![]() ![]() DotA's mysterious and anonymous developer, known only as IceFrog, was employed by Valve to work on what is, in essence, a brand new version of the same game. The original DotA was itself inspired by a StarCraft mod, called Aeon of Strife. Know your Dota lineage Dota 2 itself is a sort of official sequel to Defense of the Ancients, a modification to Warcraft 3. ![]() It's the biggest game on Steam, boasting more concurrent players than any other title on Valve's DRM and distribution platform, regularly breaking 300,000 concurrent players. In this way, it grew organically as gamers gave keys to friends or sold them for inflated prices. Even with this limited availability and a learning curve that's almost horizontal (in that additional time doesn't actually gain you much additional skill), Dota 2 has become hugely popular. You can sign up and Valve will e-mail you when you can start playing.īefore its "official" (though still limited, see sidebar) release this week, Dota 2 has been available for the last two years as an invitation-only beta, with beta keys randomly assigned to existing players. The game is being rolled out through a staggered launch, both to ensure that the game isn't inundated with people who don't know what they're doing and also to ensure that the matchmaking and game servers can tolerate the influx of new players. This means that if the leak is slower compared to the rate of adaptation of the learning algorithm, the algorithm will constantly track the leak as a normal change of behavior.Staggered releaseAlthough Dota 2 officially launched out of beta this week, Valve has stopped short of making it a free-for-all where new players will get slaughtered by beta veterans. Adaptiveness is critical when measuring businesses, as nothing is static. All known methods for modeling time series for anomaly detection (from ARIMA, Holt-Winters, LSTMs, etc), estimate trends as part of the process of learning the normal behavior and must be adaptive to small changes in the time series behavior. However, it might never get detected at the hourly timescale. We could argue that if we waited a day, the leak would show up on the hourly timescale. Is multi-scale analysis really necessary? The adaptation/detection tradeoff In this case, the increase in crashes was detected by automatically analyzing the same metric (number of crashes for iOS devices and one version of the app) at multiple time scales - although the leak was slow at the hourly time scale, and did not cause anyone hour to be anomalous, it showed up as a significant anomaly at the daily time scale, enabling early detection. These leaks typically appear as a change in trend in the metrics - revenues, conversion rate, etc. For example, metrics measuring usage of a feature, number of checkout completions, or churn rates should show gradual declines or increases.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |