README.md

In FakeDataR: Privacy-Preserving Synthetic Data for 'LLM' Workflows

FakeDataR

📘 Website • Articles • Reference

FakeDataR makes safe, synthetic stand‑ins for real datasets. It mirrors types, factor levels, NA/blank rates, and even handles sensitive columns (IDs, emails, phones) via fake or drop strategies. You can also generate fake tables from a database schema without reading any rows.

Installation

# install.packages("devtools")
devtools::install_github("zobaer09/FakeDataR")

Quick start

Generate a fake dataset that matches the structure of a real one:

library(FakeDataR)

fake_mtc <- generate_fake_data(mtcars, n = 200, seed = 1)
head(fake_mtc)
#>        mpg      cyl     disp        hp     drat       wt     qsec        vs
#> 1 16.63945 5.070033 335.2440 282.43325 4.623352 3.588993 17.57882 0.2396067
#> 2 19.14491 4.874581 145.2945 314.84395 2.834732 4.191491 20.72770 0.6477649
#> 3 23.86205 6.067187 453.7102  93.73714 4.867064 3.012021 22.34145 0.9756708
#> 4 31.74288 5.075802 431.0475 264.19953 4.376889 5.247958 20.15496 0.3779988
#> 5 15.13953 4.724673 449.4281 328.11103 3.352964 1.975893 20.39140 0.4641441
#> 6 31.51216 6.074305 361.2276 327.86627 4.229320 1.665920 21.62005 0.8122963
#>           am     gear     carb
#> 1 0.13853856 3.123219 7.102635
#> 2 0.04752457 3.710643 7.770379
#> 3 0.03391887 4.154076 7.068414
#> 4 0.91608902 4.070063 4.064007
#> 5 0.84020039 4.208546 2.343565
#> 6 0.17887142 3.972298 1.576061

Validate that classes and basic patterns were preserved:

validate_fake(mtcars, fake_mtc)
#>    column class_original class_fake class_match na_prop_original na_prop_fake
#> 1     mpg        numeric    numeric        TRUE                0            0
#> 2     cyl        numeric    numeric        TRUE                0            0
#> 3    disp        numeric    numeric        TRUE                0            0
#> 4      hp        numeric    numeric        TRUE                0            0
#> 5    drat        numeric    numeric        TRUE                0            0
#> 6      wt        numeric    numeric        TRUE                0            0
#> 7    qsec        numeric    numeric        TRUE                0            0
#> 8      vs        numeric    numeric        TRUE                0            0
#> 9      am        numeric    numeric        TRUE                0            0
#> 10   gear        numeric    numeric        TRUE                0            0
#> 11   carb        numeric    numeric        TRUE                0            0
#>    na_match blank_prop_original blank_prop_fake blank_match
#> 1      TRUE                  NA              NA          NA
#> 2      TRUE                  NA              NA          NA
#> 3      TRUE                  NA              NA          NA
#> 4      TRUE                  NA              NA          NA
#> 5      TRUE                  NA              NA          NA
#> 6      TRUE                  NA              NA          NA
#> 7      TRUE                  NA              NA          NA
#> 8      TRUE                  NA              NA          NA
#> 9      TRUE                  NA              NA          NA
#> 10     TRUE                  NA              NA          NA
#> 11     TRUE                  NA              NA          NA
#>    range_within_original
#> 1                   TRUE
#> 2                   TRUE
#> 3                   TRUE
#> 4                   TRUE
#> 5                   TRUE
#> 6                   TRUE
#> 7                   TRUE
#> 8                   TRUE
#> 9                   TRUE
#> 10                  TRUE
#> 11                  TRUE

