pmse

The propensity mean squared error is defined as
\(\frac{1}{N}\sum_{i=1}^N(p_i-c)^2\), where \(c\) is the number of
synthetic records, divided by the sum of the number of synthetic and real
records.

Data is synthesized using a combination of inverse transform
sampling using the empirical quantile functions for each variable, and
then copying the rank order structure from the original dataset. The
syntesizer method has a tunable parameter allowing to gradually move
from realistic and possibly unsafe synthetic data to decorrelated data
of less utility.

Mark der Loo

synthesizer

Synthesize Data Based on Empirical Quantile Functions and Rank
Order Matching

Mark van der Loo

pmse function

<dl><dt>synth</dt>
<dd><code>[data.frame]</code> Synthesized data.</dd>
<dt>real</dt>
<dd><code>[real]</code> Data to compare with the synthesized data.</dd>
<dt>model</dt>
<dd><code>[character]</code> Model used to compute propensity scores. Options
are <code>"lr"</code>: logistic regression, and <code>"rf"</code>: random forest.</dd>
<dt>nrep</dt>
<dd><code>[integer]</code> Number of model repetitions to average the 
pMSE value over. Ignored for <code>lr</code>.</dd></dl>

Arguments

Compute the pMSE metric between synthetic and real data — pmse

<dl>

<dt>synth</dt>
<dd><code>[data.frame]</code> Synthesized data.</dd>


<dt>real</dt>
<dd><code>[real]</code> Data to compare with the synthesized data.</dd>


<dt>model</dt>
<dd><code>[character]</code> Model used to compute propensity scores. Options
are <code>"lr"</code>: logistic regression, and <code>"rf"</code>: random forest.</dd>


<dt>nrep</dt>
<dd><code>[integer]</code> Number of model repetitions to average the 
pMSE value over. Ignored for <code>lr</code>.</dd>

</dl>

pmse: Compute the pMSE metric between synthetic and real data

Description

Usage

Value

Arguments

Examples