Curious what happens with Yahoo fantasy sports. They have been investing in new features this year for the first time in what seems like a decade. Still have millions of active users.
Sleeper is awesome! They're definitely one that I'm looking at (along with ESPN). They currently only offer a read-only API, which means I can't actually set your lineups for you, but I may reach out to them to see if that could ever change. If it does I'd love to add support.
I'm sorry, I wish there were! Unfortunately only Yahoo offers an API that lets my service directly interact with your team. Out of curiosity, which fantasy provider do you use? I think I can find a way to support ESPN but it will take some time.
Good question. I should probably make this clearer on the site.
The short answer is that if a player is injured we won't put them into your lineup unless there's an open spot (ie. we'll never bench a healthy player for an injured one).
Long-winded: we build thousands of simulations of every NBA game based on teams' recent lineups and depth charts, and incorporate current injury data into these. This lets us do some really cool stuff, such as bump up the minutes projection (and thus the value) of guys who are next in line on the depth chart at the position of the injured player. We then optimize your lineups over these simulations to maximize the chance that you'll win your matchup.
We'll see what happens. The main dynamics at play are:
- You're right, if I build out some of the more interesting AI features such as live draft assistant it will become more computationally demanding.
- As long as volume is low, that's fine. If not, I'll have to answer this question earlier ;)
- This will always be a side project. I would only want to start charging if I was really providing enough value to users where it made sense to charge. That said, I hope that by picking the right problems to solve the site can be immediately pretty useful.
I built this tool to solve the most annoying thing in fantasy sports: losing because you forgot to set your lineup. At the start of every week it sets everything for you, which you can then go modify if you want. It announces any changes it makes in an email so that you’re never surprised, and includes nice-to-haves like projected statistics and the probability that you will win the matchup. If folks like the tool, I plan to build out more AI assistant features for use-cases such as the draft, trades and waivers.
Note: You’ll need a Yahoo fantasy team to try it (they’re the only provider with a fully-featured API).
In designing this, I really wanted to build something that was useful and respects the user above all else. I require an email address because I want to make sure you know whenever the AI acts on your behalf, but include a big “Delete Account” button right on the main page once you’re logged in. This will fully and permanently delete all of your information. I also went with a privacy-respecting analytics tool (plausible.io) and will not spam or sell any information whatsoever. You should feel comfortable trying this tool and then deleting your account if it’s not for you.
Background: I’ve always been interested in building AI tools that directly interact with humans. My previous work (data scientist with the Philadelphia 76ers and research scientist at Uber AI Labs) sort of naturally led me to this idea of trying to win at fantasy sports using a pure AI approach. This tool is really just the very first step toward that long-term dream, solving the most basic decision problem in the game.
Would love any feedback or ideas! Also plan to write a blog post or two detailing both the statistical approach as well as the tech stack. I’m not a SWE so particularly appreciate any tips there.
There’s some cool stuff from OpenAI and Uber AI Labs scaling up neuroevolution (disclosure: these were my former colleagues at Uber). Check out the work from Ken Stanley and Jeff Clune, plus Tim Salimans. They solved most of the Atari suite comparably to the DQN from Deepmind (these are RL tasks).
