Jump to content

Talk:AlphaStar (software)

Page contents not supported in other languages.
From Wikipedia, the free encyclopedia

Beating MaNa and losing to Beastyqt

[edit]

It should perhaps be mentioned that when Alphastar was playing against MaNa, not only did alphastar have a seemingly unlimited APM, it also had a constant full map view - as opposed to a player who can only see a given part of the map. https://towardsdatascience.com/the-science-behind-alphastar-714bd7824d4b https://hub.packtpub.com/google-deepminds-ai-alphastar-beats-starcraft-ii-pros-tlo-and-mana-wins-10-1-against-the-gamers/

Alphastar didn't have unlimited APM. There was still a "fog of war" rather than the full map, but yes, AlphaStar could pay attention to a zoomed-out view of the map rather than having to zoom in on individual regions. If you can give us a quote from a strong WP:RS on the topic we can insert it (or you can insert it yourself).
Sorry, I didn't mean to imply that AlphaStar does not experience "fog of war". I think it's a little misleading talking of having to "zoom in on individual regions". A player is restricted to a certain part of the map, where as AlphaStar (only) had access to a view of the whole map.
I found a better source - deepmind lol ("How AlphaStar plays and observes the game" - the third section)
"During the matches against TLO and MaNa, AlphaStar interacted with the StarCraft game engine directly via its raw interface, meaning that it could observe the attributes of its own and its opponent’s visible units on the map directly, without having to move the camera - effectively playing with a zoomed out view of the game."
https://deepmind.com/blog/article/alphastar-mastering-real-time-strategy-game-starcraft-ii
The apm in this fight exceedes 1500 at some points and hovers ~1000 for several seconds at other points.
This is obviously related to AlphaStar controlling the units at 2-3 fronts simultaneously (Here the zoomed out map probably helps a lot), something any human player would find near-impossible to do effectively.
From 1:40:59 to 1:42:19 (APM comparison can be seen at the bottom)
https://www.youtube.com/watch?v=cUTMhmVh1qs&feature=youtu.be&t=6059
And as deepmind also mention, though without any data, in the same section as the camera comments:
"ALSO NOTE THAT ALPHASTAR'S EFFECTIVE APM BURSTS ARE SOMETIMES HIGHER THAN BOTH PLAYERS."
Sounds good, I added some material sourced to the ars technica source already in the article, which appears to go over some of that already. Rolf H Nelson (talk) 06:18, 3 May 2020 (UTC)[reply]

Secondly at Blizcon 2019 Beastyqt, a good player, but far from one of the best in the world, beat Alphastar 7-3, in a restricted Starcraft 2 client (Some options that is always turned was not available - like always showing units health bars). It has to be said though that Beastyqt has a lot of experience playing against a large amount of normal SC2 AI's, finding ways to exploit their behavior. https://www.youtube.com/watch?v=HBQkiT2raAQ — Preceding unsigned comment added by 152.115.42.3 (talk) 07:01, 30 April 2020 (UTC)[reply]

We also will need a WP:RS to include this information as well, otherwise it's WP:UNDUE trivia. Rolf H Nelson (talk) 03:49, 1 May 2020 (UTC)[reply]
Unfortunately I cannot find any other sources than the "replays" themselves (The youtube video above)