archive by month
Skip to content

Steamhammer at 78%

I looked up Steamhammer’s standing on SSCAIT today and was startled to see it at 78% win rate, equal with Bereaver.

standings of Bereaver and Steamhammer

The winning percentages are volatile and I don’t believe for a minute that Steamhammer is as good as Bereaver. But I have never seen it fluctuate up that high, either. Apparently the 1.0 version is substantially stronger. It is definitely better than 0.2, but how much?

Update: I documented Steamhammer 1.0’s configuration file. Because of the new features I added, somebody already expressed interest in forking from Steamhammer rather than directly from UAlbertaBot. Go for it!

Trackbacks

No Trackbacks

Comments

krasi0 on :

I don't know exactly what you have done there, but the high win% and the interest to fork from SH are proof that it's working!

At some point these days I hope to get around to export another DB dump so that you can calculate the current ELO ratings (in case you still keep those perl scripts) which should give us more insight into the up to date strength ranking.

Once I do that, would you be willing to compile and host something like https://purplepie.bitbucket.io/sscait.html for the ongoing matches with an additional ELO column? For the purpose of continuous updates to the ranking, you could just use a web crawler and a local DB similarly to how the author of purple pie has done it. At some point, we might embed it back into SSCAIT in some form, but the admin of SSCAIT is usually very busy, so I am not sure when that could be done.
What are your thoughts?

Jay Scott on :

Of course I have the scripts! My thought is that I should test the API again to see whether it is complete enough now. I know for sure that it has been updated since I first looked at it. I expect it is good now, and I can make a web page that updates not too far behind real time.

MicroDK on :

After I read this, I used a couple of hours to create this list based on the last 1000 games using the API. :D
http://scbw.holdorf.dk/sscait/ratings.php
Every bot starts at ELO 2000 and K-factor is 30.

Jay Scott on :

Nice job, and helpful for everybody! It looks like there is not enough data to fully separate the top and bottom bots, but with so many bots updated there’s no help for it.

MicroDK on :

With the API you can retrieve pages of maximum 1000 games. I have now added another 1000 games so the list is based on the last 2000 games and now the bots are a bit more separated. But it takes a little more time to get and go through all 2000 games. ;)
Most bots will have played at least 60 games and with bots being updated all the time I thinks it gives a good picture of the strength of the bots.

AIL on :

I was actively helping you be beating Bereaver twice today! ;)

Jay Scott on :

Yes: Meanwhile, AILien actually does beat Bereaver. Congratulations! You found a vulnerability. I wonder how long the window of vulnerability will be open? This version of Bereaver does not seem to be reacting to mutalisks as it should, and it doesn’t seem that hard to fix.

AIL on :

It's not so much that I found a vulnerability in Bereaver, it's more that I fixed vulnerabilities and improved unit-handling in AILien and it then went ahead to find Bereavers weaknesses on its own by adapting accordingly based on general algorithms.

This also is my general approach: Instead of trying to exploit weaknesses in my enemies' play, I want to avoid weaknesses in my own play and then just make the best use of my own strengths'.

For example your hint with the gas. AILien now has a general algorithm to a) not put so many drones into gas, when it already has a stockpile and b) even pull out some drones, when the Stockpile exceeds certain limits (I'm gonna change that limit though, as now it pulls them out of the geyser a bit too soon for my taste). The extra-minerals are put to good use and make it stronger.

A big weakness is how badly it uses Overlords. I want to revamp that completely and make it into a scouting-monster. I already have a good algorithm to decide where to attack: A score that goes up with drones and non-combat-buildings and goes down with distance and defenders. However, when the bot doesn't really scout that well and rather wastes it's overlords pointlessly, it will not find the possible targets to determine their juiciness.

I'm also thinking about adding more Squads with different target-priorities:
A squad that only contains the quicker units could prioritize units, that are slower than themselves, that are out in the open, over enemy bases.

I'm also unhappy with the UAB-shuffle and aim to somehow build concaves instead of that.

Well, overall, there's just so much to improve, that will keep me busy for a long time before I run out of ideas. Just need to try and sort them by a good metric of playing-strength-improvement/time-investment. ^^

Jay Scott on :

Your approach is good. All I really mean is that you can’t beat an opponent (on average) unless the opponent makes mistakes. The loser made mistakes, and it’s worth analyzing what the mistakes are.

MicroDK on :

I am with you there... more squads with different and better targets, taking gas on and off and using overlords better... That is also my thoughts on my upcomming bot. ;)

LetaBot on :

Hey Jay, I am currently writing up a small piece on the TL blog about the "family tree" of Bots. Which bot expressed interest in your version (if the author ofc doesn't mind you telling that) ?

Jay Scott on :

Cool chart, I like it a lot! I don’t know the bot name, only the author’s name—I guess we’ll have to wait and find out.

Jay Scott on :

I have permission to say. The author is Micky Holdorf and the bot is planned to be named Microwave.

MicroDK on :

Yes, and it will be a Zerg bot... not rushing. ;)

LetaBot on :

Looks like I will need to update the "family tree" picture from my recent blog post on TL based on that update :)


http://www.teamliquid.net/blogs/518904-family-tree-of-starcraft-bots

krasi0 on :

The following is an interesting conclusion: "So, the success of a bot mainly comes down to the skill/knowledge of the programmer. Using the code of other bots can help you get started, but the main factor for success still remains the ability to add new techniques. Simply taking the best technique(s) of each open source bot and putting them together in one giant bot won't get you 1st place at a StarCraft AI tournament."

krasi0 on :

It'd be even better if it works via the API. Does this mean that you could in theory retrieve the whole game history? Please keep me posted.

Jay Scott on :

It looks like the API is complete for bot information. Somebody could duplicate the bot list part of the score page with a nicer UI, or keep a repository of historical binaries, or whatever. But the game information seems to be not working. :-(

AIL on :

Just got queued against SH and Killerbot on my request.

Lost to SH but beat Killerbot :D

And now will revert my new concave-building, which kinda worked but really is not good in ZvZ. I'd say it's what made me lose against you because it cost too many lings early on.

Add Comment

E-Mail addresses will not be displayed and will only be used for E-Mail notifications.

To prevent automated Bots from commentspamming, please enter the string you see in the image below in the appropriate input box. Your comment will only be submitted if the strings match. Please ensure that your browser supports and accepts cookies, or your comment cannot be verified correctly.
CAPTCHA

Form options

Submitted comments will be subject to moderation before being displayed.