Why FakeDataR?
- Problem: Many users can’t run large models locally. To get help from an LLM or share examples, they hand-edit real datasets, slow, brittle, and risky for privacy.
- Solution: FakeDataR runs entirely on your machine (R only). It reads your real data locally and generates a synthetic copy that mirrors structure (schema, types, factor levels, ranges, missingness). Sensitive fields can be masked or dropped. You can safely upload the synthetic dataset to an LLM or share with collaborators, without exposing the original data.
What makes it different?
- Local by default: No cloud round-trips; your real data stays on your machine.
- Schema-faithful: Preserves column types, factor levels, NA patterns, and numeric ranges.
-
Privacy controls: Heuristics for sensitive fields (e.g., IDs, emails, phones) with strategies (
fake,drop). - LLM bundles: Exports fake data + JSON schema + README prompt in one bundle.
- DB schema support: Build synthetic data from a database schema without pulling real rows.
Related work exists (e.g., general fake-data generators, schema tools). FakeDataR focuses on privacy-preserving, schema-faithful synthesis aimed at LLM workflows and reproducible sharing.
Install
install.packages("FakeDataR") # CRAN
#> Installing package into 'C:/Users/Zobaer Ahmed/AppData/Local/Temp/RtmpeemZPG/temp_libpath2864498820af'
#> (as 'lib' is unspecified)
#> installing the source package 'FakeDataR'
# devtools::install_github("zobaer09/FakeDataR") # development (optional)Quick start
library(FakeDataR)
# 1) Start from a built-in data frame for reproducibility
df <- head(mtcars, 50)
# 2) Generate a synthetic copy
fake <- generate_fake_data(
data = df, n = nrow(df),
category_mode = "preserve",
numeric_mode = "range",
column_mode = "keep",
normalize = TRUE
)
# 3) Handle sensitive fields (auto-detect + fake)
fake_priv <- generate_fake_with_privacy(
data = df, n = nrow(df),
level = "low", sensitive_detect = TRUE,
sensitive_strategy = "fake"
)
# 4) Export an LLM bundle (fake data + schema + README prompt)
bundle <- llm_bundle(
data = df, n = min(100L, nrow(df)),
level = "medium",
formats = c("csv","rds"),
path = tempdir(), filename = "demo_bundle",
seed = 42, write_prompt = TRUE, zip = TRUE
)
bundle$zip_path
#> [1] "C:\\Users\\ZOBAER~1\\AppData\\Local\\Temp\\Rtmpc34k7g/demo_bundle.zip"When to use / not use
Use when:
- You need an LLM or collaborator to reproduce your data shape but not see the real values.
- You want quick, documented synthetic data for demos, tutorials, bug reports, or tests.
Avoid when:
- You need formal differential privacy guarantees (FakeDataR focuses on pragmatic masking and synthesis).
- You need exact statistical fidelity for modeling benchmarks. It is not a drop-in replacement for originals.