pdrhlik/parmice

Parallel Multivariate Imputation by Chained Equations

parmice

R package that speeds up multiple imputation very easily. You should be familiar with the mice package in order to use parmice.

Multiple imputation is a great example of a method that can be processed in parallel. The whole process consists of creating several data sets with imputed values. Each of the data sets is created independently - it doesn't matter in which order we create them. We can therefore create them in parallel.

Installation

You can install the package using devtools:

devtools::install_github("pdrhlik/parmice")

Usage

The easiest way to use parmice is as easy as:

imp <- parmice(data)

You can supply parmice the same arguments you would put in the mice call. The only argument that is new in parmice is ncores. This argument accepts a number of cores that will be used in the imputation process. It is set to the number of cores of your computer minus one by default. The number of cores are detected using the detectCores() function from the parallel package.

The number of imputed dataset will not always equal to m. That's because it depends on ncores. Each core will process exactly ceiling(m / ncores) which means that the total number of imputed datasets will be ncores * ceiling(m / ncores). That's because we want all the cores to work equally. It is therefore recommended that m should always be a multiple of ncores.

Remember that m == 5 by default!

Examples

m == ncores: Number of imputed data sets will be 5 * ceiling(5 / 5) == 5

imp <- parmice(data, ncores = 5, m = 5)

m > ncores: Number of imputed data sets will be 5 * ceiling(6 / 5) == 10

imp <- parmice(data, ncores = 5, m = 6)

m < ncores: Number of imputed data sets will be 5 * ceiling(4 / 5) == 5

imp <- parmice(data, ncores = 5, m = 4)

Performace comparison with mice

Graphs will be presented soon.

Author

Patrik Drhlik - Initial work - pdrhlik

Don't hesitate to contribute if you have any ideas on how to fix or improve the package.

License

This project is licensed under the GPL-3 License.

Acknowledgments

This project wouldn't be possible without the following parties:

• The University of Queensland, Australia
• Technical University of Liberec, Czech Republic
• NESSIE Erasmus Mundus project