ntnsndr/protocol-bicorder
1
0
Fork 0
Code Issues 1 Pull Requests Actions Packages Projects Releases Wiki Activity

Additional analysis: examining clustering via LDA

Browse Source
This commit is contained in:
Nathan Schneider
2025-12-19 11:40:45 -07:00
parent d62d9205ac
commit b541f85553
9 changed files with 902 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

147
analysis/visualize_clusters.py Normal file
View File

@@ -0,0 +1,147 @@
#!/usr/bin/env python3
"""
Create visualizations of k-means clusters overlaid on dimensionality reduction plots.
"""
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
from pathlib import Path
# Set up paths
output_dir = Path('analysis_results')
plots_dir = output_dir / 'plots'
data_dir = output_dir / 'data'
# Load cluster assignments
clusters = pd.read_csv(data_dir / 'kmeans_clusters.csv')
clusters['cluster'] = clusters['cluster'] # Already 1-indexed
# Load dimensionality reduction coordinates
pca_coords = pd.read_csv(data_dir / 'pca_coordinates.csv')
tsne_coords = pd.read_csv(data_dir / 'tsne_coordinates.csv')
# Merge cluster assignments with coordinates
pca_data = pca_coords.merge(clusters, on='Descriptor')
tsne_data = tsne_coords.merge(clusters, on='Descriptor')
# Set up color scheme
colors = {1: '#2E86AB', 2: '#A23B72'} # Blue for cluster 1, Purple for cluster 2
cluster_names = {1: 'Cluster 1: Relational/Cultural', 2: 'Cluster 2: Institutional/Bureaucratic'}
# ========== PCA Plot with Clusters ==========
print("Creating PCA plot with cluster colors...")
fig, ax = plt.subplots(figsize=(14, 12))
for cluster_id in [1, 2]:
cluster_data = pca_data[pca_data['cluster'] == cluster_id]
ax.scatter(cluster_data['PC1'], cluster_data['PC2'],
c=colors[cluster_id], label=cluster_names[cluster_id],
alpha=0.6, s=60, edgecolors='white', linewidth=0.5)
# Annotate some representative protocols from each cluster
# Sample a few protocols from each cluster for labeling
for cluster_id in [1, 2]:
cluster_data = pca_data[pca_data['cluster'] == cluster_id]
# Label every 8th protocol to avoid clutter
for i, row in cluster_data.iterrows():
if i % 8 == 0:
ax.annotate(row['Descriptor'],
(row['PC1'], row['PC2']),
fontsize=7, alpha=0.7,
xytext=(5, 5), textcoords='offset points')
ax.set_xlabel('PC1 (22.5% variance)', fontsize=12)
ax.set_ylabel('PC2 (22.7% variance)', fontsize=12)
ax.set_title('K-Means Clusters in PCA Space\nTwo Distinct Protocol Families', fontsize=14, fontweight='bold')
ax.legend(loc='best', fontsize=10, framealpha=0.9)
ax.grid(True, alpha=0.3)
plt.tight_layout()
plt.savefig(plots_dir / 'pca_2d_clustered.png', dpi=300, bbox_inches='tight')
print(f" Saved: {plots_dir / 'pca_2d_clustered.png'}")
plt.close()
# ========== t-SNE Plot with Clusters ==========
print("Creating t-SNE plot with cluster colors...")
fig, ax = plt.subplots(figsize=(14, 12))
for cluster_id in [1, 2]:
cluster_data = tsne_data[tsne_data['cluster'] == cluster_id]
ax.scatter(cluster_data['TSNE1'], cluster_data['TSNE2'],
c=colors[cluster_id], label=cluster_names[cluster_id],
alpha=0.6, s=60, edgecolors='white', linewidth=0.5)
# Annotate some protocols
for cluster_id in [1, 2]:
cluster_data = tsne_data[tsne_data['cluster'] == cluster_id]
for i, row in cluster_data.iterrows():
if i % 8 == 0:
ax.annotate(row['Descriptor'],
(row['TSNE1'], row['TSNE2']),
fontsize=7, alpha=0.7,
xytext=(5, 5), textcoords='offset points')
ax.set_xlabel('t-SNE Dimension 1', fontsize=12)
ax.set_ylabel('t-SNE Dimension 2', fontsize=12)
ax.set_title('K-Means Clusters in t-SNE Space\nTwo Distinct Protocol Families', fontsize=14, fontweight='bold')
ax.legend(loc='best', fontsize=10, framealpha=0.9)
ax.grid(True, alpha=0.3)
plt.tight_layout()
plt.savefig(plots_dir / 'tsne_2d_clustered.png', dpi=300, bbox_inches='tight')
print(f" Saved: {plots_dir / 'tsne_2d_clustered.png'}")
plt.close()
# ========== UMAP Plot with Clusters (if available) ==========
umap_path = data_dir / 'umap_coordinates.csv'
if umap_path.exists():
print("Creating UMAP plot with cluster colors...")
umap_coords = pd.read_csv(umap_path)
umap_data = umap_coords.merge(clusters, on='Descriptor')
fig, ax = plt.subplots(figsize=(14, 12))
for cluster_id in [1, 2]:
cluster_data = umap_data[umap_data['cluster'] == cluster_id]
ax.scatter(cluster_data['UMAP1'], cluster_data['UMAP2'],
c=colors[cluster_id], label=cluster_names[cluster_id],
alpha=0.6, s=60, edgecolors='white', linewidth=0.5)
# Annotate some protocols
for cluster_id in [1, 2]:
cluster_data = umap_data[umap_data['cluster'] == cluster_id]
for i, row in cluster_data.iterrows():
if i % 8 == 0:
ax.annotate(row['Descriptor'],
(row['UMAP1'], row['UMAP2']),
fontsize=7, alpha=0.7,
xytext=(5, 5), textcoords='offset points')
ax.set_xlabel('UMAP Dimension 1', fontsize=12)
ax.set_ylabel('UMAP Dimension 2', fontsize=12)
ax.set_title('K-Means Clusters in UMAP Space\nTwo Distinct Protocol Families', fontsize=14, fontweight='bold')
ax.legend(loc='best', fontsize=10, framealpha=0.9)
ax.grid(True, alpha=0.3)
plt.tight_layout()
plt.savefig(plots_dir / 'umap_2d_clustered.png', dpi=300, bbox_inches='tight')
print(f" Saved: {plots_dir / 'umap_2d_clustered.png'}")
plt.close()
# ========== Summary Statistics ==========
print("\n=== Cluster Summary ===")
print(f"Total protocols: {len(clusters)}")
print(f"\nCluster 1 (Relational/Cultural): {len(clusters[clusters['cluster'] == 1])} protocols")
print(f"Cluster 2 (Institutional/Bureaucratic): {len(clusters[clusters['cluster'] == 2])} protocols")
print("\nSample protocols from each cluster:")
print("\nCluster 1 (Relational/Cultural):")
for protocol in clusters[clusters['cluster'] == 1]['Descriptor'].head(10):
print(f" - {protocol}")
print("\nCluster 2 (Institutional/Bureaucratic):")
for protocol in clusters[clusters['cluster'] == 2]['Descriptor'].head(10):
print(f" - {protocol}")
print("\n=== Visualization Complete! ===")