Edit mesures_clustering.py

clustering/mesures_clustering.py

 from sklearn.metrics import silhouette_score
+
 def compute_silhouette_scores(reduced_embeddings, labels):
     """
     Calcule les scores de silhouette pour différents nombres de clusters.
@@ -8,4 +9,34 @@ def compute_silhouette_scores(reduced_embeddings, labels):
     :return: silhouette score
     """
     silhouette_avg = silhouette_score(reduced_embeddings, labels)
-    return silhouette_avg
\ No newline at end of file
+    return silhouette_avg
+
+
+def compute_dunn_index(reduced_embeddings, labels):
+    """
+    Calcule l'index de Dunn pour évaluer la qualité du clustering.
+    
+    :param reduced_embeddings: Matrice des embeddings réduits
+    :param labels: Labels prédits par les algorithmes de clustering
+    :return: Index de Dunn
+    """
+    unique_labels = np.unique(labels)
+    # calcul des distances intra-cluster
+    intra_cluster_distances = []
+    for label in unique_labels:
+        cluster_points = reduced_embeddings[labels == label]
+        if len(cluster_points) > 1:
+            distances = pdist(cluster_points)
+            intra_cluster_distances.append(np.max(distances))
+    #calcul des distances inter-cluster 
+    inter_cluster_distances = []
+    for i, label_i in enumerate(unique_labels):
+        for j, label_j in enumerate(unique_labels):
+            if i < j:
+                points_i = reduced_embeddings[labels == label_i]
+                points_j = reduced_embeddings[labels == label_j]
+                distances = pdist(np.vstack([points_i, points_j]))
+                inter_cluster_distances.append(np.min(distances))   
+    #calcul de l'index de Dunn
+    dunn_index = np.min(inter_cluster_distances) / np.max(intra_cluster_distances)
+    return dunn_index
\ No newline at end of file