dataprep: upgrading from 0.1.5 to 0.1.7 • dataprep

library(dataprep)

What changed

Three behaviour changes affect row and column counts. All three are bug fixes, but each one changes the output on real data. If your downstream analysis depends on exact row counts, read the quantified comparison in the last section before upgrading.

Interface changes

Besides the behaviour changes above, several functions changed their argument interface between 0.1.5 and 0.1.7. The old start / end pair was replaced by a single cols argument (character names, integer indices, or logical mask):

Function	0.1.5 arguments	0.1.7 arguments
`varidele`	`start`, `end`	`cols`
`obsedele`	`start`, `end`	`cols`
`condextr`	`start`, `end`	`cols`
`percoutl`	`start`, `end`	`cols`
`optisolu`	`start`, `end`	`cols`
`dataprep`	`start`, `end`	`cols`
`descdata`	`start`, `end`	`cols`
`descplot`	`start`, `end`	`cols`
`percdata`	`start`, `end`	`cols`
`percplot`	`start`, `end`	`cols`
`shorvalu`	`start`, `end`	`cols`
`melt`	`cols` (id columns)	`id` / `measure.vars`

A call such as varidele(data, 3, 15) was interpreted as start = 3, end = 15 in 0.1.5, but in 0.1.7 it is parsed as cols = 3 (the 15 is dropped or becomes fraction). You must rewrite it as varidele(data, cols = 3:15). The same applies to every function in the table.

obsedele() now scans each column independently. The 0.1.5 implementation collapsed all selected columns into one long vector before computing missing runs; this changed NA run boundaries and could both over-delete boundary rows and retain rows that should have been deleted. The 0.1.7 implementation scans each column independently: a row is deleted when any selected column has a missing run longer than half minutes on both sides.
The half-minute boundary is now inclusive. Rows whose nearest anchor is exactly half minutes away are retained (“within half minutes” is a <= condition).
optisolu() no longer crashes with cores > 16. The 0.1.5 parallel::makeCluster() path exhausted memory when the worker processes each received a full copy of the input. The 0.1.7 implementation loads the package on each worker, exports the input data only once per worker, and runs each (interval, times) case in a separate task, so cores = 64 and cores = NULL (automatic) are both safe. Note that the optimal parameter values returned by optisolu() may differ slightly between 0.1.5 and 0.1.7 because the underlying outlier-marking and observation-deletion backends have changed. Re-run optisolu() after upgrading if exact parameter values matter.

The full description is in news(package = "dataprep"). The design reasoning behind the pipeline is in vignette("dataprep-philosophy").

Minimal reproduction of the boundary change

Consider five observations sampled every 10 minutes, with a valid value only at the two ends:

df <- data.frame(
  date = as.POSIXct("2024-01-01 00:00:00", tz = "UTC") + 0:4 * 600,
  x    = c(1, NA, NA, NA, 5)
)

obsedele(df, cols = "x", half = 30)
#>                  date  x
#> 1 2024-01-01 00:00:00  1
#> 2 2024-01-01 00:10:00 NA
#> 3 2024-01-01 00:20:00 NA
#> 4 2024-01-01 00:30:00 NA
#> 5 2024-01-01 00:40:00  5

The three interior rows are each 10, 20, or 30 minutes from the nearest valid anchor. Under 0.1.5 the third row was deleted because the delete condition was inclusive (dl >= half); under 0.1.7 it is retained because the delete condition is strict (dl > half && dr > half), so dl == half satisfies “within half minutes”.

Minimal reproduction of the multi-column change

Consider two channels with mutually exclusive missing runs:

df <- data.frame(
  date = as.POSIXct("2024-01-01 00:00:00", tz = "UTC") + 0:9 * 600,
  x    = c(1, NA, NA, NA, NA, NA, NA, NA, NA, 5),
  y    = c(NA, NA, NA, NA, 2, NA, NA, NA, NA, NA)
)
nrow(obsedele(df, cols = c("x", "y"), half = 60))
#> [1] 10

Under 0.1.5, the x and y NA runs were merged before computing the run length. This changed the run boundaries and could either over-delete rows or retain rows that should have been deleted. Under 0.1.7 the two channels are checked independently: a row is deleted when any selected column has a run longer than half minutes on both sides.

