Vern
- Start date
KungCheops
Warlord
Which?
pineappledan
Deity
The only legitimately challenging ones to use that I could see are:
- Sofa - the AI doesn’t know its mounted units get a combat bonus near its archer units
- Suffet - the GAdmiral/GGeneral switching is tough to deal with, even for humans
- Armada - the switch to a faster/weaker unit when damaged Vs slower/stronger at full HP is something the AI doesn’t know about. Not sure if it matters
- Tarkhan - the AI doesn’t know that if it escorts settlers etc with their UU they will move faster... It does that anyways, but it doesn't know this unit does it Particularly well
- Baochuan - the AI doesn’t know it can generate influence by parking their UU in CS territory.
- Iron chariot - the AI won’t be able to recognize the resource management ability, though foregoing the resource management CS bonus to spam a SR-free horseman is a perfectly legitimate strategy with the unit
How feasible is to add amendments to general AI, so it takes into consideration its UC?
4UC Wat apparently has 2 writing slots and a theme better than Amphitheater's, with no downsides compared to the current VP version. Could've contributed to their win rate a bit (all those CVs they shouldn't have bonus towards).
Havent played the mod much but I recall going mass archers with india was completely broken.
Also the aztec 10 kill upgrade was good.
More uu meant more agression and dangerous AIs.
Wouldnt mind the migration but I def want some stuff toned down.
Also the aztec 10 kill upgrade was good.
More uu meant more agression and dangerous AIs.
Wouldnt mind the migration but I def want some stuff toned down.
I can confirm Indian archers are absolutely wild
Basically any ranged UU is by default S-tier given ranged land units make up the majority (or even super majority) of total damage output
L. Vern
Warlord
| Victory Type | Cultural | Diplomatic | Domination | Science | Time |
|---|---|---|---|---|---|
| Correlation With Score (At Game End) | -0.300 | -0.197 | 0.708 | 0.074 | 0.205 |
I might not understand how the correlation score works, but consider Time IS a score victory....shouldn't that be 1? You can't win a time victory without having the highest score after all.
Thanks! That's what I expected more or less. So score from tourism might be a good idea.
L. Vern
Warlord
Correlation is just a measure of how linearly related two variables are - if correlation is positive, higher score = higher chance of winning that type of victory. (strictly speaking that's not correct to say, that would be a logistic regression interpretation, but same idea). This calculation was also done on only the 650 win instances, if all 5200 points were used I think all of them would be positive; these correlations are conditional on knowing a civ wins and predicting what kind of win they got based on their score
Could you elaborate why?
Perhaps I’ve got the results backwards, let me check. I feel this correlation answers one of the following two questions, but which one?
1) I won a culture victory, how likely was it I had the highest score?
2) I had the highest score at the end of the game, how likely was it I won a culture victory? (Do the stats also correlate for not winning at all, or does it assume you won in the stat?)
L. Vern
Warlord
Yea, let me set this up in a way that's easier to understand, using all civ game instances. Basically, we begin by plotting Wins (1/0) against Score and using linear regression to get a line of best fit
(ignore the duplicated losses plotted in every color, too much effort to clean it up lol)
The correlation is a measure of how strong the linear relationship between score and victory type is, whereas the slope of the line of best fit is a measure of how much one variable changes as a response to the other.
The statistical definition of correlation probably goes against how its used in common language, which is why I wanted to include both in the explanation - correlation is a measure of how tightly coupled the relationship is - a 20 degree line or a 45 degree line can both have a correlation of 1 for example, if they were both both perfect linear functions. The slope on the other hand, describes how much change in one variable we would expect to see from a change in another. Higher correlations means there's a stronger coupling, so higher correlations on domination and science means that changes in scores, on average, will more "reliably" change victory outcomes, whereas a low correlation means it won't matter as much.
Hopefully that makes it a bit more clear, let me know if there's any other questions
(ignore the duplicated losses plotted in every color, too much effort to clean it up lol)
| Cultural | Diplomatic | Domination | Science | Time | |
|---|---|---|---|---|---|
| Slope of fit line (*1e5) | 1.850 | 1.974 | 1.694 | 3.338 | 0.193 |
| Correlation | 0.176 | 0.190 | 0.308 | 0.315 | 0.09 |
The correlation is a measure of how strong the linear relationship between score and victory type is, whereas the slope of the line of best fit is a measure of how much one variable changes as a response to the other.
The statistical definition of correlation probably goes against how its used in common language, which is why I wanted to include both in the explanation - correlation is a measure of how tightly coupled the relationship is - a 20 degree line or a 45 degree line can both have a correlation of 1 for example, if they were both both perfect linear functions. The slope on the other hand, describes how much change in one variable we would expect to see from a change in another. Higher correlations means there's a stronger coupling, so higher correlations on domination and science means that changes in scores, on average, will more "reliably" change victory outcomes, whereas a low correlation means it won't matter as much.
Hopefully that makes it a bit more clear, let me know if there's any other questions
Last edited:
Could you do a candlestick plots for each victory type, where each point is:
1) Score of the winner
2) Score of the winner - average score of all civs including dead
3) Score of the winner - median score of all civs including dead
I would like to see score distribution for each victory type.
1) Score of the winner
2) Score of the winner - average score of all civs including dead
3) Score of the winner - median score of all civs including dead
I would like to see score distribution for each victory type.
L. Vern
Warlord
Not a candlestick, but here's a visualization of score distributions for each victory type
That's what I had in mind
Thanks!
bonniepbilly
Warlord
- Joined
- Jul 4, 2007
- Messages
- 156
I feel like I see more snowballing civs lately.
Would you have the stats on the score difference between the winner and 2nd place (at time of win), as a way to measure this?
Would you have the stats on the score difference between the winner and 2nd place (at time of win), as a way to measure this?
Similar threads
