News blog

pomp version 6.3 released

9 May 2025

pomp version 6.3 has been released to CRAN and will soon appear on a mirror near you. This release contains a relatively minor feature upgrade: the built-in examples that use simulated data now have a seed argument. One can adjust this argument to generate distinct datasets.

pomp version 6.2 released

16 April 2025

pomp version 6.2 has been released to CRAN and will soon appear on a mirror near you. This release contains only performance enhancements and internal changes needed to keep up with developments in R.

Specifically, the basic Kalman filter (kalmanFilter) has been refactored, resulting in a roughly three-fold speed up.

In addition, internal changes have been made in package C codes in anticipation of forthcoming changes to the C API for R.

pomp version 6.1 released

8 January 2025

pomp version 6.1 has been released to CRAN and will soon appear on a mirror near you. This release contains breaking changes as well as some additional features.

User-visible changes

New interface for `userdata` becomes mandatory

Since version 5.8.4, the manner in which one provides extra elements to basic model components (i.e., beyond parameters, latent state variables, covariates, time, and observations: the so-called userdata) has changed. During a grace period, the old method still worked, though it generated a warning. In versions 6.X, an error will be generated. To supply additional elements to the basic model components, pass them as elements of a named list via the userdata argument, which can be furnished to any elementary algorithm or estimation algorithm, and of course, to the pomp constructor itself.

Passing arguments by position now results in an error in most cases

In calls to pomp elementary and inference algorithms, it is now necessary to pass arguments by name and not by position. This has always been good practice, but from this release, calls that rely on the position of arguments will typically generate errors.

Feature enhancements

Initial value parameters in `mif2`

It is now possible to specify more than one lag in the ivp function, which is evaluated only when the mif2 perturbations are specified. See ?mif2.

Keeping a database of explorations

In conducting an extensive exploration of a likelihood surface, it is useful to maintain a database of places visited, together with associated likelihood values. The new function append_data assists in this. It simply appends a given data frame to an existing CSV file (creating the file if it does not exist).

Expectation of an Euler-multinomial random variable

The new function eeulermultinom gives the expectation of an Euler-multinomial random variable. There is also an interface to this function in the C API.

Other changes

The save.states option to pfilter has changed. See pfilter and saved_states for details. The deprecated options will still work for the present, but will generate a warning, with advice about how to change. Ultimately, these options will be removed.

pomp version 6

10 December 2024

The current release contains two breaking changes over pomp versions 5.11.X.X.

With version 5.8.4, the manner in which one supplies extra model elements (the so-called userdata, elements other than parameters, state variables, observations, and covariates that are available to the basic model components) was changed. The old way of doing things remained available, but using it generated a warning message, with the promise that the warning would someday become an error. That day has come. As of version 6, to supply userdata to the basic model components, one must explicitly use the userdata argument in pomp constructors, elementary algorithms, or inference algorithms. Users who have already adjusted their codes to eliminate the aforementioned warning should experience no change in pomp behavior.
While it has always been good practice in pomp function calls to pass arguments by name (as opposed to by position), with version 6.0.1, this becomes (almost) mandatory. This is meant to prevent a class of difficult-to-trace errors in which an unnamed argument is passed to a lower-level function and potentially ignored, silently. The chief exception to this rule is the first argument in many pomp functions, which is typically a pomp object or a data frame. Users who are in the habit of passing named arguments will not notice any change in pomp behavior.

Blog archive
Subscribe to this newsfeed

This software has been made possible by support from the U.S. National Science Foundation (Grants #EF-0545276, #EF-0430120), by the “Inference for Mechanistic Models” Working Group supported by the National Center for Ecological Analysis and Synthesis (a Center funded by N.S.F. (Grant #DEB-0553768), the University of California, Santa Barbara, and the State of California), and by the RAPIDD program of the Science & Technology Directorate, Department of Homeland Security and the Fogarty International Center, U.S. National Institutes of Health.

Top Back Close