2.6 KiB
CUSF Standalone Predictor - Version 2
Working on improving the Cambridge University Spaceflight landing predictor, a web-based tool for predicting the flight path and landing location of latex balloons.
Install
The source for the predictor itself is in pred_src/ and instructions for building it can be found there.
The following items need to be executable (chmod +x ./predict.py) by the user under which the predictor runs:
predict.pypred_src/pred(once compiled)cron/clear-pydap-cache-cronjob.shcron/purge-predictions-cronjob.sh
The predict/preds/ and gfs/ directories need to have rwx access by the PHP interpreter and the predict.py python script. You will need to install the following python packages: pydap, numpy, json, simple-json. We use at to automatically background the predictor, so you will need that installed.
Other than that, just clone this repo to a non web-accessible folder and create symlinks to the predict/ and hourly-predictions/ directories in the repo.
There are useful configuration options in predict/includes/config.inc.php.
Information
The two bash scripts in the cron/ directory should both be run daily. clear-pydap-cache-cronjob.sh clears the cache used by pydap so that old data does not build up. purge-predictions-cronjob.sh deletes scenarios and predictions not accessed or modified within the last 7 days. Re-running a prediction for a scenario will therefore reset its time to live to 7 more days.
The directory names are UUIDs comprised of an SHA1 hash of the launch parameters, and re-running predictions will overwrite data in the existing directory, rather than create a new one.
We use GFS data provided by the NOAA, accessed via NDAP and their NOMADS distribution system. The 1.0x1.0 degree data (26 vertical pressure levels) is used for standard predictions, and the 0.5x0.5 degree data (47 vertical pressure levels) is used for the high definition (HD) predictions.
Credits & Acknowledgments
Credit as detailed in individual files, but notably:
- Rich Wareham - The new predictor and the hourly predictor system
- Fergus Noble, Ed Moore and many others
Adam Greig - http://www.randomskk.net - random@randomskk.net
Jon Sowman - http://www.hexoc.com - jon@hexoc.com
Copyright CUSF 2010 - All Rights Reserved