Research snowballing mitigation

Creation, discussion, and balancing of game content such as techs, buildings, ship parts.

Moderators: Oberlus, Committer

Post Reply
Message
Author
User avatar
Oberlus
Cosmic Dragon
Posts: 5714
Joined: Mon Apr 10, 2017 4:25 pm

Research snowballing mitigation

#1 Post by Oberlus »

Another balancing suggestion to mitigate snowballing effect on research.
If it is welcome I could do something similar for the production techs/buildings.

I usually check Game Graphs to see how I'm going againts AIs. Unfortunately, this let me know if I've already won and makes me lose interest on the current game.
I've seen a pattern: unless I began with bad planets to colonise around homeworld, once I reach Distributed Thought Computing I can claim victory. See the attached image (it is turn 77). I'm the dark blue one (Trith). I got two desert planets to colonise early on and a few tundras so that I could keep on par with the best industry-focused AI and beat all of them on research easy peasy. The first big bump on the graph, a bit before the third quartile, corresponds with the discovery of Distributed Thought Computing. The last sharp bump is due to the confluence of a new Trith Colony and Quantum Networking. At turn 77 current RP is 96 and target is 144, so 50% of the spike is still to be shown once the target RP is reached. Once in there I'm the badass of the universe and AIs are just my puppets. Boring.

This is all about snowballing. We agree it isn't good, right?

I've reviewed the different techs and buildings that increase RP:

Algorithmic Elegance
Research focused planets get +0.1 per pop.
This means an increase of 50% over base pop-dependant RP.
Fine.

Nascent Artificial Intelligence
All planets get +2
Compared with the base pop-dep. RP of planets with 20 pop (i.e. 4 RP), this means +50% RP.
Fine.

Distributed Thought Computing
All planets get +0.1 per pop.
This means a +50% pop-dep research if all planets were research focused, more otherwise.
I guess it's fine, but dunno why AIs don't get this earlier, given that this tech is the one that tend to give me the unsurmountable advantage.

Quantum Networking
Research focused planets get +0.5 per pop.
This means +250% pop-dep research.
Brutal. It is 5 times bigger than the previous increases. I find it totally unbalanced. I would try to set it as 0.15 (+75%) or 0.2 (+100%).

Pure-Energy Metabolism
Research focused planets get +0.5 per pop.
Another +250% pop-dep research.
Again, brutal. Again, 0.2 could be better.

Collective Thought Network (BLD)
Research focused planets get +0.5 per pop.
Another +250% pop-dep research.
Can be disrupted and does not get easy to keep that planet quiet until you already have some map control, so I think it is not as unbalanced as the Pure-Energy Met. and Quantum Net., but still I think some nerfing could be good. I'd give it 0.2 or 0.25 (+150%) tops.

Enclave of the Void (BLD)
Research focused planets get +0.75 per pop.
+375% pop-dep research.
Brutalisch. 0.2 should be enough, it's not disruptable.

Computronium Moon (special)
Research focused planets get +0.2 per pop.
+100% pop-dep research.
Getting one of those could be either the necessary bump for your battered RPs or the final blow to decide who wins. I hope they are well guarded.

Stellar Tomography
Assuming expected distribution of star types for each galaxy age, research focused planets get +0.44 per pop. (0.45 for ancient galaxies)
This means a +220% pop-dep research. Also a bit excesive. I would change the current multipliers like this:
Blue and White: 0.2
Yellow, Orange and Red: 0.1
Neutron: 0.3
Black: 0.5
Thus the average increase for expected star distribution would be +0.195 for young, +0.197 for mature and +0.205 for ancient (i.e. around a +100% pop-dep. research).

If you devs are interested in this kind of balancing, I could try this changes on my own machine and report back my experience for your consideration. However, if you think it is not the moment nor the way to tackle this snowballing problem, or if there is no consensus on it being a problem, I will leave it be.
Attachments
ResearchGraphTrithTurn77.jpg
ResearchGraphTrithTurn77.jpg (36.74 KiB) Viewed 6200 times

User avatar
Dilvish
AI Lead and Programmer Emeritus
Posts: 4768
Joined: Sat Sep 22, 2012 6:25 pm

Re: Research snowballing mitigation

#2 Post by Dilvish »

Oberlus wrote:Another balancing suggestion to mitigate snowballing effect on research.
Civil feedback is always welcome. I think MatCG was planning on doing another round of overall balancing once he does a revamp of Stealth. I am doubtful that we'd be doing any significant rebalancing before then, but your input could be helpful to him, and of course you can implement your ideas as custom content for yourself (and could make them available at your github fork) and then report on how well they seem to work out, for anyone else interested in trying them in the meantime. Be sure to play the full gamut of starting species and galaxy setups to be able to give informed balance feedback.
I usually check Game Graphs to see how I'm going againts AIs. Unfortunately, this let me know if I've already won and makes me lose interest on the current game.
I'll make a couple suggestions-- try starting with a less powerful species, like Humans. Also, although the game defaults to Medium Monsters the AI still is not so great with dealing with them, I think that Low Monsters works out for a more challenging game. I also think the AI does better when there are 35-40 or more systems per empire, and I recommend you try out the different galaxy shapes and different planet densities. I personally tend to like 6-10 AI's with 300-400 systems and Low or Medium Planets, and Low Monsters. The variety should help keep things interesting while we keep working on improving the AI. :D
Distributed Thought Computing
dunno why AIs don't get this earlier
Trith are telepathic, and so can get Psionics (and thus Distrub Thought) very cheaply. When the AI has a telepathic species they do get Distrib Thought pretty early, but otherwise it is very expensive and so they put it off.
If I provided any code, scripts or other content here, it's released under GPL 2.0 and CC-BY-SA 3.0

User avatar
Oberlus
Cosmic Dragon
Posts: 5714
Joined: Mon Apr 10, 2017 4:25 pm

Re: Research snowballing mitigation

#3 Post by Oberlus »

Thanks for this feedback, Dilvish.

