suggest_k {ulrb}

Suggest k

Description

Tool to help decide how many clusters to use for partition around medoids algorithm.

Usage

suggest_k(
  data,
  range = 3:10,
  samples_col = "Sample",
  abundance_col = "Abundance",
  index = "Average Silhouette Score",
  detailed = FALSE,
  ...
)

Arguments

data

a data.frame with, at least, the classification, abundance and sample information for each phylogenetic unit.

range

The range of values of k to test, default is from 3 to 10.

samples_col

String with name of column with sample names.

abundance_col

string with name of column with abundance values. Default is "Abundance".

index

Index used to select best k. Can be one of: "Average Silhouette Score", "Davies-Bouldin" or "Calinski-Harabasz".

detailed

If False (default) returns an integer with best overall k. If TRUE, returns a list with full details.

...

Extra arguments.

Details

The best k is selected for each sample, based on the selected index. If different k's are obtained for different samples (probable) then we calculate the mean value of k and return it as an integer. Alternatively, we can return a more detailed result in the form of a list.

Note: this function is used within define_rb(), with default parameters, for the optional automatic selection of k.

Detailed option

If detailed = TRUE, then the output is a list with information to help decide for k. More specifically, the list will include:

• A data.frame summarizing what information each index provides and how to interpret the value.

• A brief summary indicating the number of samples in the dataset and the range of k values used.

• A data.frame with the best k for each sample, based on each index.

Automatic k selection

If detailed = FALSE, this function will provide a single integer with the best k. The default decision is based on the maximum average Silhouette score obtained for the values of k between 3 and 10. To better understand why the average Silhouette score and this range of k's were selected, we refer to Pascoal et al., 2024 (in peer-review) and to vignette("explore-classifications").

Alternatively, this function can also provide the best k, as an integer, based on another index (Davies-Bouldin and Calinski-Harabasz) and can compare the entire of possible k's.

Value

Integer indicating best k from selected index. Optionally, can return a list with details.

See Also

evaluate_k(), evaluate_sample_k(), check_DB(), check_CH(), check_avgSil(), cluster::pam()

Examples

# Get the best k with default parameters
suggest_k(nice_tidy)


# Get detailed results to decide for yourself
suggest_k(nice_tidy, detailed = TRUE, range = 2:7)

# Get best k, based on Davies-Bouldin index
suggest_k(nice_tidy, detailed = FALSE, index = "Davies-Bouldin")