Ordering clusters in plotClusterTrajectories()

[hichew22]

Hello Niek,

I am using the dtw method in latrend to plot some clusters like so: plotClusterTrajectories(dtw_model_2)

or plot(dtw_model_2)

Is there a way within plotClusterTrajectories() where I can specify that the cluster with more values is always listed as cluster "A" (highest frequency to lowest frequency)?

Or would I need to reorder them manually, something like:

# Create dataframe with cluster assignments and UPNs
cluster_dtw <- trajectoryAssignments(dtw_model_2)
upn <- ids(dtw_model_2)
df_cluster_dtw <- cbind(cluster_dtw, upn) %>%
  as.data.frame()

# Recode clusters from largest (1) to smallest (2)
cluster_freq <- table(df_cluster_dtw$cluster_dtw)
ordered_clusters <- names(sort(cluster_freq, decreasing = TRUE))
df_cluster_dtw$cluster_dtw <-
  factor(
    df_cluster_dtw$cluster_dtw,
    levels = ordered_clusters,
    labels = seq_along(ordered_clusters)
  ) 

# Create column with cluster labels and percentages
cluster_percentages <- prop.table(table(df_cluster_dtw$cluster_dtw))
cluster_labels = sprintf("%s (%d%%)", names(cluster_percentages), round(cluster_percentages * 100))

df_cluster_dtw <- df_cluster_dtw %>%
  mutate(cluster_label = factor(
    cluster_dtw,
    levels = names(cluster_percentages),
    labels = cluster_labels
  ))

# Add cluster assignments to df_long
df_long <- df_long %>%
  left_join(df_cluster_dtw, by = "id")

# Plot cluster trajectories
latrend::plotClusterTrajectories(
  df_long,
  response = "value",
  cluster = "cluster_label",
  trajectories = TRUE,
  facet = TRUE,
  size = 2
)

?

Thank you!

[niekdt]

Hi @hichew22, thanks for the suggestion. It's a useful feature to have, especially when comparing similar cluster solutions.

I'll start with adding an argument to plotClusterTrajectories for specifying which clusters to plot, and the ordering. Ultimately, having a wrapper lcModel class for which you can specify the ordering logic would be best. But that'll take some more effort.

[niekdt]

plotClusterTrajectories now has a clusterOrder argument allowing you to specify which clusters to plot, and the order thereof. Either by name or index

[hichew22]

Awesome, thanks, Niek! Do you have an example, and does this also allow ordering clusters from most to least frequent?

[niekdt]

You're welcome!

data(latrendData)
method <- lcMethodLMKM(Y ~ Time, id = "Id", time = "Time")
model <- latrend(method, latrendData, nClusters = 3)

# change cluster order
plotClusterTrajectories(model, clusterOrder = c('B', 'C', 'A'))

# show only specific clusters
plotClusterTrajectories(model, clusterOrder = c('B', 'C'))

It's intended as a quick way to set a custom order, but to dynamically order by cluster size, you can use:

plotClusterTrajectories(model, clusterOrder = order(-clusterSizes(model)))

In the future I intend to add some lcModel wrapper classes that would automatically relabel clusters based on some criterion, so the ordening would then be handled during the latrend fitting procedure.

[hichew22]

I tried doing that for a 4-cluster DTW model as such: plotClusterTrajectories(dtw_model_4, clusterOrder = order(-clusterSizes(dtw_model_4)))

However, the clusters do not appear ordered. I did make sure to download the most recent installation of latrend. Could you help me with this?

[niekdt]

Did you install the latest commit (not release)?

remotes::install_github('philips-software/latrend')

[hichew22]

Yes, I just did but seems like the function does not work.

─ preparing ‘latrend’: (555ms) ✔ checking DESCRIPTION meta-information ... ─ installing the package to process help pages (811ms) Loading required namespace: latrend ─ saving partial Rd database (3.1s) ─ checking for LF line-endings in source and make files and shell scripts (335ms) ─ checking for empty or unneeded directories ─ building ‘latrend_1.5.1.tar.gz’

[niekdt]

I can't spot any issues in the source code. Could you let me know what the output is of:

latrend:::make.orderedClusterNames(clusterNames(dtw_model_4), order(-clusterSizes(dtw_model_4)))

[hichew22]

I just tried re-running it and it works now! Perhaps I had to restart my R session. Thank you so much!