I'll do (almost) as you say. Next game I'll do human on ring or disc, 400 systems, low monsters, natives and specials against 9 AIs. I guess the AI gets stressed and resorts to suboptimal industry focus when meets neighbours, so the extra space makes very much sense.

Up to now I've played roughly the same for each species (around two times each right now except Egassem and Chato that got three each). Usually I use low everything, 15 systems per player and 9 AIs (so 150 systems), but the Trith game of this thread is medium monsters. I've abused 4-arm spirals but tried every galaxy shape and age (I'll never again pick young+chato, too much of an advantage).

Regarding Psionics, I always rush to conquer a telepathic native. I'm not sure, but I think Getting Distrib. Thought on turn 55 isn't my best mark, I usually get it around turn 50 starting with non-telepathic species 8) (ah, no, I wasn't remembering it right, that may be true on lucky starts but not usually).

User avatar
MatGB
Creative Contributor
Posts: 3310
Joined: Fri Jun 28, 2013 11:45 pm

Re: Research snowballing mitigation

#4 Post by MatGB »

Dilvish wrote:
Oberlus wrote:Another balancing suggestion to mitigate snowballing effect on research.
Civil feedback is always welcome. I think MatCG was planning on doing another round of overall balancing once he does a revamp of Stealth. I am doubtful that we'd be doing any significant rebalancing before then, but your input could be helpful to him,
Yeah, not sure when to prioritise what, I'm increasingly convinced that the growth and research tech trees are the ones that need the most work for overall balance and player sastisfaction, production is OK as a tree but the mid to late game bonuses are too big.

Whereas stealth-as-a-strategy is an objective to actually introduce, currently stealth's just a tool that's fairly dull and blunt edged.
although the game defaults to Medium Monsters the AI still is not so great with dealing with them, I think that Low Monsters works out for a more challenging game. I also think the AI does better when there are 35-40 or more systems per empire, and I recommend you try out the different galaxy shapes and different planet densities. I personally tend to like 6-10 AI's with 300-400 systems and Low or Medium Planets, and Low Monsters. The variety should help keep things interesting while we keep working on improving the AI.
I find that the best settings for a challenging AI are about 30 systems, High planets (important), Low monsters and, weirdly, galaxy layouts that're generated via Python scripts not the old C++ system—I picked up the last after someonne, I think LGM-Doyle, mentioned it ona thread on Github and I did some testing, the AI seems to do worse on spiral galaxies.
Distributed Thought Computing
dunno why AIs don't get this earlier
Trith are telepathic, and so can get Psionics (and thus Distrub Thought) very cheaply. When the AI has a telepathic species they do get Distrib Thought pretty early, but otherwise it is very expensive and so they put it off.
that's how I do it, try to rush a telepathic conquest if there is one otherwise it's a very low priority.

Anyway, to actual topic: I think the numbers for all the research techs are bad. I also think where they appear in the tech tree are bad. I tend to find that you get the first two, then there's nothing for ages, then you get either Distributed or Networking, follow up with Force Energy Structures (which is the most overpowering tech and is top priority change) and then the rest snowball very quickly.

I want to spread the bonuses out so you get some early, mid and late game boosts, and change most of the numbers: this'll probably change the RP output curve, so lots of other techs then need recosting.

My first specific plan is to move Stellar Tomography forward, reduce the initial bonus but have a refinement later. I like the idea of this one because it's a nice SF flavourful tech and if you get it early it adds strategic choice to colonising decisions that's currently lacking as by the time you've normally got it you're either able to grab wherever you want or in full on conquest mode.

The other is to move Enclave forward, massively reduce the initial output but give all the "*of the Void" an Industrial Centre style bonus. Another idea here is something Geoff suggested ages ago, link some of the bonus to the number of starlanes linking either the system the Enclave is in or the system getting the bonus: I think I prefer the former as it means a choice of location for the building is interesting not just 'best defended'/homeworld.

Then move some others later and possibly tone down bonuses: for what it's worth Pure Energy Metabolism is stupidly overpowered and I want tor educe all the bonuses it gives...

I've been mulling over the idea of putting a PR up with some of these changes very soon after we declare Release as I suspect the AI (and players) will need to learn new priorities.
Mat Bowles

Any code or patches in anything posted here is released under the CC and GPL licences in use for the FO project.

User avatar
Vezzra
Release Manager, Design
Posts: 6095
Joined: Wed Nov 16, 2011 12:56 pm
Location: Sol III

Re: Research snowballing mitigation

#5 Post by Vezzra »

This is a known issue. Practically all the resource output boni are way too high, we've already talked repeatedly about having to flatten these curves. Just didn't get around to it yet. ;)

Post Reply