Page 5 of 8
Posted: Sun Jan 10, 2010 6:54 am
The current problem with improving my script, other than not have enough time, is that it simply puts so many resources into tech that it doesn't have enough to build any sizable force to do some sort of push until what is essentially late game when it maxs tech (this applies to both counter Zerg and counter Protoss). Which obviously is a problem since an opponent that can notice this can get 3-4 expands without being in any real danger, but this is more of a strategic issue than an efficiency one so it's not something that can be fixed just by making it do the same thing faster.
Here are the thing's that I have noted in general and from watching your replays,
-as you said it should early expand if opponent is; it probably can trigger off of cannon detection, also might be able to expand earlier if cybernetics core is detected though I'm not sure about this.
-I haven't checked the numbers but I've been wondering if the infantry armour has any segnificant effect against protoss since their units are mostly high damage mitigating armor usefullness.
-I've been trying to decide if it needs more barracks or more factories and am stuck on the issue that the AI uses mech rather poorly and the production times are rather bad but infantry doesn't seem very effective once the protoss has tech'ed up (tempar/reaver/carrier).
-The issue with expanding is that I had it expand a lot so that in the case an expansion got killed it wouldn't mine out without trying to expand again. I could try a different expand control format similar to its first expand but expand control is a pain to try and work with since it has a tendency to fail.
-I've been wondering if other AI's get the top expand on that map properly since it seem strange that the bottom AI expands correctly but the top one seems to get stuck (I don't see any scv going there and I've checked and the 2nd ex is actually a forced one so it can't be getting stopped by troop number).
-it should stop firebat and vulture production on fleet beacon detect.
I'd appreciate further idea or comments on these but it could take a while for changes to get implemented since it's no longer christmas break so I can't turn out a version every two days that is actually better than those before it.
Posted: Sun Jan 10, 2010 8:54 am
Posted: Sun Feb 14, 2010 5:13 am
I think I've fixed my AI so that it doesn't adversely affect other AIs.
I'd appreciate some help in verifying that other AIs perform normally when running at the same time as Accel_T v1.49, as I was under the impression until recently that the previous version didn't adversely affect other AIs. However I am now fairly certain that it does.
Posted: Sun Feb 14, 2010 9:03 am
[quote name='Equalizer' post='9258' date='Feb 13 2010, 09:13 PM']I think I've fixed my AI so that it doesn't adversely affect other AIs.
I'd appreciate some help in verifying that other AIs perform normally when running at the same time as Accel_T v1.49, as I was under the impression until recently that the previous version didn't adversely affect other AIs. However I am now fairly certain that it does.[/quote]
Was there anything specific that you did or was it just moving around things until they didn't mess up the AI running alongside it?
I mean, it looks just about the same as the last one, with just some minor changes, just going by my memory.
Posted: Sun Feb 14, 2010 11:06 am
Have you downloaded the latest 2/12 ver iof bwailauncher ?? The size limitation bug was fixed so thus the problems might go away with the extra space.
I'm somewhat drunk right now, but I'll go play a game or two maybe.
edit:uploaded.... one of them didn't expand for some reason. But it was still very hard.
Posted: Sun Feb 14, 2010 3:38 pm
The main difference is that it doesn't spam multiruns for unit production, so its marco isn't quite the same which required changes to the script in general to compensate for that.
While other AIs do function when running concurrently with the previous version they seem to get slowed down considerably. In general it seem that if multirun calls build up they start slowing everything down even in the best case. (Worst case is that they make the multiruns in other scripts get missed).
I've been mostly using Archon_Wing's protoss for testing this since it uses multiruns throughout the duration of the game for various things and it was noted in one of the AIBashes that it didn't play properly when running with AccelDefault v1.1x. I use the 3v3 replays and make sure that nothing stranged happens when having that many AI's going at once.
Also I found that having an expansion call repeated over and over in a loop while it does ensure that the expansion gets placed it causes problems with how all other AIs in the game function.
