data_transformer: Data Transformer

In RGAN: Generative Adversarial Nets (GAN) in R

Data Transformer

Description

Provides a class to transform data for RGAN. Method $new() initializes a new transformer, method $fit(data) learns the parameters for the transformation from data (e.g. means and sds). Methods $transform() and $inverse_transform() can be used to transform and back transform a data set based on the learned parameters. Currently, DataTransformer supports z-transformation (a.k.a. normalization) for numerical features/variables and one hot encoding for categorical features/variables. In your call to fit you just need to indicate which columns contain discrete features.

Value

A class to transform (normalize or one hot encode) tabular data for RGAN

Methods

Public methods

• data_transformer$new()

• data_transformer$fit_continuous()

• data_transformer$fit_discrete()

• data_transformer$fit()

• data_transformer$transform_continuous()

• data_transformer$transform_discrete()

• data_transformer$transform()

• data_transformer$inverse_transform_continuous()

• data_transformer$inverse_transform_discrete()

• data_transformer$inverse_transform()

• data_transformer$clone()

---

Method new()

Create a new data_transformer object

Usage

data_transformer$new()

---

Method fit_continuous()

Usage

data_transformer$fit_continuous(column = NULL, data = NULL)

---

Method fit_discrete()

Usage

data_transformer$fit_discrete(column = NULL, data = NULL)

---

Method fit()

Fit a data_transformer to data.

Usage

data_transformer$fit(data, discrete_columns = NULL)

Arguments

data

The data set to transform

discrete_columns

Column ids for columns with discrete/nominal values to be one hot encoded.

Examples

data <- sample_toydata()
transformer <- data_transformer$new()
transformer$fit(data)

---

Method transform_continuous()

Usage

data_transformer$transform_continuous(column_meta, data)

---

Method transform_discrete()

Usage

data_transformer$transform_discrete(column_meta, data)

---

Method transform()

Transform data using a fitted data_transformer. (From original format to transformed format.)

Usage

data_transformer$transform(data)

Arguments

data

The data set to transform

Examples

data <- sample_toydata()
transformer <- data_transformer$new()
transformer$fit(data)
transformed_data <- transformer$transform(data)

---

Method inverse_transform_continuous()

Usage

data_transformer$inverse_transform_continuous(meta, data)

---

Method inverse_transform_discrete()

Usage

data_transformer$inverse_transform_discrete(meta, data)

---

Method inverse_transform()

Inverse Transform data using a fitted data_transformer. (From transformed format to original format.)

Usage

data_transformer$inverse_transform(data)

Arguments

data

The data set to transform

Examples

data <- sample_toydata()
transformer <- data_transformer$new()
transformer$fit(data)
transformed_data <- transformer$transform(data)
reconstructed_data <- transformer$inverse_transform(transformed_data)

---

Method clone()

The objects of this class are cloneable with this method.

Usage

data_transformer$clone(deep = FALSE)

Arguments

deep

Whether to make a deep clone.

Examples

## Not run: 
# Before running the first time the torch backend needs to be installed
torch::install_torch()
# Load data
data <- sample_toydata()
# Build new transformer
transformer <- data_transformer$new()
# Fit transformer to data
transformer$fit(data)
# Transform data and store as new object
transformed_data <-  transformer$transform(data)
# Train the default GAN
trained_gan <- gan_trainer(transformed_data)
# Sample synthetic data from the trained GAN
synthetic_data <- sample_synthetic_data(trained_gan, transformer)
# Plot the results
GAN_update_plot(data = data,
synth_data = synthetic_data,
main = "Real and Synthetic Data after Training")

## End(Not run)

## ------------------------------------------------
## Method `data_transformer$fit`
## ------------------------------------------------

data <- sample_toydata()
transformer <- data_transformer$new()
transformer$fit(data)

## ------------------------------------------------
## Method `data_transformer$transform`
## ------------------------------------------------

data <- sample_toydata()
transformer <- data_transformer$new()
transformer$fit(data)
transformed_data <- transformer$transform(data)

## ------------------------------------------------
## Method `data_transformer$inverse_transform`
## ------------------------------------------------

data <- sample_toydata()
transformer <- data_transformer$new()
transformer$fit(data)
transformed_data <- transformer$transform(data)
reconstructed_data <- transformer$inverse_transform(transformed_data)

RGAN documentation built on March 30, 2022, 1:07 a.m.