Columbia-PRIME/pcpr

PCP function in R.

pcpr

PCP functions in R including adaptations for environmental data.

to run:

To use this package, clone the repo or download the .zip file. Locate the folder, unzip if applicable, and use the following code. Replace path_to_folder with your local path.

install.packages("path_to_folder", repos = NULL, type="source")
library(pcpr)

includes:

1. stable_pcp

   • This includes a non-negativity constraint on the L solution matrix. It does not include a LOD (limit of detection) penalty.
   • It takes 4 inputs:
     • D the original dataset
     • lambda parameter
     • mu parameter
     • verbose parameter (optional)
   • For more info: ?stable_pcp

2. pcp_lod

   • This includes a non-negativity constraint on the L solution matrix and a separate penalty function for values <LOD.
   • Values <LOD should be pre-processed as -1.
   • It takes 5 inputs:
     • D the original dataset
     • lambda parameter
     • mu parameter
     • LOD may be a scalar, vector, or matrix
     • verbose parameter (optional)
   • For more info: ?pcp_lod

3. root_pcp

   • This changes the objective function by removing the squaring of the error term so that the mu parameter does not rely on the unknown sigma value. It does not include a non-negativity constraint on the L matrix or a LOD penalty.
   • It takes 4 inputs:
     • D the original dataset
     • lambda parameter
     • mu parameter
     • verbose parameter (optional)
   • For more info: ?root_pcp

4. root_pcp_nonnegL

   • This includes a non-negativity constraint on the L matrix with the squareroot version of the objective function. It does not include a LOD penalty.
   • It takes 4 inputs:
     • D the original dataset
     • lambda parameter
     • mu parameter
     • verbose parameter (optional)
   • For more info: ?root_pcp_nonnegL

5. root_pcp_na

   • This allows for missing values with the squareroot version of the objective function. It does not include a LOD penalty or a non-negativity constraint on the L matrix.
   • Missing values should be pre-processed as NA.
   • It takes 4 inputs:
     • D the original dataset
     • lambda parameter
     • mu parameter
     • verbose parameter (optional)
   • For more info: ?root_pcp_na

6. root_pcp_na_nonnegL

   • This includes a non-negativity constraint on the L matrix and allows for missing values with the squareroot version of the objective function. It does not include a LOD penalty.
   • Missing values should be pre-processed as NA.
   • It takes 4 inputs:
     • D the original dataset
     • lambda parameter
     • mu parameter
     • verbose parameter (optional)
   • For more info: ?root_pcp_na_nonnegL

7. root_pcp_na_nonnegL_lod

   • This includes a non-negativity constraint on the L matrix, allows for missing values, and includes a separate penalty function for values <LOD with the squareroot version of the objective function.
   • Missing values should be pre-processed as NA.
   • Values <LOD should be pre-processed as -1.
   • It takes 5 inputs:
     • D the original dataset
     • lambda parameter
     • mu parameter
     • LOD may be a scalar, vector, or matrix
     • verbose parameter (optional)
   • For more info: ?root_pcp_na_nonnegL_lod

8. root_pcp_noncvx

   • This replaces the nuclear norm in the objective function with a projection to a lower rank. It does not include a LOD penalty or a non-negativity constraint on the L matrix.
   • It takes 5 inputs:
     • D the original dataset
     • lambda parameter
     • mu parameter
     • r the desired rank
     • verbose parameter (optional)
   • For more info: ?root_pcp_noncvx

9. root_pcp_noncvx_nonneg

   • This replaces the nuclear norm in the objective function with a projection to a lower rank. It includes a non-negativity constraint on the L matrix. It does not include a LOD penalty.
   • It takes 5 inputs:
     • D the original dataset
     • lambda parameter
     • mu parameter
     • r the desired rank
     • verbose parameter (optional)
   • For more info: ?root_pcp_noncvx_nonneg

10. root_pcp_noncvx_w_na

    • This replaces the nuclear norm in the objective function with a projection to a lower rank. It does not include a LOD penalty or a non-negativity constraint on the L matrix. It does allow missing values.
    • Missing values should be pre-processed as NA.
    • It takes 5 inputs:
      • D the original dataset
      • lambda parameter
      • mu parameter
      • r the desired rank
      • verbose parameter (optional)
    • For more info: ?root_pcp_noncvx_w_na

11. root_pcp_noncvx_nonneg_w_na

    • This replaces the nuclear norm in the objective function with a projection to a lower rank. It includes a non-negativity constraint on the L matrix. It does not include a LOD penalty. It does allow missing values.
    • Missing values should be pre-processed as NA.
    • It takes 5 inputs:
      • D the original dataset
      • lambda parameter
      • mu parameter
      • r the desired rank
      • verbose parameter (optional)
    • For more info: ?root_pcp_noncvx_nonneg_w_na

12. root_pcp_na_nonneg_noncvx_LOD

    • This replaces the nuclear norm in the objective function with a projection to a lower rank. It includes a non-negativity constraint on the L matrix and a LOD penalty. It does allow missing values.
    • Missing values should be pre-processed as NA.
    • It takes 5 inputs:
      • D the original dataset
      • lambda parameter
      • mu parameter
      • r the desired rank
      • LOD may be a scalar, vector, or matrix
      • verbose parameter (optional)
    • For more info: ?root_pcp_na_nonneg_noncvx_LOD