Portal Forecsts

Portal Forecasting Website

This is the main repository for predictions made on the Portal rodent census data Portal Project.

Predictions are made and archived weekly. Approximately once a month, one of these forecasts is made immediately prior to a new trapping session (trapping occurs as close to each new moon as possible).

The website code is now located in its own repository.

How to add a new model

Modeling is driven by the portalcasting package. New models should be added there following instructions in the "adding a model" vignette.

Docker builds

Forecasts are run using continuous integration based on a docker image. This makes the builds faster and more reproducible. The code in this repo uses the latest portalcasting image

Developer notes

This code runs weekly on the UF HiPerGator using a cron job on daemon2 using Ethan White's account. The cron job runs a version of portal_weekly_forecast.sh that is separate from the one in the repository, but generally just a copy of it. The version of portal_weekly_forecast.sh in the repo is automatically updated to the one that is run on HiPerGator after the weekly forecasts are complete.

portal_weekly_forecast.sh does the following:

• Updates the portal-forecasts repository to it's current version
• Updates the forecasts repository to it's current version (this repository is used for archiving full forecasts)
• Runs the forecasts
• Pushes the results of the forecasts in the forecasts and models and data directories to the portal-forecasts repository
• Pushes the results of the forecasts in the forecasts and models and data directories and also the fits directory to the forecasts repository
• Tests to see if forecasts ran correctly

portal_dryrun_forecast.sh is automatically run twice each week to check to see if the forecasts are working prior to a production run. It does the same thing as portal_weekly_forecast.sh but does not actually push a release.

The root directory for all work is /orange/ewhite/PortalForecasts/.

There are 4 log files:

• portal_weekly_forecast_log.out - the main log file for the weekly forecast
• portal_dryrun_forecast_log.out - the main log file for the dryrun forecast
• testthat.log - the log file for the tests of whether the forecasts ran correctly
• cron.log - the cron log file, which is just a list of SLURM submissions

If necessary to create a fresh setup this system:

1. Create a root directory (PortalForecasts)
2. Clone portal-forecasts into that directory
3. Clone forecasts into that directory
4. Copy portal_weekly_forecast.sh and portal_dryrun_forecast.sh from portal-forecasts into the root directory.