如何从diana (diana，cluster R package)输出中获得最优聚类数？

install.packages("cluster")
library(cluster)

set.seed(123) # 设置随机种子以便结果可复现
diana_result <- diana(df)

library(factoextra)

# 计算不同聚类数下的轮廓系数
silhouette_scores <- numeric()
for (k in 2:10) { # 假设你想检查2到10个簇
  diana_k <- cutree(diana_result, k = k)
  score <- silhouette(diana_k, dist(df))
  silhouette_scores[k] <- mean(score[, "sil_width"])
}

# 找到最大轮廓系数的聚类数
optimal_clusters <- which.max(silhouette_scores) + 1 # 加1是因为我们从2开始检查
print(paste("Optimal number of clusters:", optimal_clusters))

db_scores <- numeric()
for (k in 2:10) {
  diana_k <- cutree(diana_result, k = k)
  score <- index.DB(df, diana_k, centrotypes = "medoids")
  db_scores[k] <- score$DB
}

# 找到最小Davies-Bouldin指数的聚类数
optimal_clusters_db <- which.min(db_scores) + 1
print(paste("Optimal number of clusters (Davies-Bouldin):", optimal_clusters_db))

ch_scores <- numeric()
for (k in 2:10) {
  diana_k <- cutree(diana_result, k = k)
  score <- index.G1(df, diana_k, diana_result$medoids, centrotypes = "medoids")
  ch_scores[k] <- score$value
}

# 找到最大Calinski-Harabasz指数的聚类数
optimal_clusters_ch <- which.max(ch_scores) + 1
print(paste("Optimal number of clusters (Calinski-Harabasz):", optimal_clusters_ch))

fviz_nbclust(df, FUNcluster = diana, method = "silhouette") +
  theme_minimal()