Wrong normalization in rho plotting?

[lorenzo-zuccato]

Hi,
I wanted to ask clarifications about the rho plotting from the function plot.BayesMallows. I was perplexed about the normalization in line 69 of the file plot.BayesMallows.R. In case of multiple clusters I think the results does not sum to 1 for every cluster. I would use something like

df$pct <- ave(df$n, df$cluster, df$item, FUN = function(x){x / sum(x)})

instead of

df$pct <- df$n / sum(df$n)

Hope that this makes sense :)

[wleoncio]

It would be useful to get a reproducible example for the case of multiple clusters. I've probably misunderstood, but at least in the 2 examples in the test-plot.R file the current code adds to 1 and the proposal doesn't (see images below, where a breakpoint was added just before that line).

(...)

Lorenzo, can you confirm you're running the latest build (currently, 1.4.0.9002) of the package?

[lorenzo-zuccato]

The sum you are performing returns the number of clusters fitted by the model. This is because I think there is a different distribution of the consensus for every cluster.

set.seed(1234)
bm <- compute_mallows(mini_test, n_clusters = 3, nmc = 10000)

bm$burnin <- 1000

plot(bm, parameter = "rho", item = 1)

Here you see that you obtain 1 if you sum the values in all graphs. Instead I would expect something where every single graph sums to 1.

[wleoncio]

Thanks for the clarification! This goes over my head, so I'll leave it to @osorensen to analyse the case. :)

[osorensen]

Thanks @lorenzo-zuccato and @wleoncio. I'll take a look.

[osorensen]

Issue fixed now. Thanks for reporting @lorenzo-zuccato. It actually turned out the ggplot2 could do much of the data processing job, so was able to remove a couple of codelines at the same time.

Note that the plot design is different now, since I'm using ggplot2::geom_histogram(), but since a ggplot object is returned, it's possible to define exactly the style you like.

library(BayesMallows)
set.seed(1234)
bm <- compute_mallows(potato_visual, n_clusters = 3, nmc = 1000)

bm$burnin <- 500

plot(bm, parameter = "rho", item = 1)

^{Created on 2023-11-23 with reprex v2.0.2}

[osorensen]

Closed by #343