Requirements

Install dependencies for this to run:

pip install -r requirements.txt

Commands to run for this

Parse open text objections which are grouped together into bullet points:

python main.py parse --responses "responses.xlsx" --output parsed_points.csv

Run clustering analysus on the coded data:

python main.py cluster --coded "pathtocoded.xlsx" --taxonomy "pathtotaxonomy.xlsx" --output output --k N --pcoa-k N --pcoa-sub-k N 

NOTE: --k N argument is optional

Specifying K for hierarchical

--k N is optional. If you pass it, hierarchical clustering will use that exact N clusters. If you leave it out, script picks K automatically using the silhouette score from KMeans, and hierarchical clustering reuses same K.

Specifying K for PCoA

--pcoa-k N

Same function as --k N, but for PCoA.

Specifying K for sub-clusters within each K for PCoA

--pcoa-sub-k N

Same function as --k N, but for subclusters within the PCoA clsuters.

Output

Cluster command writes these files into the output folder:

• binary_matrix.csv: participants by codes table with 0 or 1.
• pcoa_coordinates.csv: 3D PCoA coordinates for each participant.
• kmeans_assignments.csv: Participant # and KMeans cluster, sorted by cluster.
• hierarchical_assignments.csv: Participant # and hierarchical cluster (complete linkage), sorted by cluster.
• top_codes_clusters.csv: top 3 taxonomy codes per KMeans cluster (Group, Code, Count, Proportion, Group Size).
• top_codes_clusters.csv: top 3 codes per subcluster. Group ids are composite (parent*100 + sub,e.g. 201 = cluster 2 / subcluster 1). Inly written when --pcoa-sub-k is passed.
• pcoa_2d.png, pcoa_3d.png, elbow_silhouette.png, dendrogram.png: plots.
• pcoa_2d_subclusters.png: PCoA 2D scatter colored by subcluster. Only written when --pcoa-sub-k is passed