如何在R中基于pam聚类结果拆分时间序列数据框

Got it, let's break down how to pull out just the first cluster's time series from your dataframe. I'll cover the most common scenarios you might be working with:

Step 1: Confirm Your Clustering Output #

First, make sure you have your PAM clustering result saved as an object (let's call it pam_result for this example). The cluster labels for each of your 157 time series are stored in pam_result$clustering—this is a vector where each value corresponds to the cluster number for the matching row (or column) in your dataframe.

Step 2: Extract Cluster 1 Data #

Scenario 1: Time series are stored as rows (each row = 1 day) #

If your original dataframe (ts_df) has each time series as a row, use logical indexing to filter rows where the cluster label equals 1:

# Load the cluster package if you haven't already
library(cluster)

# Assuming you've already run PAM (replace k with your actual number of clusters)
pam_result <- pam(ts_df, k = 3)

# Extract only rows belonging to Cluster 1
cluster1_df <- ts_df[pam_result$clustering == 1, ]

Scenario 2: You've already added cluster labels to your dataframe #

If you merged the cluster labels into a column (e.g., named cluster) in ts_df, the code is even simpler:

# Filter rows where the cluster column equals 1
cluster1_df <- ts_df[ts_df$cluster == 1, ]

Scenario 3: Time series are stored as columns (each column = 1 day) #

If your dataframe has time series as columns instead, adjust the indexing to target columns instead of rows:

# Extract only columns belonging to Cluster 1
cluster1_df <- ts_df[, pam_result$clustering == 1]

Verify Your Result #

Quickly check that you got the right subset by counting the number of time series in Cluster 1:

# For row-based storage
nrow(cluster1_df)

# For column-based storage
ncol(cluster1_df)

That should give you a clean dataframe with only the time series from Cluster 1!

内容的提问来源于stack exchange，提问作者Maider