I don't really get it. Firstly, this isn't logistic regression, and secondly what exactly is the purpose of regression here? The power and benefit of regression is to be able to make predictions and derive formulas from data, here you are literally just estimating equations that you already typed out? Why not just use the equations that you used to generate the data in the first place lol The actual design problem here is coming up with those original equations you typed out to generate the data, how do you decide which type of function is suitable for which attribute? How do you generalize the concepts of player health, player damage, enemy health and enemy damage if those things are dependent on items the player finds, upgrades or equips (like most rpgs)?
I think the point is people would randomly come up with numbers that fit their ideas and then want a function instead of lookup table, and then eventually want to be able to tweak the values without having to mess with the function directly. I assume he just didn't want to spend time punching in all the values by hand, but also the "functions" he uses require the output at the previous input, so it's not straightforward what the function would be for a (typical?) non-math-y game designer. A lot of RPGs have "base stats" and equipment is a modifier. To give a specific example: you might want your levels planned out such that the player's base damage is just under one-hit-killing enemies of the same level by some amount. Let's assume that gap also expands per level, and assume that gear of the same level perfectly fills that gap. Now if a player is fighting on level, but under-geared we know it will take at least 2 rounds of combat to kill each enemy of the same level. This would let you control how it feels to play as the "worst character" and maybe inform how any modifiers should be paced out. I'm not saying that would be a great design in itself, rather just how you can focus on design and derive numbers to fit that intent.
Is there a reason to start with extremely specific concepts like player/enemy health/damage? You go on to point out the actual interesting questions in design later. (3:45 Level 1 takes 3 hits to kill per enemy, Level 2 takes 2) So wouldn't it be a lot easier for the designer to deal directly with these concepts? Then derive a curve for exact health/damage numbers based on how they want that enemy to be used (horde vs normal vs boss) and game feel progression later? Maybe I'm wrong, but it really seems like the only thing that matters are the thresholds between an enemy being a threat or not. So deciding an enemy should take 45 or 46 damage is pointless when what you want (for example) is that players cannot one-shot LV1 enemies until LV4. Then the next interesting question is, if the player augments their base numbers (via gear or whatever) what should the game feel like? I.e. Does a LV1 weapon allow a LV2 player to one-shot LV1 enemies? Can I use the concepts discussed in your video to derive the lowest level numbers that answer the questions a *designer* is interested in? I haven't wrapped my head around it enough yet to know how I would do this.
This is exactly what I did for an action RPG project (think loot-based game like Diablo). We modeled time-to-kill goals at ALL stages of progression, including how gear of various qualities would affect that time-to-kill. The models became extremely complex due to the sheer number of variables in play, but I think it was at least 60-70% successful. With a simpler progression model I think it would perform very, very well. Maybe I should make a video about that some time?
I’m a bit confused. Isn’t logistic regression used for modeling the probability of a dependent variable belonging to a category and not for modeling a continuous, nonlinear curve as you’ve shown here? Isn’t this nonlinear regression, not logistic regression?
Yes this is just nonlinear regression/curve fitting. I thought he might use a logistic curve at some point. Which is a simple S shaped curve you could use for something like this. But that does not appear in the video.
That was exaclty what i was searching for, trying to code an rpg rn and wanted to now how to get formulas for it ! Thanks a lot for that tip !
wtf? I guess I should be grateful Google listens to all my convos cuz this is exactly what I need
Thanks evil privacy-invasive overlords!
Thank you! It is really helpful for designing games.
I don't really get it. Firstly, this isn't logistic regression, and secondly what exactly is the purpose of regression here? The power and benefit of regression is to be able to make predictions and derive formulas from data, here you are literally just estimating equations that you already typed out? Why not just use the equations that you used to generate the data in the first place lol The actual design problem here is coming up with those original equations you typed out to generate the data, how do you decide which type of function is suitable for which attribute? How do you generalize the concepts of player health, player damage, enemy health and enemy damage if those things are dependent on items the player finds, upgrades or equips (like most rpgs)?
I think the point is people would randomly come up with numbers that fit their ideas and then want a function instead of lookup table, and then eventually want to be able to tweak the values without having to mess with the function directly.
I assume he just didn't want to spend time punching in all the values by hand, but also the "functions" he uses require the output at the previous input, so it's not straightforward what the function would be for a (typical?) non-math-y game designer.
A lot of RPGs have "base stats" and equipment is a modifier. To give a specific example: you might want your levels planned out such that the player's base damage is just under one-hit-killing enemies of the same level by some amount. Let's assume that gap also expands per level, and assume that gear of the same level perfectly fills that gap. Now if a player is fighting on level, but under-geared we know it will take at least 2 rounds of combat to kill each enemy of the same level. This would let you control how it feels to play as the "worst character" and maybe inform how any modifiers should be paced out.
I'm not saying that would be a great design in itself, rather just how you can focus on design and derive numbers to fit that intent.
thanks for this!
Thank you alot for sharing this
Is there a reason to start with extremely specific concepts like player/enemy health/damage?
You go on to point out the actual interesting questions in design later. (3:45 Level 1 takes 3 hits to kill per enemy, Level 2 takes 2)
So wouldn't it be a lot easier for the designer to deal directly with these concepts?
Then derive a curve for exact health/damage numbers based on how they want that enemy to be used (horde vs normal vs boss) and game feel progression later?
Maybe I'm wrong, but it really seems like the only thing that matters are the thresholds between an enemy being a threat or not.
So deciding an enemy should take 45 or 46 damage is pointless when what you want (for example) is that players cannot one-shot LV1 enemies until LV4.
Then the next interesting question is, if the player augments their base numbers (via gear or whatever) what should the game feel like?
I.e. Does a LV1 weapon allow a LV2 player to one-shot LV1 enemies?
Can I use the concepts discussed in your video to derive the lowest level numbers that answer the questions a *designer* is interested in? I haven't wrapped my head around it enough yet to know how I would do this.
This is exactly what I did for an action RPG project (think loot-based game like Diablo). We modeled time-to-kill goals at ALL stages of progression, including how gear of various qualities would affect that time-to-kill.
The models became extremely complex due to the sheer number of variables in play, but I think it was at least 60-70% successful. With a simpler progression model I think it would perform very, very well.
Maybe I should make a video about that some time?
As for your question: absolutely! Give it a shot and let me know what you think about modeling it. It's a very interesting topic!
@@polatrite I think that would be a very interesting video!
I’m a bit confused. Isn’t logistic regression used for modeling the probability of a dependent variable belonging to a category and not for modeling a continuous, nonlinear curve as you’ve shown here? Isn’t this nonlinear regression, not logistic regression?
Yes this is just nonlinear regression/curve fitting. I thought he might use a logistic curve at some point. Which is a simple S shaped curve you could use for something like this. But that does not appear in the video.
This helped a lot, thank you.
Log is a reverse exponential curve... x and y are flipped.
This is amazing, thanks!
can you explain how fixing those curves make our normal game equations better?
Thank you i'll use that 😃
👍