Early version of an AI you can practice TFM against

Forum rules
Please DO NOT POST BUGS on this forum. Please report (and vote) bugs on : https://boardgamearena.com/bugs
FrankJones
Posts: 2456
Joined: 30 June 2024, 00:24

Re: Early version of an AI you can practice TFM against

Post by FrankJones »

A couple other things to take a look at:

https://boardgamearena.com/gamereview?table=888933443

1) Gen 1 - the bot sems to have Ecological experts confused with eccentric sponsor. It played Eco Experts into ... Io Mining.

It had Farming in its starting cards. I don't know whether Exo-experts -> Farming is a good use of a prelude (seems reasonable enough), but it did not buy Farming. It did buy Lichen, which it could have played after Eco experts.

2) Gen 3: Despite numerous placement bonuses, and mineral deposit in hand, it spent all its steel (unnecessarily) on Space Elevator. It missed out on activation that generation and also one later generation.

I realize the awards are tough. It funded a dangerous Landlord which I could have tied if I hadn't misplayed my hand at the end. I realized I needed 3 or 4 standard projects at the end to finish the game, and I forgot I had standard technology. I caught it halfway through, but if I hadn't wasted 6 dollars first, I could have played another tile and tied the award.
User avatar
Lytical
Posts: 82
Joined: 07 May 2026, 00:51

Re: Early version of an AI you can practice TFM against

Post by Lytical »

Bot stuck on Vitor free funding on Hellas, didn't play a single move the whole game.

https://boardgamearena.com/table?table=888964042
AReallyGoodName123
Posts: 11
Joined: 01 January 2025, 21:12

Re: Early version of an AI you can practice TFM against

Post by AReallyGoodName123 »

Lytical wrote: 26 July 2026, 08:48 Bot stuck on Vitor free funding on Hellas, didn't play a single move the whole game.

https://boardgamearena.com/table?table=888964042
Saw that one. Fixed.
FrankJones
Posts: 2456
Joined: 30 June 2024, 00:24

Re: Early version of an AI you can practice TFM against

Post by FrankJones »

Some other minor stuff, depending on how much you want to fine tune your creation:

https://boardgamearena.com/table?table=889376122


Gen 1 it got off to a fast start, I thought I would get buried.

It ended the gen with an unused power plant card for 2 dollars (After plating Earth Catapult), and holding Development center as a likely gen 2 play.

Gen 2 it plays Olympus conference for no reason. It had no other science cards, and maybe Earth Office would come in the next draft. Also makes me even more likely to hate draft science cards.

Gen 3 it drafted indentured workers over Asteroid as first draft pick, and then did not buy the card. Then it took Mineral Deposit (which saved it 4 MC) over advanced Ecosystems. That's odd, I would figure the card hierarchy you programmed would always prioritize adv. Eco over Mineral Deposit. Same draft it didn't buy indentured workers, it bought Windmills with oxygen at 0.

Gen 4 draft it drafted Mining Area over Algae. Questionable. I ended up with great value from Algae while it was spending 23 dollars to do a lot of SP greeneries. It prioritizes animals in the draft (seems to always take them first, often without actually buying them), but seems to snooze on valuable plant tag cards.

It then played Colonizer Training Camp for no reason (Oxygen was at 0 or 1, and I am Saturn systems).

Gen 6 it prioritized Tardigrades in the Draft over Phobos Space Station when I had a big pile of titanium, and miner was perhaps in contention. It ended up buying and playing Tardigrades, getting one point. Seems kind of like playing Dust seals, not considered a good play.

Gen 7 it was doing weird things with prioritization. It played Deel Well heating, leaving energy dangling even though it seemed to be planning to play a city. Which is fine, I guess, because it had two moves for the city: Lava Tube and then Mining area, which it had for 3 or 4 generations but could not use (partly because of bad placement of its commercial district on a metal spot instead of on a card spot next to a metal spot).

But instead of doing Lava Tube -> mining area, it did something els first, then Lava tube, allowing me to place a blocking tile on its intended mining area spot, which it then could not use.

It eventually paid 25 MC for a SP city in a terrible spot solely to be able to then use mining area (very late in the game). This bad city ended up being worth 1 point when it could have ben 5.


It started to stall out in generations 8-10, doing mostly SP cities and greeneries. Less necessarily if it prioritized Algae more earlier.

