This function lets the user cluster a whole data.frame automatically. As you might know, the goal of kmeans is to group data points into distinct non-overlapping subgroups. If needed, one hot encoding will be applied to categorical values automatically with this function. For consideration: Scale/standardize the data when applying kmeans. Also, kmeans assumes spherical shapes of clusters and does not work well when clusters are in different shapes such as elliptical clusters.
clusterKmeans(
df,
k = NULL,
wss_var = 0,
limit = 15,
drop_na = TRUE,
ignore = NULL,
ohse = TRUE,
norm = TRUE,
algorithm = c("Hartigan-Wong", "Lloyd", "Forgy", "MacQueen"),
dim_red = "PCA",
comb = c(1, 2),
seed = 123,
quiet = FALSE,
...
)List. If no k is provided, contains nclusters and
nclusters_plot to determine optimal k given their WSS (Within
Groups Sum of Squares). If k is provided, additionally we get:
df data.frame with original df plus cluster column
clusters integer which is the same as k
fit kmeans object used to fit clusters
means data.frame with means and counts for each cluster
correlations plot with correlations grouped by clusters
PCA list with PCA results (when dim_red="PCA")
tSNE list with t-SNE results (when dim_red="tSNE")
Dataframe
Integer. Number of clusters
Numeric. Used to pick automatic k value,
when k is NULL based on WSS variance while considering
limit clusters. Values between (0, 1). Default value could be
0.05 to consider convergence.
Integer. How many clusters should be considered?
Boolean. Should NA rows be removed?
Character vector. Names of columns to ignore.
Boolean. Do you wish to automatically run one hot encoding to non-numerical columns?
Boolean. Should the data be normalized?
character: may be abbreviated. Note that
"Lloyd" and "Forgy" are alternative names for one
algorithm.
Character. Select dimensionality reduction technique.
Pass any of: c("PCA", "tSNE", "all", "none").
Vector. Which columns do you wish to plot? Select which two variables by name or column position.
Numeric. Seed for reproducibility
Boolean. Keep quiet? If not, print messages.
Additional parameters to pass sub-functions.
Other Clusters:
clusterOptimalK(),
clusterVisualK(),
reduce_pca(),
reduce_tsne()
Sys.unsetenv("LARES_FONT") # Temporal
data("iris")
df <- subset(iris, select = c(-Species))
# If dataset has +5 columns, feel free to reduce dimenstionalities
# with reduce_pca() or reduce_tsne() first
# Find optimal k
check_k <- clusterKmeans(df, limit = 10)
check_k$nclusters_plot
# Or pick k automatically based on WSS variance
check_k <- clusterKmeans(df, wss_var = 0.05, limit = 10)
# You can also use our other functions:
# clusterOptimalK(df) and clusterVisualK(df)
# Run with selected k
clusters <- clusterKmeans(df, k = 3)
names(clusters)
# Cross-Correlations for each cluster
plot(clusters$correlations)
# PCA Results (when dim_red = "PCA")
plot(clusters$PCA$plot_explained)
plot(clusters$PCA$plot)
Run the code above in your browser using DataLab