topicmodels.etm (version 0.1.1)

plot.ETM: Plot functionality for an ETM object

Description

Convenience function allowing to plot

• the evolution of the loss on the training / test set in order to inspect training convergence

• the ETM model in 2D dimensional space using a umap projection. This plot uses function textplot_embedding_2d from the textplot R package and plots the top_n most emitted words of each topic and the topic centers in 2 dimensions

Usage

# S3 method for ETM
plot(
  x,
  type = c("loss", "topics"),
  which,
  top_n = 4,
  title = "ETM topics",
  subtitle = "",
  encircle = FALSE,
  points = FALSE,
  ...
)

Value

In case type is set to 'topics', maps the topic centers and most emitted words for each topic to 2D using summary.ETM and returns a ggplot object by calling textplot_embedding_2d.

For type 'loss', makes a base graphics plot and returns invisibly nothing.

Arguments

x

an object of class ETM

type

character string with the type of plot to generate: either 'loss' or 'topics'

which

an integer vector of topics to plot, used in case type = 'topics'. Defaults to all topics. See the example below.

top_n

passed on to summary.ETM in order to visualise the top_n most relevant words for each topic. Defaults to 4.

title

passed on to textplot_embedding_2d, used in case type = 'topics'

subtitle

passed on to textplot_embedding_2d, used in case type = 'topics'

encircle

passed on to textplot_embedding_2d, used in case type = 'topics'

points

passed on to textplot_embedding_2d, used in case type = 'topics'

...

arguments passed on to summary.ETM

See Also

ETM, summary.ETM, textplot_embedding_2d

Examples

if(require(torch) && torch::torch_is_installed())
{

library(torch)
library(topicmodels.etm)
path  <- system.file(package = "topicmodels.etm", "example", "example_etm.ckpt")
model <- torch_load(path)
plot(model, type = "loss")
# \dontshow{
}
# End of main if statement running only if the torch is properly installed
# }

if(require(torch) && torch::torch_is_installed() && 
             require(textplot) && require(uwot) && require(ggrepel))
{

library(torch)
library(topicmodels.etm)
library(textplot)
library(uwot)
library(ggrepel)
path  <- system.file(package = "topicmodels.etm", "example", "example_etm.ckpt")
model <- torch_load(path)
n_epochs <- NULL
# \dontshow{
  n_epochs <- 5
# }
plt   <- plot(model, type = "topics", top_n = 7, which = c(1, 2, 14, 16, 18, 19),
              metric = "cosine", n_neighbors = 15, n_epochs = n_epochs,
              fast_sgd = FALSE, n_threads = 2, verbose = TRUE,
              title = "ETM Topics example")
plt
# \dontshow{
}
# End of main if statement running only if the torch is properly installed
# }