Zoltar Home API

    Welcome to the Zoltar forecast archive, a web application to demonstrate ideas for a repository of model forecast results along with tools to compare, etc. The main ideas are outlined on the About page.

    Projects API

    The archive has 3 projects:

    Project Owner Description Summary
    CDC Flu challenge
    nick Guidelines and forecasts for a collaborative U.S. influenza forecasting project.
    21 models, 4724 forecasts, 23.4 million rows (estimated), 27.0 million rows (actual) as of 2019-07-23
    Impetus Province Forecasts
    nick Impetus Project forecasts for real-time dengue hemorrhagic fever (DHF) in Thailand. Beginning in May 2017, this project contains forecasts for biweekly DHF incidence at the province level in Thailand. Specifically, each timezero date is associated with a biweek in which data were delivered from the Thai Ministry of Public Health to servers in the US. We use standard biweek definitions described in the supplemental materials of Reich et al. (2016). Each timezero also has a data-version-date that represents the day the forecast model was run. This can be the same as the timezero, but cannot be earlier.

    Files follow the naming conventions of `[timezero]-[modelname]-[data-version-date].cdc.csv`, where dates are in YYYYMMDD format. For example, `20170917-gam_lag1_tops3-20170919.cdc.csv`.

    For each timezero, a forecast contains predictive distributions for case counts at [-1, 0, 1, 2, 3] biweek ahead, relative to the timezero. Predictive distributions must be defined according to this binned-interval structure:{[0,1), [1, 10), [10, 20), [20, 30), ..., [1990, 2000), [2000, Inf)}.

    One model, 37 forecasts, 81,585 rows (estimated), 73,405 rows (actual) as of 2019-07-23
    TEST cdc flusight network (No logo) khouse This is a project to test the cdc flusightnetwork upload functionality.
    Two models, 179 forecasts, 2.4 million rows (estimated), 1.1 million rows (actual) as of 2019-07-23