To begin, the presenter’s work in this area is performed under a philosophy of open access. All of the work presented during this session, both in terms of the code and data, are openly available. Access to everything is through this site: https://covidcast.cmu.edu. The data server can be found at: https://github.com/cmu-delphi/delphi-epidata and the COVID-19-specific client libraries are at https://github.com/cmu-delphi/covidcast. And, here is a link to their group's website https://delphi.cmu.edu which shares everything beyond COVID (papers, blog, flu forecasting and nowcasting tools, etc.).
The Delphi research group was formed in 2012 at Carnegie Mellon University “to develop the theory and practice of epidemic forecasting, and its role in decision-making”. In 2019 it was awarded the CDC’s National Center of Excellence for flu forecasting. When the COVID-19 pandemic hit in 2020, this group and their work was clearly in place to make a difference right away. As a result, the Delphi group launched an effort called COVIDcast to help tracking and even forecast the pandemic. During this webinar, founding members of the Delphi group, Carnegie Mellon research scientists Roni Rosenfeld (Department of Machine Learning) and Ryan Tibshirani (Department of Machine Learning and Department of Statistics) provided an overview of the many parts of COVIDcast which in turn led a number of fascinating examples.
Roni began by providing an overview of the hierarchy of a variety of indicators that they have woven into this application as the basis for their analyses. From here he described the various components of the COVIDcast ecosystem that work together, everything from the data and software infrastructure that make up the foundation through a public API that then allows for display of real-time interactive maps and graphics and modeling. In addition, covidcast and evalcast R packages as well a covidcast Python package are made available so that others may fully use/leverage the tools we've built or may perhaps add to this work and its capabilities.
Ryan spent his time concentrating more on the technical aspects of COVIDcast. He first provided a quick overview of both the API and what is included in the R and Python packages. This led to a number of eye-opening examples of output that tracked the following indicators: deaths, hospitalizations, doctor’s visits, symptoms and mask use, as well as an interactive maps of new cases and cumulative cases of COVID in the United States. ALL of the examples included the snippets of relevant code so that others can work with these data! Perhaps the most exciting topic discussed was Ryan’s description of the work they have done regarding the development of forecasting.
Rob Tibshirani (Stanford University) led an active Question and Answer portion of the session. In fact, there were more questions than could be answered in the time provided. While some questions tended to be technical in nature, others were curious as to whether the CDC is using any of their forecasting information, or do they know if others are using their work. Another question asked about whether they have thought about indicators related to vaccine development or deployment. Still other questions got both Roni and Ryan to talk about where they see this work going in the future, the types of models they envision and the forecasting that could become available. Both presenters answered a host of questions responding to a number issues related to this work and the challenges they face. Please review the recording of the session below for all of the detailed discussion that took place.
Want to learn more? Mark your calendar for every 1st and 3rd Thursday from noon to 1 pm ET. (See the NISS website for event details and to register for these sessions!)
Below, please find a recording of this session along with a link to the slides that the speaker used. The slides not only provide you with the key points that were offered but also include links to additional resources that should not be ignored!
Recording of the Session
Slides used by the Speakers
Roni Rosenfeld and Ryan Tibshirani (Carnegie Mellon University)
“Delphi’s COVIDcast Project: An Ecosystem for Tracking and Forecasting the Pandemic”
About this Webinar Series
The COPSS-NISS COVID-19 Data Science webinar series is co-organized by the Committee of the Presidents of Statistical Societies (COPSS) and its five charter member societies (ASA, ENAR, IMS, SSC, and WNAR), as well as NISS. This bi-monthly webinar features the latest research that is positioned on the cusp of new understanding and analysis of COVID-19 pandemic data, and promotes data-driven research and decision making to combat COVID-19. Find out more about this series and view all the previous sessions on the Webinar Series page.