kmeans.py

import numpy as np
import pandas as pd
from k_means_constrained import KMeansConstrained
from sklearn.preprocessing import OneHotEncoder, StandardScaler
# 2. Vectorize Categories (One-Hot Encoding)
encoder = OneHotEncoder(sparse_output=False)
encoded_features = encoder.fit_transform(df[['Contract', 'Domain', 'Portfolio', 'Department']])
feature_names = encoder.get_feature_names_out()

# 3. Apply Hierarchy Weights
# We multiply the columns to prioritize Project > Portfolio > Discipline
weights = {
    'Contract': 10.0,   # Top priority
    'Domain': 7.0,  # Medium priority # todo - need supragroupings
    'Portfolio': 3.0,  # Medium priority # todo - need supragroupings
    'Department': 1.0  # Lowest priority
}

weighted_features = encoded_features.copy()
for i, col in enumerate(feature_names):
    for category, weight in weights.items():
        if category in col:
            weighted_features[:, i] *= weight

# 4. Run Constrained K-Means
# Setting size_min=15 and size_max=20 to meet your specific grouping needs
n_clusters = len(df) // 17  # Aim for ~17 per group
clf = KMeansConstrained(
    n_clusters=n_clusters,
    size_min=15,
    size_max=20,
    random_state=42
)

df['cluster'] = clf.fit_predict(weighted_features)

# 5. Output Results
print(df.groupby('cluster').size()) # Verify sizes are between 15-20
print(df.sort_values('cluster').head(20))

df.sort_values('cluster').to_csv('clustered.csv', index=False)