|
I made a webpage that can filter and generate some statistics from selected SC2 tournaments (2019-2020 and those that have released replay packs freely).
Here are unfiltered stats currently available (imgur albums):
• Proxy locations (Barracks, Stargates and Photon Cannons built in the first 5 minutes and in at least 40 distance from owner's starting location) • Various charts • Pie charts for unit types killed/created/lost in total number or equivalent resource value
As always, data is the main limiting factor. I'm not claiming this will provide some deep insight into the game. With StarCraft II, you either have data that are relevant, but the dataset is too small, or the dataset is of good size but gathered across too wide skill level and different metagames.
Webapp Webpage for filtering and generating stats. It might be slow as it's a free host – especially if you are generating more charts. I'm sorry about that.
https://tournamentanalysis.herokuapp.com/
Filtering There are several ways to filter, and it can be combined. By:
• Tournament • Map • Matchup • Minimum and maximum game length • Date interval in which games were played • Minimum number of certain unit types created during each game
Few examples
Other notes If there is a replay pack that I missed, or someone has a replay pack that isn't freely available and would want to add it here, let me know. The same if you have an idea for interesting stats to add or a way to filter.
Added tournaments: Assembly Summer 2019, Cheeseadelphia XI, HomeStory Cup XIX, HomeStory Cup XX, IEM Katowice 2020 - Server Qualifiers, IEM Season XIII Katowice, Nation Wars 2019, QLASH StarCraft II Invitational 2 Offline Finals, WCS 2019 Global Finals, WCS Fall 2019, WCS Spring 2019, WCS Summer 2019, WCS Winter 2019 Americas, WCS Winter 2019 Europe, WESG 2018 Grand Finals
Accompanying post on my blog.
Main Python libraries used: sc2reader, flask, pandas, matplotlib.
|
That's sick. For proxy locations is there a way to differentiate which side of the map that proxying player spawned?
|
Pretty solid nice work, keep it up! I think there's a lot to achieve.
|
On February 22 2020 13:00 ZigguratOfUr wrote: That's sick. For proxy locations is there a way to differentiate which side of the map that proxying player spawned? Not right now. I set a limit for the distance from owner's starting location to be at least 40. So early Barracks/Stargate/Cannons in natural or third base won't be shown. I think you can guess most proxies by their position. They will be closer to your opponents location by ground for Barracks or by air for Stargates.
On February 22 2020 18:43 derkopf wrote: Pretty solid nice work, keep it up! I think there's a lot to achieve. Thank you. I will add more tournaments as they go. ESL already released replay packs for open qualifiers, so I'm hoping they will post replays from the main event after the next week. If anyone has a good idea for additional stats, let me know.
|
I added an option to see how winrates change in games where players have invested at least certain amount of resources into each unit type. This doesn't mean a unit with positive change is necessary good, it is more complicated than that. However, it's interesting to see the numbers and think of causes behind those numbers.
Here are two examples for Terran showing 100 and 2000 resources invested into each unit type. This is without any filtering, so all games from 2019 onward.
• Imgur album for all races. • On the webpage you can set resource amount to any value and filter games.
|
Really cool graphs, thanks for the hard work
|
|
Some improvements:
- Added an option to filter games where only certain race won
- Expanded "hero unit" and "worker slayer" options
- Some new charts (examples from IEM)
- Added ability frequency for some abilities. Accuracy of some data might vary, but those abilities that creates units (e.g. Purifier Nova) are accurate.
Few examples from IEM games:
265 Oracles used: 246 Revelation (0.93 per Oracle) 159 Stasis Ward (0.6 per Oracle) 591 Phoenixes used: 1899 GravitonBeam (3.21 per Phoenix) 433 Disruptors used: 830 Purifier Nova (1.92 per Disruptor) 8 Motherships used: 9 Time Warp (1.12 per Mothership) 2 Mass Recall (0.25 per Mothership)
|
Added few new charts:
• Typical upgrade/research time • Median units created in each non-mirror matchup (~typical amount of units of that type you will see if the unit type is used)
|
|
|
|
|