Preserve factor levels & numeric ranges (example with palmerpenguins):

if (requireNamespace("palmerpenguins", quietly = TRUE)) {
  peng <- na.omit(palmerpenguins::penguins[, c("species","island","bill_length_mm","sex")])
  fake_peng <- generate_fake_data(peng, n = 400, seed = 11, category_mode = "preserve")
  validate_fake(peng, fake_peng)
}
#>           column class_original class_fake class_match na_prop_original
#> 1        species         factor     factor        TRUE                0
#> 2         island         factor     factor        TRUE                0
#> 3 bill_length_mm        numeric    numeric        TRUE                0
#> 4            sex         factor     factor        TRUE                0
#>   na_prop_fake na_match blank_prop_original blank_prop_fake blank_match
#> 1            0     TRUE                   0               0        TRUE
#> 2            0     TRUE                   0               0        TRUE
#> 3            0     TRUE                  NA              NA          NA
#> 4            0     TRUE                   0               0        TRUE
#>   range_within_original
#> 1                    NA
#> 2                    NA
#> 3                  TRUE
#> 4                    NA

Sensitive columns

Detect and handle PII by name. Use strategy "fake" (default) or "drop":

df <- data.frame(
  id    = 1:50,
  email = sprintf("[e-mail]", 1:50),
  phone = sprintf("(415) 555-%04d", 1:50),
  spend = runif(50, 10, 500)
)

# Keep the columns but replace values with synthetic ones
fake_keep <- generate_fake_data(
  df, n = 80, seed = 12,
  sensitive_detect = TRUE, sensitive_strategy = "fake"
)
intersect(df$email, fake_keep$email)  # should be character(0)
#> character(0)

# Drop sensitive columns entirely
fake_drop <- generate_fake_data(
  df, n = 80, seed = 13,
  sensitive_detect = TRUE, sensitive_strategy = "drop"
)
names(fake_drop)  # no id/email/phone
#> [1] "spend"

From a database schema (no data read)

Create fake rows using only the schema (types, nullability, etc.). This works even when you cannot access the underlying data.

library(DBI); library(RSQLite)

con <- dbConnect(RSQLite::SQLite(), ":memory:")
dbExecute(con, "
  CREATE TABLE employees (
    id INTEGER,
    email TEXT,
    phone TEXT,
    is_active BOOLEAN,
    hired_at TIMESTAMP,
    salary NUMERIC,
    dept TEXT
  )
")

sch  <- schema_from_db(con, "employees")
fake <- generate_fake_from_schema(sch, n = 50, seed = 14)

str(fake$hired_at)  # POSIXct
dbDisconnect(con)

Exporting

Write fake data to common formats (CSV, RDS, Parquet) and optionally produce an LLM bundle containing a schema and README.

# CSV / RDS
export_fake(fake_mtc, file.path(tempdir(), "fake_mtc.csv"))
export_fake(fake_mtc, file.path(tempdir(), "fake_mtc.rds"))

# Parquet (requires the 'arrow' package)
# install.packages("arrow")
export_fake(fake_mtc, file.path(tempdir(), "fake_mtc.parquet"))

# End-to-end bundle
b <- llm_bundle(
  mtcars, n = 200, seed = 10, level = "high",
  formats = c("csv","rds"),
  path = tempdir(), filename = "mtcars_fake",
  write_prompt = TRUE, zip = TRUE
)
b$zip_path

Reproducibility

All generators respect seed. Given the same inputs and a fixed package version, results are reproducible:

a1 <- generate_fake_data(CO2, n = 123, seed = 42)
a2 <- generate_fake_data(CO2, n = 123, seed = 42)
identical(a1, a2)
#> [1] TRUE

Learn more

• Getting started vignette: https://zobaer09.github.io/FakeDataR/articles/getting-started.html
• Database workflow vignette: https://zobaer09.github.io/FakeDataR/articles/database-workflow.html
• Full function reference: https://zobaer09.github.io/FakeDataR/reference/

Contributing

Issues and pull requests are welcome! If you find a bug or have a feature request, please open an issue on GitHub.

License

This package is distributed under the terms of the license included in the repository (see LICENSE).

FakeDataR documentation built on Nov. 6, 2025, 1:15 a.m.