In total, the current version of OpenAI Five has consumed 800 petaflop/s-days and experienced about 45,000 years of Dota self-play over 10 realtime months (up from about 10,000 years over 1.5 realtime months as of The International), for an average of 250 years of simulated experience per day. A steep slope after any of these indicates OpenAI Five adapting to that change depending on the change the evaluation may be unfair to the versions before. This graph evaluates all bots on the final game rules (1 courier, patch 7.21, etc)-even those trained on older ones. The graph is roughly linear, meaning that OpenAI Five benefited continually from additional compute (note this is a log-log plot, since the x-axis is logarithm of compute and TrueSkill corresponds roughly to exponential progress). OpenAI Five's TrueSkill as we've applied additional training compute, with lines demarcating major system changes (moving to single courier increasing LSTM size to 4096 units upgrading to patch versions 7.20 and 7.21 and starting to learn buyback). So we increased the scale of compute in the only way available to us: training for longer. But after The International, we'd already dedicated the vast majority of our project's compute to training a single OpenAI Five model. In many previous phases of the project, we'd drive further progress by increasing our training scale. OpenAI Five's victories on Saturday, as compared to its losses at The International 2018, are due to a major change: 8x more training compute. This isn't the end of our Dota work-we think that Dota is a much more intrinsically interesting and difficult (and now well-understood!) environment for RL development than the standard ones used today. We are retiring OpenAI Five as a competitor today, but progress made and technology developed will continue to drive our future work. But we think decreasing the amount of experience is a next challenge for RL. This limitation may not be as bad as sounds-for example, we used Rapid to control a robotic hand to dexterously reorient a block, trained entirely in simulation and executed on a physical robot.
![dota replay player dota replay player](https://clevercolorado.weebly.com/uploads/1/2/6/2/126266407/168413316.jpg)
The surprising power of today's RL algorithms comes at the cost of massive amounts of experience, which can be impractical outside of a game or simulated environment. The results exceeded our wildest expectations, and we produced a world-class Dota bot without hitting any fundamental performance limits.
![dota replay player dota replay player](https://storage.indoesports.com/images/strategi-dota-2-untuk-menjadi-pro-player-3.jpg)
![dota replay player dota replay player](https://7ckngmad.files.wordpress.com/2016/04/replay.png)
To build OpenAI Five, we created a system called Rapid which let us run PPO at previously unprecedented scale. It uses the same general-purpose learning code whether those numbers represent the state of a Dota game (about 20,000 numbers) or robotic hand (about 200). Dawnbreaker and generally unknown/unprocessed units lack images.OpenAI Five sees the world as a bunch of numbers that it must decipher. The viewer currently hardcodes a 7.23 map background, so loading up a match played on the newest patch will most likely look very odd. Mobile/tablet support is flaky at best, so for now best to try it on a desktop. Jumping ahead in the timeline is supported, but depending on the length of the jump may freeze up your browser. Seeking backwards in time is currently not supported.
![dota replay player dota replay player](https://i.ytimg.com/vi/g7ruT1bCG3Y/maxresdefault.jpg)
There are plenty of issues and enhancements currently logged on GitHub, but a few known issues are worth mentioning in particular: Keep in mind that replay files can get pretty massive, especially pro commentated ones (50MB+), so if you'd rather download a smaller replay, try this link for a pub game.Īs before: feedback, issue reports or code contributions are more than welcome! There's a getting started link on the website to find the pro match between Na'Vi and OG from the screenshot below. All this match and replay data is kept locally in your browser and is never uploaded anywhere. This should work as long as the match data is available through OpenDota's API and the replay is still available from Valve. The project has matured a bit, featuring a new website and I am happy to announce that the most popular request is now possible: to find matches by match ID, automatically download the replay and load them up in the viewer. Thanks so much to everyone who provided feedback on this initial version, Redditors and community members from ModDota and OpenDota alike. Three months ago I posted about the initial release of ReDota, a replay viewer to revisit past Dota 2 matches in a web browser, visualizing the game as an interactive map with units moving around as the game progresses.