datDota

Recently upgraded servers to a new Ryzen 9 3900 setup and doubled up total RAM and storage. Key cause for upgrade was increased performance needs (especially with the AD sub-site), but also a postgres 13 upgrade, a better backup solution, and general hardware replacement. 1/3

A bit of downtime - adding some new data tracking before TI starts!

Once the BetBoom Dacha ticket has been released we'll move the games (currently played on the EPL World Series: SEA ticket) over.

Sorry, had some unexpected downtime on @windrunio earlier today.

Sorry we just had some downtime as one particularly long game was unable to parse within a threshold and was killing the RAM usage.

@Dota_Science @Noxville @LGDgaming @virtuspro @Team__Spirit Updated elo32 btw

@Noxville @LGDgaming @virtuspro @Team__Spirit @datDota The friend, who did it got the task ~ at 15:00 yesterday with no prior knowledge of dota2, elo, glicko etc. At 05:33 he sent me some graphs saying all the other versions were «degenerate» and only elo_32 (3 spikes) looked reasonable. He is obviously sleeping now, so...

Elo has been too generous to CIS this time: according to Monte Carlo simulations the highest probabilities to win #ti10 are: @LGDgaming - 30% @virtuspro - 26% @Team__Spirit - 21% The only problem is all the simulations use ELO_32 from @datDota, which seem not to be updated =(

Did some maintenance tonight: [Pro stats + AD stats] * Added 7.30 (abilities/spells/talents) * Fixed up three older broken games (if you notice stats Δ1) [AD] * Retro-added 7.29d * Fix for parsers just dying * Leaderboard # games req. dropped to 150 (from 75th percentile)

Replay parser bug fixed up, should be caught up in a few minutes. Overall data quality this year has been pretty good - only 19 missing/broken replays (out of 2987 games, so 0.63% error rate).

Due to a bug in clarity related to the new patch, there might be a delay in parsing games with match_id >= 6055758217. Will try get the issue fixed upstream but could take a day or so - then will catch up missing games. Pesky undecodable teams!

Mostly caught up with the AD backlog. Ended up not just throwing more hardware at it, but also rewriting the parser manager to be quite a bit more efficient. Popularity is skyrocketing. Currently ~4k in the queue (although ~500 of them are too recently received to be parsed).

@arian_taheri Popularity in China mostly, apparently the top streamers are all playing it at the moment.

Due to the super high load in Ability Draft, we've got a bit behind (~55k) in matches. As a result we've just tripled the number of parsers running and underlying IO/compute resources. Will see if this makes a dent on the backlog.

There's some Steam API issue on some endpoints we use (GetMatchDetails specifically). This is affecting pro matches, but not Ability Draft ones (we only get basic data for Ability Draft games) - so pro matches will be delayed for a bit.

Valve configured their missing CDNs, so some replays are available from the affected EU regions - but sadly some are lost. Of pro matches that's 18 missing. I spoke to the CEO at my work (@BayesEsports) and he was okay with me ingesting the match data from our db - huge props!

Users have increasingly been asking for a way to donate to the site - and whilst the costs are not that high to run the site (<$150/mo), I finally got round to setting up a Ko-Fi page so people who are interested to drop a few dollars when they want. ko-fi.com/datdota 3/3

This should also satisfy replay + audio storage needs for the next ~3 years, and give room for Source 1 commentary audio. On the Ability Draft side we'll be announcing something soon™ If you notice anything weird on either the main site or abilitydraft.datdota.com poke us! 2/2

Recently upgraded servers to a new Ryzen 9 3900 setup and doubled up total RAM and storage. Key cause for upgrade was increased performance needs (especially with the AD sub-site), but also a postgres 13 upgrade, a better backup solution, and general hardware replacement. 1/3

Spent the last week or so on a passion project: a high level, game-play independent, player-centric rating system for @DOTA2 based on historical performance, based off of @datDota 's Professional + Premium games dataset. This spans from May 2012 to end of December 2020.

Earlier today, Aster took on IG.Vitality in what was the 50,000th professional game played of Dota 2. Let's look back at some numbers over that time ...

After introducing the bug ~16 months ago, I've finally fixed the one random query endpoint which effectively acts as a memory leak - slowly killing the server. Annoying that a single borked line can create so much chaos!

'Pucker Up For A Punch' - a discussion about teamfight data in Dota 2. medium.com/datdota/pucker…