Though it seems that it will need some form of backup plan to ensure the expansions get placed. Since Archon_Wing found that the first expansion can sometimes be missed.
(I think his drone might have confused the AI /unsure.gif' class='bbc_emoticon' alt=':unsure:' /> )
I believe its infinite loops particularly ones that have, multiruns, expand, or build statements imbeded in them, that lead to strange behaviour. I've tried adding back the rerunning of expand to get the first expand down and it doesn't seems to cause issues long as the loop is finite under normal conditions and ends relitivly quickly.
Posted: Mon Feb 15, 2010 12:30 am
For the AI bash, I actually think it's more with the map being crap. Unleaded AIs use a lot of multiruns but they were still working very well in their bash even though we had 4 in the game. Though obviously your script is more complex , I think we do underestimate how much stuff the AI can do at once.
One thing that helps is to place wait(1) between large groups of multiruns. I think it forces the AI to stop and read the lines. Otherwise the AI reads ahead of itself and weird crap happens, causing things to be ignored.
The script randomly not taking an expo I swear is a hardcoded issue. I've seen it with a number of scripts. You can try increasing the priority of an expo cc so it'll go ahead and stop other stuff for it. For 2 rax expo I I didn't bother running a resource_check for it to expand, instead just forcing it in.
Overall, there's pretty much no more dispute anymore; this is easily the best script for tvz. One issue I noted was that marine range was gotten really late which makes it very weak to someone with better micro, though the bigger issue here is a lurker attack. I think a earlier bunker, 2nd comsat and turret at the natural could help.
Posted: Mon Feb 15, 2010 1:58 am
I really don't like bunkers since once the AI puts units in them it almost never takes them out which really hurts the AIs early game if lings can by-pass the bunker. I'll see if having one being built later on or if lurk tech is detected is benifical.
Multiruns are a real pain as they are incredibly unpredictable. I tried pushing them all to the beginning of the script so if the AI can't handle X number of multiruns it should be obvious fairly early in the game. I get the impression though its more whats in the multirun than how many of them there are when dealing with multiruns initializing at the start.
As for expands I think I've issolated the expand scriping issue down to if the AI thinks its under attack, enemy units in its base, it will ignore any expand orders. I think a time triggered forced expand might help the AI try to grab an expand before it runs out of resources, as that seems to be a bit of an issue.
Regarding academy tech I find the AIs use of stim against mutas is actually fairly good so I have it get that first then range. Also it tends to be rather low on gas since it is also is trying to upgrade infantry wep. Perhaps it should pospone the wep upgrade to get range faster. Any opinions on that idea?
Posted: Mon Feb 15, 2010 3:41 am
Yea, getting stim and than range is typical but both need to be done ASAP. +1 weapon is much less important because that takes a while to research. In the above rep, my attack was very late due to me losing an ol early, so take that into consideration.
It's just that the AI does not preemptively stim, making it less useful. It's still highly effective and needed. I think it's easier for a script to hold stuff off with range. This is even more true with vs toss where unranged marines are completely useless vs ranged goons.
It's not necessary to have a bunker, but I would recommend at least 1 missile turret upon lair tech detection and can add more with spire detection. The main issue is that the AI does not scan sometimes if lurkers are attacking buildings
Posted: Fri Feb 19, 2010 1:58 am
that unused labels can cause a huge disaster.
I'm not sure what happened here, but I think a single unused label caused my entire Protoss script to simply break. The Protoss script by itself does not break, but once I copied it over to Entropy III, the results were bad. PYAI has had a problem with unused labels for quite a while.
For this reason, I would ask if you have the time to remove unused labels from your script to avoid this from possibly happening to your script in Entropy III. Thanks. Your script will be even deadlier when mixed in with others.
Here's a rep of 1.52. One major gap in Accel T's otherwise flawless defense is 9 pool. It actually does get the raxes up fast enough, but it built too many depots.
Unforunately one of their terrans decided to build their CC in a funny spot so it didn't do as well as it could.