What's a microprediction API?
When you publish numbers periodically to the community API (using the Python client or directly) you automatically lure reward-seeking forecasting algorithms that monitor streams and make short-term distributional predictions of future data values you are yet to publish.
Predictions produced range from a minute to an hour ahead. They can use whatever external data is known to them. Their success is reported on leaderboards
- although due to the combining mechanism
the whole is greater than the sum of the parts.
This method of "nowcasting" live data is eventually hard to beat
. Anyone, anywhere in the world, can launch an algorithm to improve the overall quality by modifying and running a crawler
, which is merely a Python script - usually.
We also produce offline ratings
of autonomous open-source prediction algorithms. These power so-called forever functions
which stack the best performing algorithms drawn from many different open-source packages.
Microprediction.com is an educational site for humans. The action takes place at microprediction.org
where machines fight it out. If you are researching time-series and want to release your new method into the wild, or if you wish to collaborate on open-source timeseries algorithms (such as the timemachines
package, or maybe benchmarking your own) then consider joining our slack
These efforts are sponsored by Intech Investments
out of the belief that open, collective, bespoke business optimization is vastly more powerful when models and feature spaces are shared. No single company or group can solve repeated prediction, but collectively we can make a dent.
If you'd just like to use free on-tap prediction, then what really matters is your imaginative use of this statistical power. That's up to you but here are some articles from our blog
providing in-depth descriptions of use cases. They consider the determination of the relationships between cryptocurrencies, the creation of surrogate models for a physical system, and the normalization of streaming social data.