new bot BetaStar
Based on its authors, BetaStar looks to be associated with China’s State Key Laboratory for Novel Software Technology (page in Chinese) at Nanjing University; it is sometimes translated as “National Key Laboratory” instead. Here are a couple papers that seem to be from the same research group that produced it. They are both about Starcraft II.
• On Reinforcement Learning for Full-length Game of StarCraft (September 2018)
• Efficient Reinforcement Learning with a Mind-Game for Full-Length StarCraft II (March 2019; by “mind game” they mean a simplified abstract version of the real game)
The papers have similar titles, but the contents are different. The first is about hierarchical reinforcement learning for macro action choices; the second builds on it with the “mind game” to make learning faster and more effective.
I did not find information about how BetaStar is related to this past research. Presumably a new research paper is in preparation, or published somewhere I did not look. Included in the binary are 3 files with the air of learned data, less than 1M each in size, named pvp_params.csv, pvt_params.csv, and pvz_params.csv. If BetaStar uses the methods of the research papers above, then these will be data for choosing macro actions—what to build next, what to research, etc.
BetaStar is derived from CSE. The line of descent is UAlbertaBot > Steamhammer > Locutus > CSE > BetaStar. It has strong inherited skills. CSE finished #3 in AIIDE 2018, ahead of Locutus. I could not figure out where the name BetaStar comes from. In this context, it sounds like an algorithm name, but if so I guess it is a new algorithm.
So I found a bunch of clues about BetaStar, but I don’t actually know a thing about how it works! It finished #9 in the CoG tournament (formerly CIG), with a 59.04% win rate, ahead of MetaBot and behind ZZZKBot. As I write, it has the BASIL elo of 2314, which is above average but well below the top; it is ranked #31 out of 83 active bots, between Arrakhammer and TyrProtoss. It has 55 wins, 23 losses, and—this is the amazing part—49 crashes. It often fails to start games at all. Its basic play is quite strong, with Locutus dragoon micro and other skills. So far, between SSCAIT and BASIL, Steamhammer has several losses and only 1 win against it. BetaStar scored 2-0 versus #3 PurpleWave on BASIL. But besides crashing, BetaStar has play bugs that other bots do not: It likes to build 2 cybernetics cores. It sometimes plays an unsafe opening and loses to a rush, then repeats the opening against the same opponent and loses the same way (see vs legacy 1 and vs legacy 2—#50 legacy is 3-0 versus BetaStar). It seems that BetaStar has both great strengths and great weaknesses.
Without much to go on, I nevertheless read BetaStar as similar to past Chinese research projects I’ve looked into: It is meant to be a one-off project to demonstrate a specific research goal, and aspects outside the goal were afforded no more effort than absolutely needed; solving crashes was not part of the goal. If so, it may have a successor next year in the same way that it succeeded CSE, but I don’t expect updates in the meantime.
It’s good to have a new bot! They have been few lately.
Trackbacks
No Trackbacks
Comments
Dan on :
There has already been one update since the initial upload (I haven't investigated the changes) which offers hope for further updates, but on the other hand it's not competing in AIIDE.
Jay Scott on :
Jay Scott on :
MarcoDBAA on :
It feels as if you want to win at all costs (lazy shortcut) instead of demonstrating artificial intelligence. Would be ok for a single author or student, but I am not impressed here. They already use a strong base bot anyway.
P.S: As I said before, there won´t be many new (especially from scratch) bots, because there are no/few weaker bots left active and noone wants to upload a bot that looses (nearly) all games.
Joseph Huang on :
Bruce on :
Jay Scott on :
MarcoDBAA on :
I am just saying, that it isn´t (only) positive. There is a danger, that a community dries up, and/or most people will only use existing base bots, thereby reducing innovation (and many bots play very similar). And this seems to be happening. It wasn´t like that before (I am following it for a long time).
I also don´t really get the problem. People can and will vote for top match ups anyway. The problem existed, when there was no voting. People like me, with limited internet data volume, just search for (interesting) replays. And yes, I mainly search for top bots, but this isn´t the point.
You could create tiers, and/or higher chance to get randomly chosen for bots with a higher elo... It doesn´t have to be this clear cut gate keeping. I also like to watch medium tier bots more than clones (or nearly clones) anyway.
P.S: @Jay
You can tell me, if this is too much OT here. Originally I only wanted to make this "I predicted, that this will happen" comment.
Jay Scott on :
Joseph Huang on :
MarcoDBAA on :
I think my approach is superior in sustaining and enlarging the community.
Yes, there are more high quality matches at first (just randomly), if you disable weaker bots. But a diminishing community (and less innovation, because of new bots being descendents of older bots) can even change this positive aspect to the opposite in the long run. And doesn´t it look like that now?
And I really don´t get it. Open up the voting site, when you watch, and only vote for the top 10?
A real problem are weak bots, that lag, crash, and/or cannot finish the game, and I always wanted them to be disabled (and some of them were disabled because I mentioned it), if they weren´t updated anymore, or if the problem was too severe.
But disabling just for having a low elo isn´t helpful in my opinion. Even if you do not disable the newer weak ones at first, you create an environment, where authors do not believe, that their bot is wanted there (or do not even try to create one), but believe that they are too late to the party (or have to use an existing template), and/or believe that their bot is still too weak, and they do not upload and may get discouraged completely and stop working on it.
But well, I already said that here before. It just looks that my prediction will be correct.
Hao Pan on :
On the flip side, the seemingly strong bots which are shallow forks of existing top ones, have become a mild concern for me. From last year's fork clown (Locutoids) fiesta, to the new comer BetaStar (not sure if it's a shallow fork of Locutus, I hope it's not). In a game of it vs my bot, it did exactly the same build Locutus would, and I couldn't tell the difference between unit micro (especially goons'). This, plus the aforementioned opponent-specific strategies described by Jay, really cast shadows on a potentially well-intended research project. And there doesn't seem to be a clearly-defined way to regulate forks. This IMO, is not good to advance bot dev work further. My bot itself being a fork, I welcome more of such rules/regulations on this matter.
MarcoDBAA on :
Yes, there were/are other reasons too, to not try to create a bot myself (it is difficult and I am only a hobbyist programmer), but it is another discouragement for sure.
Ok, what is a weak bot?
1.
I already said, that bots should not crash too often, should not slow down the game, should be able to finish games (not only attacking the starting position of another player for example).
2.
I also think, that you could compare it to normal computer game AIs. A bot doesn´t have to be super strong, but it should be somewhat interesting. Even players that play a RTS on easy difficulty setting do not want the AI to only build marines or zerglings for example.
There are/were some weaker bots, that fit this description, being interesting enough, stable and playing the role of the beatable competition for newcomers.
Elo or Winrate is also a bad metric because both will fall even more of course, if you disable weaker bots. I talked about this here for the "Roman Danielis" bot:
http://satirist.org/ai/starcraft/blog/archives/803-new-bot-RedRum.html
Hao Pan on :
As to the two criteria you listed to determine a weak bot:
1. Completely agree. IMO these are some bottom lines on this matter.
2. Also agree. And I use the Blizzard AI for a lot of testing purposes. For I know, if my bot struggles with this AI, very likely it'll struggle vs other bots. Not necessarily the built-in AI is weak, but, it has its own problems which can help us to determine a bot's strength. For example, only doing 1-2 builds, and the micro is sub-optimal and often units get stuck.
I checked your article on Roman Danielis. And I agree once again, that "Elo or Winrate is also a bad metric". ELO/Win rate is more of an illusion. They are relative and change over time. And they sort of manipulate the focus here. Is it more about AI dev work, or more about competition and the vanity (to some extreme extent)?