search

mroberts / stmBrowser

Visualizations for the stm package
Other
59 stars 7 forks source link

default labeltype='prob' fails when stm object only contains content covariate, but not prevalence covariate #3

Open cbjrobertson opened 5 years ago

cbjrobertson commented 5 years ago

To reproduce:

library(stm)
data(poliblog5k)
#Create date
dec312007 <- as.numeric(as.Date("2007-12-31"))
poliblog5k.meta$date <- as.Date(dec312007+poliblog5k.meta$day,
                                origin="1970-01-01")
out2 <- prepDocuments(poliblog5k.docs, poliblog5k.voc, poliblog5k.meta)
stm.out <- stm(out2$documents, out2$vocab, K=10,
               content=~ date,
               data=out2$meta,
               max.em.its=1)

stmBrowser(stm.out, data=out2$meta, c("date"),
           text="text")
>>> Error in 1:nrow(topics) : argument of length 0

This is because the topics object doesn't contain an attribute called probs when the model is called with only a content covariate, so topics <- labelTopics(stm.out, n=3)[[labeltype]] in line 103 of the function produces a NULL object and line 105 then errors out.

The error can be avoided by including labeltype = 'topic' in the function call to stmBrowser, so perhaps it's an issue for the docs... or a tryCatch() could be added. Sorry, I wouldn't have written a PR but I don't know R that well. Thanks for the cool function!