6. GAN_SAT.Rmd

---
title: "GAN_SAT"
author: "Steffen Moritz"
date: "1/27/2022"
output: html_document
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```


```{r dataset, include=FALSE}
# Load Source Dataset
load(here::here("satgpa.rda"))
sat <- satgpa
```

```{r reticulate, include=FALSE}
# Use reticulate, set python up - so ctgan can be later used
library("reticulate")
#install_python()
#py_available(initialize = TRUE)
use_python("/usr/local/bin/python3")
py_version()
#virtualenv_create("synthchallenge")
#virtualenv_install("synthchallenge", "sdnist")
#virtualenv_install("synthchallenge", "CTGAN")
use_virtualenv("synthchallenge")
#py_module_available("ctgan")
```

```{python pycode}
# import ctgan
from ctgan import CTGANSynthesizer

# define which columns shall be discrete
discrete_columns = [
  'sex',
  'hs_gpa'
]

# Define Algorithm
ctgan = CTGANSynthesizer(epochs=10)

# Fit - r.sat is the syntax to use variables from r scope 
ctgan.fit(r.sat, discrete_columns)

# Create samples from model
res = ctgan.sample(1000)

#print(samples)
# save model (could be used to create different samples later)
ctgan.save('gan_model.pkl')
```

```{r}
# Save result file p$res is the syntax used to access variables in py from r
result_gan <- py$res
save(result_gan, file = "results/sm_gan_ctgan_epoch1000.rda")
```