ModelOriented / EMMA

Evaluation of Methods for dealing with Missing data in Machine Learning algorithms
23 stars 7 forks source link

Changes to "paradox" package will break NADIA #186

Open mb706 opened 10 months ago

mb706 commented 10 months ago

We are in the process of overhauling the paradox package, on which your NADIA package depends: At some point in the near future, we are going to merge this PR.

This will remove the ParamDbl, ParamInt etc. classes, instead only the ParamSet class will remain, representing the parameter space as a table. Constructing a ParamSet will then need to be done using the shorthand forms ps(), p_dbl(), p_int() etc. See this diff of PipeOpImputeOOR.R in mlr3pipelines as an example of what changes are necessary.

Note that ps(), p_dbl() etc. is already possible (and the recommended way of doing this!), we are just phasing out the old ParamDbl$new() way.


If it helps, I did this semi-automatically by running the following in the R directory of affected packages. Please check these commands before running them and check the result, I can't guarantee that it works as intended in your setup.

sed -i 's/\(^ *\)ParamInt\$new("\([^"]*\)", /\1\2 = p_int(/' ./*.R
sed -i 's/\(^ *\)ParamDbl\$new("\([^"]*\)", /\1\2 = p_dbl(/' ./*.R
sed -i 's/\(^ *\)ParamFct\$new("\([^"]*\)", /\1\2 = p_fct(/' ./*.R
sed -i 's/\(^ *\)ParamUty\$new("\([^"]*\)", /\1\2 = p_uty(/' ./*.R
sed -i 's/\(^ *\)ParamLgl\$new("\([^"]*\)", /\1\2 = p_lgl(/' ./*.R
sed -i 's/ParamSet\$new()/ps()/g' ./*.R
sed -i 's/ParamSet\$new(list/ParamSet$new(params = list/g' ./*.R
for f in ./*.R ; do awk '/ParamSet\$new\(params = list\($/ { sub(/ParamSet\$new\(params = list\(/, "ps(");  print; in_block=1;  next; } in_block && /^ *\)\)$/ { sub(/\)\)/, ")"); in_block=0; } { print }'  $f > tmp ; mv tmp  $f ; done

It is not perfect -- it e.g. fails when the closing parentheses of the ParamSet$new(params = list( are not on the same line (i.e. they are not ))), so I had to do some manual cleaning afterwards.

Some other changes could also be necessary: Instead of accessing param_set$params[["<id>"]]$<field>, it is now necessary to access param_set$<field>[["<id>"]]. Both of these were possible in the past, we are phasing out the first one now.

Please @ me if you have any questions, I will try to help.

eccentric003x commented 3 weeks ago

Hello! I'm encountering an issue when trying to use PipeOpMice from the NADIA package. When running the following example code, I receive an error.

library(NADIA) library(mlr3oml) task <- mlr3oml::OMLTask$new(55) task <- task$task graph <- PipeOpMice$new()%>>% PipeOpEncodeImpact$new() %>>% lrn('classif.glmnet') Error in isTRUE(lhs) : 找不到对象'ParamDbl'

Thank you for any assistance you can provide! The error seems related to the ParamDbl object, but I'm not sure if there's a specific configuration or setup required to resolve this. Any suggestions would be greatly appreciated.