It seems to like Adaptation technology a lot, even with no apparent use for the effect.

It drafted Birds first in one gen, but didn't buy it. Birds at the end would have cost me a 2-point greenery. It also passed on Livestock earlier. It does not seem to value getting one of the four 1-point-per animal cards.

Gen 10 draft it Took Insulation in the draft - Wise, because banker might be contested. Id then immediately funded banker - meaning Thermalist was no longer possible. Temperature was done. For 3 Dollars, it could have insulted 10 heat prod into money prod, but it didn't buy the card.

With Miner now almost deadlocked, it then spent all its steel, partly to play mining area (it could have just used 2 dollars), and partly to play a completely unnecessary Fusion power. It had no cards requiring energy, and temp was done, and if it wanted a card from Olympus, it could have played Media Group and special design for free and Gotten the same card from Olympus, while also having Media group in play for the final gen. And saving its 4 steel.

It then spent all its titanium on Callisto Penal mines, a very mediocre gen 10 play against Saturn Systems. I would want it to do media group - special design and maybe it draws a better space card or just a better card. Antigrav was still floating around.

Gen 11 draft, final gen, it took optimal aero with no space events in hand (and really none worth playing anyway at that point with all the good ones having been played or irrelevant.

With Miner still possibly contested, it then drafted Convoy From Europa (useless) over building industries, a 4-steel swing. If it takes building industries, I can't use all my metal. Then it actually bought Optimal aero. waste of 2 dollars.

Then it did two SP greeneries and sold its cards, even though it had no yet used inventor's Guild or Space elevator. Definitely better to look at the IG card. It could have kept a 1-point card and played it.
FrankJones
Posts: 2456
Joined: 30 June 2024, 00:24

Re: Early version of an AI you can practice TFM against

Post by FrankJones »

Small sample size, but:

I have 24 wins and 5 losses. Based on my current Elo of 438, that would mean tfmBot is playing at a 165 Elo.

I think it's better than that, so either
1) My true level is higher than 438 (Possible, and I certainly hope so; I was as high as 535 at one point), or
2) I have been winning at a higher rate than expected.

Or, I am mistaken, and the bot is 165 Elo, in which case my win rate would be exactly as expected based on my 438 Elo.

If my actual level is my high of 535, that would mean the bot's Elo is 263.

If my true Elo is halfway between my current and my highest (not a bad estimate method), then my Elo would be 486 and the Bot would be 213.
GersonK
Posts: 1
Joined: 21 May 2020, 23:09

Re: Early version of an AI you can practice TFM against

Post by GersonK »

https://boardgamearena.com/table?table=889950650

Looks like it needs some tweaking on how it values cards late game:
Last moves of the game, I passed with Mars fully terraformed
Bot sells its last card, then buys one from Inventor's Guild
Then sells that card
It should have had enough info to realize it wasn't going to play that card, but I see where that's a special case to change up the logic
It of course still ties me thanks to a last second success on S4L, but loose the tie break by 5MC
AReallyGoodName123
Posts: 11
Joined: 01 January 2025, 21:12

Re: Early version of an AI you can practice TFM against

Post by AReallyGoodName123 »

FrankJones wrote: 27 July 2026, 22:24 Small sample size, but:

I have 24 wins and 5 losses. Based on my current Elo of 438, that would mean tfmBot is playing at a 165 Elo.

I think it's better than that, so either
1) My true level is higher than 438 (Possible, and I certainly hope so; I was as high as 535 at one point), or
2) I have been winning at a higher rate than expected.

Or, I am mistaken, and the bot is 165 Elo, in which case my win rate would be exactly as expected based on my 438 Elo.

If my actual level is my high of 535, that would mean the bot's Elo is 263.

If my true Elo is halfway between my current and my highest (not a bad estimate method), then my Elo would be 486 and the Bot would be 213.

One thing to consider is that it's just a baby. 8 days old! It's learning.

You should find you win less and less as time goes on. Today's AI beats 8 days ago AI over 75% of the time in a head to head
FrankJones
Posts: 2456
Joined: 30 June 2024, 00:24

Re: Early version of an AI you can practice TFM against

Post by FrankJones »

Yeah, I just lost with point Luna.

I decided to have fun and gamble.
It did not pay off.

I ended up with NINETY dollars worth of unused alloyed metal.

