---
title: "Clustering with K-Means and UMAP"
author: "Steven P. Sanderson II, MPH"
date: "`r Sys.Date()`"
output: 
  rmarkdown::html_vignette:
    toc: true
    toc_depth: 2
vignette: >
  %\VignetteIndexEntry{Clustering with K-Means and UMAP}
  %\VignetteEncoding{UTF-8}
  %\VignetteEngine{knitr::rmarkdown}
editor_options: 
  markdown: 
    wrap: 72
---

```{r, echo = FALSE, message = FALSE, warning = FALSE}
knitr::opts_chunk$set(
    message = FALSE,
    warning = FALSE,
    fig.width = 8, 
    fig.height = 4.5,
    fig.align = 'center',
    out.width='95%', 
    dpi = 100,
    collapse = TRUE,
    comment = "#>"
)
```

> healthyR.ai: A toolkit for hospital data
# Libaray Load

First things first, lets load in the library:

```{r setup}
library(healthyR.ai)
```

# Information

K-Means is a partition algorithm initially designed for signal processing.
The goal is to partition *`n`* observations into *`k`* clusters where
each *`n`* is in *`k`*. The unsupervised k-means algorithm has a loose
relationship to the k-nearest neighbor classifier, a popular supervised
machine learning technique for classification that is often confused
with k-means due to the name. Applying the 1-nearest neighbor classifier
to the cluster centers obtained by k-means classifies new data into the
existing clusters.

The aim of this vignette is to showcase the use of the `healthyR`
wrapper for the `kmeans` function the the wrapper and plot for the
`uwot::umap` projection function. We will go through the entire workflow
from getting the data to getting the fina `UMAP` plot.

# Generate some data

```{r get_data}
library(healthyR.data)
library(dplyr)
library(broom)
library(ggplot2)

data_tbl <- healthyR_data %>%
    filter(ip_op_flag == "I") %>%
    filter(payer_grouping != "Medicare B") %>%
    filter(payer_grouping != "?") %>%
    select(service_line, payer_grouping) %>%
    mutate(record = 1) %>%
    as_tibble()

data_tbl %>%
  glimpse()
```

Now that we have our data we need to generate what is called a user item
table. To do this we use the function *`hai_kmeans_user_item_tbl`* which
takes in just a few arguments. The purpose of the user item table is to
aggregate and normalize the data between the users and the items.

The data that we have generated is going to look for clustering amongst
the *`service_lines`* (the user) and the *`payer_grouping`* (item)
columns.

Lets now create the user item table.

# User Item Tibble

```{r uit}
uit_tbl <- hai_kmeans_user_item_tbl(data_tbl, service_line, payer_grouping, record)
uit_tbl
```

The table is aggregated by item for the various users to which the
algorithm will be applied.

Now that we have this data we need to find what will be out optimal k
(clusters). To do this we need to generate a table of data that will
have a column of k and for that k apply the k-means function to the data
with that k and return the **`total within sum of squares`**.

To do this there is a convienent function called *`hai_kmeans_mapped_tbl`*
that takes as its sole argument the output from the
*`hai_kmeans_user_item_tbl`*. There is an argument *`.centers`* where the
default is set to 15.

# K-Means Mapped Tibble
```{r kmm_tbl}
kmm_tbl <- hai_kmeans_mapped_tbl(uit_tbl)
kmm_tbl
```

As we see there are three columns, `centers`, `k_means` and `glance`.
The k_means column is the `k_means` list object and `glance` is the
tibble returned by the `broom::glance` function.

```{r kmm_tbl_glance}
kmm_tbl %>%
  tidyr::unnest(glance)
```

As stated we use the `tot.withinss` to decide what will become our *`k`*, an easy
way to do this is to visualize the Scree Plot, also known as the elbow plot. This 
is done by ploting the `x-axis` as the `centers` and the `y-axis` as the `tot.withinss`.

# Scree Plot and Data

```{r scree_plt}
hai_kmeans_scree_plt(.data = kmm_tbl)
```

If we want to see the scree plot data that creates the plot then we can use another function `hai_kmeans_scree_data_tbl`.

```{r scree_data}
hai_kmeans_scree_data_tbl(kmm_tbl)
```

With the above pieces of information we can decide upon a value for *`k`*, in this
instance we are going to use 3. Now that we have that we can go ahead with creating the umap list object where we can take a look at a great many things associated with the data.

# UMAP List Object

Now lets go ahead and create our UMAP list object.

```{r umap_list, message=FALSE, warning=FALSE}
ump_lst <- hai_umap_list(.data = uit_tbl, kmm_tbl, 3)
```

Now that it is created, lets take a look at each item in the list. The `umap_list` function returns a list of 5 items.

  * umap_obj
  * umap_results_tbl
  * kmeans_obj
  * kmeans_cluster_tbl
  * umap_kmeans_cluster_results_tbl
  
Since we have the list object we can now inspect the `kmeans_obj`, first thing we will do is use the `hai_kmeans_tidy_tbl` function to inspect things.

```{r kmeans_obj_inspect}
km_obj <- ump_lst$kmeans_obj
hai_kmeans_tidy_tbl(.kmeans_obj = km_obj, .data = uit_tbl, .tidy_type = "glance")
hai_kmeans_tidy_tbl(km_obj, uit_tbl, "augment")
hai_kmeans_tidy_tbl(km_obj, uit_tbl, "tidy")
```

# UMAP Plot

Now that we have all of the above data we can visualize our clusters that are colored by their cluster number.

```{r umap_plt}
hai_umap_plot(.data = ump_lst, .point_size = 3, TRUE)
```