Error: DataBackend did not return the queried rows correctly

[bommert]

I am encountering the error

Error: DataBackend did not return the queried rows correctly: 781 requested, 593 received.
        The resampling was probably instantiated on a different task.
This happened PipeOp performance's $train()

when I run the following code:

task = tsk("ames_housing")

# remove columns with missing values (not of interest to problem)
mi = task$missings()
keep = setdiff(names(mi[mi == 0]), task$target_names)
task$select(keep)

# create graph learner: 
# impact encoding -> filter for feature selection -> linear regression model
learner = lrn("regr.lm")
filter = flt("performance", learner = learner, 
  resampling = rsmp("holdout", ratio = 2/3), measure = msr("regr.rmse"))
enc_po = po("encodeimpact", affect_columns = selector_type("factor"))
filt_po = po("filter", filter = filter, filter.nfeat = 1)
gl = as_learner(enc_po %>>% filt_po %>>% learner)

# reampling the graph learner results in the error
resample(task, gl, rsmp("cv", folds = 5))

There seems to be no problem when there are no factor variables in the dataset, e.g. when task = tsk("mtcars") is used as task in resample() in the code above.

[mb706]

Thanks for the report! I believe this happens because the resampling is not cloned between different calls to the filter. It is an mlr3filters bug...

[mb706]

Can you check if #173 solves it?

remotes::install_github("mlr-org/mlr3filters@filter_performance_clone_resampling")

[bommert]

Yes, it solves the problem. Thank you very much!