What would the experts do with this tarting hand?

https://boardgamearena.com/table?table=890488859


Go full crazy gamble on Jovians (which failed)? Or , something else?
Last edited by FrankJones on 29 July 2026, 23:11, edited 1 time in total.
Peach 2d
Posts: 10
Joined: 10 April 2026, 08:18

Re: Early version of an AI you can practice TFM against

Post by Peach 2d »

What would the experts do with this tarting hand?

https://boardgamearena.com/table?table=890488859


Go full crazy gamble on Jovians (which failed)? Or , something else?
Technically, BGA calls me an "Expert", albeit at the lower reaches of the expertise hill :-)
Galilean Mining comes with an MC penalty. The appeal of Io Mining Industries and Jovians is clear, but it also gives a slow start with little development in generation 1. I'd probably go for:
- Point Luna and Orbital Construction Yard, as per your selection
- Skip Galilean Mining and pick Research Network instead. It's a strong prelude for PL, gives you cards, MC production, and a wild tag.
- as cards, buy Space Station, Solar Wind Power, Opti Aerobraking, Satellites, Convoy and Natural Preserve. That leaves you with 20MC and 12MC's worth of titanium from Orbital Construction Yard.
- There's always a risk of sabotage or hired raiders disrupting the plan, but assuming there's no such hit:
- Key plays become: Space Station, Opti Aerobraking, Solar Wind Power and Satellites. That gets you down to 4MC and zero titanium. And 7MC production (Satellites, SWP, Point Luna, Research Network, Space Station, Opti Aerobraking: 6 space tags - giving a great 6 MC prod boost from Satellites in gen 1. Plus you get 1MC prod from Research Network), 2 titanium production, and 1 energy production. The cards from Research Network may offer a better line, and you want to evaluate the cards you get first, but this is a solid fall-back plan. All these plays somehow "make sense" - Space Station's discount is used to full effect immediately, for 3 space cards this generation. It all sets up a Satellites play that is pretty outrageous: which other cards could you possibly play in generation 1 for 8MC (Satellites placement cost after the space station discount) that give that kind of economy boost? Space Station and Opti Aero set up the Convoy play for generation 2, so are useful infrastructure that get almost instantly a solid throughput. And Solar Wind Power, played at 9MC and rebating 2 titanium that you can use instantly, means you get a science/space tag and 1 energy prod for 3MC. What's not to like?

It all leaves you with few cards - and that's another reason for picking Research Network. You have the beginnings of a science track (1 science tag, 1 wild tag, Natural Preserve as 3rd science tag waiting in the wings), decent titanium production, cards from Research Network, and solid MC production that helps buy and play cards.
For generation 2, you have Natural Preserve and Convoy as perfectly decent cards for this game phase. Natural Preserve plus the wild tag from Research Network also offer a glimmer of hope for Builder as a milestone in case good building cards appear.
Convoy gives another card, uses the Space Station discount and gives an Opti Aero activation - and this early in the game, will allow for premium ocean placements (cards/titanium/steel). In this game, the bot took the 2 cards and 2 titanium pickups, so going for the 2 steel pickup and using that for Natural Preserve makes sense - and pick up a card with Natural Preserve, or other placement benefits. At the time of selecting your preludes and cards, you don't know that, of course, except that it's reasonably likely that a premium ocean will still be available.

Based on this hand, I'd probably skip Io Mining Industries. There is no ongoing card draw that would help you find more Jovians. So with this hand, you'd be reliant on the draft. If you really want to stretch, keeping IoMI is an option - but the opening does become very tight, and vulnerable not just to sabotage, but also to hired raiders.
The line I have in mind sets you up for good development with a strong early economy, space discounts and rebates, cards (Research Network and Convoy, plus probably a card from Natural Preserve placement). It would also avoid an overinvestment in the more narrow, space-only spending possibilities of titanium, and instead get more flexible MC production - and very strong MC production for that matter, starting at +7MC and rising. Cards from Convoy, Research Network and the draft in generations 2 and 3 can then give further direction from which you can move towards a coherent strategy. That would be my suggestion.
FrankJones
Posts: 2456
Joined: 30 June 2024, 00:24

Re: Early version of an AI you can practice TFM against

Post by FrankJones »

Analysis looks good. Thanks :)

I just now saw this because I didn't even remember posting this last night.

I remember when choosing the line I chose that it was likely a silly approach, but it seemed like a fun thing to try.
Locked

Return to “Terraforming Mars”