It is easy to write programs that look for streams to predict and submit scenarios.
0. Install microprediction
pip install microprediction
Or see microprediction
1. Get a key
See muid instructions
. Note especially the table of bankruptcy levels in the microprediction package's README
. At minimum, visit muid.org
to create a minimal strength MUID for prediction submission.
2. Run the default crawler
from microprediction import SimpleCrawler
crawler = MicroCrawler(write_key="YOUR_WRITE_KEY_HERE")
3. Watch it run
You can use the dashboard
to monitor performance. Performance is also available via the API, or Python client:
You are advised to read the MicroWriter code
, the MicroCrawler code
and package readme. There is also a series of articles
on LinkedIn covering topics such as crawler navigation
, predicting bivariate
streams and overview
of the mechanics of prediction and reward.
4. Improve it
Take a look at how this example
modifies the all important sample
method of the crawler. Crawlers provide samples, which can be interpreted as percentile estimates. Any time series model can be shoe-horned into a crawler. It is worth reviewing this article
also that explains the mechanics of this site and scoring.
You can also read this article
how to change the nagivation of the crawler, thus determining which streams it visits and how far ahead it decides to try to predict.
5. Let it go further afield
If you derive from MicroCrawler rather than SimpleCrawler you can visit zscore, bivariate and trivariate streams. The z1~ streams are described in the article
mentioned above. You can also read this article
about bivariate prediction of badminton, or this article
about trivariate prediction of crytocurrencies to understand the z2~ and z3~ streams you see.
6. Let it run indefinitely
There are many ways to do this. One is suggested towards the end of this article