NEWS.md

In umich-cphds/lodi: Limit of Detection Imputation for Single-Pollutant Models

lodi v0.9.2

Minor changes

• Added badges to README.Rmd / README.md to show the CRAN version of lodi, Github version of lodi, and a Travis-CI badge showing whether or not lodi tests are passing. If the Github version and CRAN versions differ, the badges appear as different colors.
• Added some unit tests to be run when updating the package / before CRAN submission.
• Added check to ensure exposure data that is below the lod is coded as NA. Users might input an already log transformed exposure in clmi, while not transforming the lod variable. If the transformation is logarithmic, this will catch it
• Changed the optimization algorithm uses in the MLE procedure from L-BFGS-B to BFGS with reparameterized variance (ie exp(var)) to ensure non-negativity instead of box constraints. Using L-BFGS-B could cause non-finite values of the objective function due to large gradients when evaluating the initial value of the function.

Bug fixes

• Fixed a subtle bug when calling lodi inside of a function when using a custom exposure function. lodi was internally using the parent frame to enclose the transformation, which is the lodi package namespace, instead of the caller's frame.

lodi v0.9.1

Minor changes

• Added additional checks to clmi to protect against improper variable ordering.
• Added verbose option to clmi to print out parsing information.
• Updated clmi documentation to provide additional exposition on proper use of the package.

Bug fixes

• Corrected calculation of censored likelihood maximum likelihood estimates, and named them.
• clmi now returns imputed data.frames that have the same row ordering as the original.

lodi v0.9.0

Initial CRAN release.

umich-cphds/lodi documentation built on Feb. 23, 2020, 9:43 a.m.