Augmenting small training sets for machine learning5 days ago
What augmentation can and cannot do | Tabular data: 60 observations | Time series: forecasting from a short history | Checklist
No articles match
Controlling how close synthetic data are to the real data5 days ago
One knob: closeness | Record level: further from real records, same distributions | Distribution level: scenarios | Hitting a target distance | Use with augmentation
Evaluating and visualising synthetic data5 days ago
Everything at once: compare_synthetic() | Marginal fidelity | Dependence | General utility: can a classifier tell them apart? | Specific utility: do analyses give the same answers? | Disclosure risk: distance to closest record | Plots from a comparison object | References
Generating data from multivariate distributions5 days ago
Correlation matrices | Classical multivariate distributions | Copulas | Arbitrary margins: r_mvdist() | References
Getting started with SynthesizerPlus5 days ago
The model in one paragraph | Fit once, generate many times | Supported column types | Missing values | Stratified models | Controlling the dependence | Heavy tails: the t copula | Identifiers, model size and privacy | Multiple synthetic data sets | Where next
Reading and writing data: CSV, Parquet, HDF5 and more5 days ago
Supported formats | Writing and reading | Parquet and Feather (Apache Arrow) | HDF5 | JSON, NDJSON, Excel, SPSS, Stata, SAS, fst | Synthesising straight from a file to a file | Very large synthetic data sets
Regularising high-variance models with synthetic data5 days ago
The idea | How much synthetic data? | Where the gain comes from: bias and variance | Seeing the variance | Compared with bagging | Tuning the amount of synthetic data | Summary | References
Releasing confidential microdata as synthetic data5 days ago
The confidential file | Step 1: how risky would releasing the real data be? | Step 2: a synthetic file | Step 3: near-copies and the amount of noise | Step 4: set the noise to meet a risk threshold | Step 5: check the released file | Targeted noise | Good practice | References
Synthetic time series5 days ago
Copula-VAR | Several series | Bootstraps | Practical notes | References