Implementation evolution

The retention criterion has been stable across releases, but the implementation has improved steadily. The table below summarises the three generations:

Release	Strategy	Complexity
0.1.0	Borrowed running mean: expand the series onto a regular grid with `tidyr::complete()`, compute a 59-minute centred moving average on a temporary column, and use its emptiness pattern to flag long runs.	`O(grid length)` per subset
0.1.5	Run-length encoding: use `data.table::rleid()` and `rowid()` to collapse consecutive `NA`s into runs, then compare each run length to the number of grid points covered by `half` minutes.	`O(n)` time, `O(n)` temporary storage
0.1.7	Anchor scan: for each missing value, look up the nearest non-missing anchor on each side and compare the two time distances directly. No grid, no run-length state.	`O(n)` time, `O(1)` extra allocation per column

Each generation produces the same deletion decision on the same input, but the constant factors shrink. The 0.1.7 anchor scan is the first version that is fast enough to run interactively on full-year data: on SMEAR I Varrio 2025 (49,422 rows × 61 numeric channels), obsedele() now runs in 0.05 s against 11.6 s in 0.1.5 on Ubuntu 25.10 (about 232×), and in 0.035 s against 22.5 s on Windows 11 Pro for Workstations (about 648×). See vignette("dataprep-performance") for the full benchmark.

Quantified effect on a full-year dataset

On SMEAR I Varrio 2025 (49,422 rows × 61 numeric channels, 10-minute sampling), running the same pipeline with the same parameters:

Stage	0.1.5	0.1.7	Δ
`varidele`	25 columns deleted	25 columns deleted	0
`obsedele`	1,494 rows deleted	1,496 rows deleted	+2
`condextr`	1,868 rows deleted	1,863 rows deleted	−5
`shorvalu`	50,376 NAs filled	50,387 NAs filled	+11
`dataprep` final	46,060 rows	46,063 rows	+3

Net change: 0.006% of the input. The six rows that differ between versions all sit at run boundaries where the anchor distance is within one sampling interval of half minutes.

Where the difference comes from

The six differing rows fall into two groups:

Rows kept by 0.1.7, deleted by 0.1.5 (3 rows). These are rows whose nearest anchor is exactly half minutes away. Under the 0.1.5 delete condition (dl >= half) they were deleted; under the 0.1.7 condition (dl > half && dr > half) they are retained. Each of these rows has a valid anchor within one sampling interval of half, so retaining them is consistent with the physical constraint described in vignette("dataprep-philosophy").
Rows deleted by 0.1.7, kept by 0.1.5 (3 rows, overlapping with the above). These are rows that pass the check on some columns but fail on at least one. Under 0.1.5 the merge-columns approach effectively widened the anchor window on these rows; under 0.1.7 each column is checked independently, so the row is deleted. These rows would have been interpolated across a gap longer than half minutes in at least one channel, which contradicts the design.

The net result is 3 additional rows in the final output.

Migration checklist

Re-run any downstream analysis that uses the output of obsedele(), condextr(), percoutl(), or dataprep(). The three-row difference on SMEAR I Varrio 2025 is representative of the magnitude you should expect on other datasets (well under 0.1% of rows).
If you use prep_fit() / prep_transform(), re-fit the plan on the new training data. The stored thresholds are not affected by the change, but the row indices that fed into them are.
If you were limiting optisolu() to cores <= 16 as a workaround for the 0.1.5 crash, remove the cap. 0.1.7 accepts up to 64. Also re-run optisolu() because the returned optimal parameters may differ slightly from 0.1.5.
If downstream behaviour depends on specific boundary rows, verify the difference with dplyr::anti_join() between the 0.1.5 and 0.1.7 outputs. The example below shows how.

result_015 <- dataprep::dataprep(data, cols = 5:65, group = 4)
result_017 <- dataprep(data, cols = 5:65, group = 4)

kept_only_by_017 <- dplyr::anti_join(result_017, result_015, by = "date")
kept_only_by_015 <- dplyr::anti_join(result_015, result_017, by = "date")

What did not change

For users who only call melt() and dcast(), or who only use dataprep for descriptive statistics (descdata, na_diagnose, percdata, percplot, descplot), there is no behaviour change. Those functions have been re-implemented in 0.1.7 for speed (melt and dcast) or reorganised internally (data_report, dry_run), but the output on every tested input is byte-identical to 0.1.5 except where noted in the news file.

Performance reference

The 0.1.7 release rewrites every heavy cleaning routine in C++. The table below compares against 0.1.5 on three dataset sizes from the same source (SMEAR I Varrio forest). All numbers are speed-up ratios (0.1.5 time / 0.1.7 time); a value below 1.0× means 0.1.7 is slightly slower on that cell.

