cp_hierarchical_clustering

by Derfel Terciano

A hierarchical clustering library for Cytological profiling projects written in rust

Purpose

This is a submodule that will be used for HistDiff Processor to generate HistDiff heatmaps.

Implements:

• Linkages:
  • Complete Linkage
  • Single Linkage
  • Average Linkage
• Metrics/Matrices:
  • Distance
  • Pearson Correlation

Outputs:

• Dendrogram as json
• Row and column ordering

main function:

create_hierarchy(
    raw_data: &[Vec<f64>],
    mat_metric: Metric,
    link_method: LinkageMethod,
) -> Result<ClusterHierarchy, Box<dyn Error>>

• This function takes a &[Vec<f64>] and creates a ClusterHierarchy struct that stores the clustering output.

create_hierarchy_from_df(
    df: &DataFrame,
    mat_metric: Metric,
    link_method: LinkageMethod,
    ignore_cols: &Option<Vec<usize>>,
) -> Result<ClusterHierarchy, Box<dyn Error>> {

• This is the same main function except it converts a polars dataframe into a hierarchy

Documentation

Run cargo doc --open

TODO:

• Implement clustering foundation
  • Implement Agglomerative clustering
• Implement similarity/distance calculations
  • Use native vectors instead of ndarray
  • Implement Distance matrix calculation
  • Implement Pearson similarity matrix
• Implement tree outputs
  • Output tree JSON
  • Output simple JSON output
  • Output leaf orderings
  • Output dendrogram nodes themselves