---
title: "Using dbscan from Python"
author: "Michael Hahsler"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Using dbscan from Python}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

```{r setup, include=FALSE}
### run with: 
#DBSCAN_RUN_PYTHON_VIGNETTE=true Rscript -e 'rmarkdown::render("vignettes/python.Rmd")'

run_python <- identical(
  tolower(Sys.getenv("DBSCAN_RUN_PYTHON_VIGNETTE")),
  "true"
)
knitr::opts_chunk$set(collapse = TRUE, comment = "#>")

if (run_python) {
  Sys.setenv(RETICULATE_PYTHON = "managed")
  reticulate::py_require(c("numpy", "pandas", "rpy2"))
}
```

R, the R package **dbscan**, and the Python package **rpy2** need to be
installed. The following example reads the iris data and calls the R
implementation of DBSCAN from Python.

The Python chunks are not run during a normal package build. To execute them
while rendering this vignette, set the environment variable
`DBSCAN_RUN_PYTHON_VIGNETTE=true`. The first run uses **reticulate** to create a
managed Python environment and install **numpy**, **pandas**, and **rpy2**. The
environment is cached for later use, and all Python chunks run in one shared
session.

```{python, eval=run_python}
import pandas as pd
import numpy as np
import rpy2.robjects as ro

# Prepare data.
iris = pd.read_csv(
    "https://archive.ics.uci.edu/ml/machine-learning-databases/iris/iris.data",
    header=None,
    names=["SepalLength", "SepalWidth", "PetalLength", "PetalWidth", "Species"],
)
iris_numeric = iris[["SepalLength", "SepalWidth", "PetalLength", "PetalWidth"]]

# Import the R dbscan package.
from rpy2.robjects import packages
from rpy2.robjects import pandas2ri
dbscan = packages.importr("dbscan")

# Convert the pandas data frame using local conversion rules.
conversion_rules = ro.default_converter + pandas2ri.converter
with conversion_rules.context():
    iris_r = ro.conversion.get_conversion().py2rpy(iris_numeric)

db = dbscan.dbscan(iris_r, eps=0.5, minPts=5)
print(db)
```

Extract the cluster assignment vector as a NumPy array:

```{python, eval=run_python}
labels = np.asarray(db.rx2("cluster"), dtype=int)
labels
```

Cluster label `0` identifies noise; positive integers identify clusters.