Function	500 rows	7,640 rows	49,422 rows (Ubuntu 25.10)
`varidele`	1.1×	1.1×	11.6×
`obsedele`	203×	424×	232×
`condextr`	196×	217×	1146×
`optisolu`	188×	77×	109×
`dataprep`	185×	228×	247×

On Windows 11 Pro for Workstations, the same full-year pipeline gives obsedele ≈ 648×, condextr ≈ 839×, shorvalu ≈ 81×, optisolu ≈ 25× (at cores = 32), and the integrated dataprep call ≈ 173×. varidele is around 1.17× on this cell; this is expected, since varidele is a single colMeans(is.na(.)) in both versions and the new code path has little room for improvement.

Note on optisolu cores. The 0.1.5 implementation could crash when cores > 16. The benchmark above used cores = 16 for both versions to keep the comparison fair. 0.1.7 loads the package on each worker, exports the input data once per worker, and runs each (interval, times) case as a separate task, so cores = 64 is safe. The practical speed-up on a many-core host is larger than the table above. Also note that the optimal parameter values returned by optisolu() may differ slightly between versions; re-run optisolu() after upgrading if exact values matter.

Test environments

Benchmarks and checks were run on two reference hosts. Only the core configuration is listed here; full details are in vignette("dataprep-performance").

Ubuntu 25.10 — R 4.5.1, g++ 15.2.0; 2× AMD EPYC 9965 192-Core (384 physical / 768 logical cores), 1.0 TiB (16 × 64 GiB Micron, DDR5-5600, Multi-bit ECC), full AVX-512.
Windows 11 Pro for Workstations — R 4.6.1 (ucrt), GCC 14.3.0; 2× AMD EPYC 7B12 64-Core (128 physical / 128 logical cores), about 224 GiB RAM.

Where to go next

Design philosophy — vignette("dataprep-philosophy"). Why the pipeline has the shape it does.
Cleaning pipeline walkthrough — vignette("dataprep-cleaning"). Step-by-step execution on a real dataset.
Performance and cross-engine consistency — vignette("dataprep-performance"). Full benchmark tables and 8-engine consistency checks.
Fast reshaping — vignette("dataprep-melt-dcast").

Session info

sessionInfo()
#> R version 4.6.1 (2026-06-24)
#> Platform: x86_64-pc-linux-gnu
#> Running under: Ubuntu 24.04.5 LTS
#> 
#> Matrix products: default
#> BLAS:   /usr/lib/x86_64-linux-gnu/openblas-pthread/libblas.so.3 
#> LAPACK: /usr/lib/x86_64-linux-gnu/openblas-pthread/libopenblasp-r0.3.26.so;  LAPACK version 3.12.0
#> 
#> locale:
#>  [1] LC_CTYPE=C.UTF-8       LC_NUMERIC=C           LC_TIME=C.UTF-8       
#>  [4] LC_COLLATE=C.UTF-8     LC_MONETARY=C.UTF-8    LC_MESSAGES=C.UTF-8   
#>  [7] LC_PAPER=C.UTF-8       LC_NAME=C              LC_ADDRESS=C          
#> [10] LC_TELEPHONE=C         LC_MEASUREMENT=C.UTF-8 LC_IDENTIFICATION=C   
#> 
#> time zone: UTC
#> tzcode source: system (glibc)
#> 
#> attached base packages:
#> [1] stats     graphics  grDevices utils     datasets  methods   base     
#> 
#> other attached packages:
#> [1] dataprep_0.1.7
#> 
#> loaded via a namespace (and not attached):
#>  [1] vctrs_0.7.3       cli_3.6.6         knitr_1.52        rlang_1.3.0      
#>  [5] xfun_0.61         otel_0.2.0        generics_0.1.4    textshaping_1.0.5
#>  [9] jsonlite_2.0.0    glue_1.8.1        htmltools_0.5.9   ragg_1.5.2       
#> [13] sass_0.4.10       rmarkdown_2.32    tibble_3.3.1      evaluate_1.0.5   
#> [17] jquerylib_0.1.4   fastmap_1.2.0     yaml_2.3.12       lifecycle_1.0.5  
#> [21] compiler_4.6.1    dplyr_1.2.1       fs_2.1.0          pkgconfig_2.0.3  
#> [25] Rcpp_1.1.2        systemfonts_1.3.2 digest_0.6.39     R6_2.6.1         
#> [29] tidyselect_1.2.1  pillar_1.11.1     parallel_4.6.1    magrittr_2.0.5   
#> [33] bslib_0.12.0      tools_4.6.1       pkgdown_2.2.1     cachem_1.1.0     
#> [37] desc_1.4.3