The R-package RaProR can be used to calculate a sketch of a large data set. That is a substitute data set of the same dimension but smaller number of observations. As we show in [1], the sketch can be used to perform approximate Bayesian or frequentist linear regression. More specifically, the likelihood as well as the posterior will be close to the ones obtained on the original data. Any algorithm for the regression analysis will run much faster on the sketch than on the original data set given that its running time depends on the number of observations.
Available on CRAN: https://CRAN.R-project.org/package=RaProR.
[1] Geppert, LN, Ickstadt, K, Munteanu, A, Quedenfeld, J, Sohler, C: Random projections for Bayesian regression, In Statistics and Computing 27, 79-101 (2017). DOI 10.1007/s11222-015-9608-z.