Request `as.dfm.DocumentTermMatrix` & `as.dfm.TermDocumentMatrix`

[trinker]

It is nice to be able to convert between data structures that are common in the text world.
As the dfm becomes one of the preferred structures it'd be nice to convert from another structure that is common the DocumentTermMatrix & TermDocumentMatrix . Granted, I'm sure you'd prefer people just go from Corpus to dfm in quanteda but sometimes you have one of thes tm structures already because of preference or some pakage output or just to give the user flexibility.

quanteda already provides the as.dfm conversion for matrices and data.frames. I'd like to request extensions of as.dfm.DocumentTermMatrix & as.dfm.TermDocumentMatrix that allow for easy conversion from these data forms. Below I provide the current code I use for such conversion though I realize you're using more sophisticated methods with as_dfm_constructor under the hood. I also get that this flexibility may be in antithesis of the quanteda in which case I'll accept "This is out of scope" as a response. But I figure I might ask and see.

Also, I may have overlooked the fact that quanteda already does this. If this is the case kindly tell me I'm a bone head and the next time we meet up your first beer is on me :-)

Crude code for converting to dfm

as.dfm.DocumentTermMatrix <- function(x, ...){

    as.dfm(
        Matrix::sparseMatrix(
            i = x$i, 
            j = dtm$j, 
            x = x$v, 
            dimnames = list(seq_len(nrow(x)), colnames(x))
        )
    )

}

[kbenoit]

There's already an app for that: convert(x, format = "tm"). We don't do TermDocumentMatrix formats since these are an affront to nature. 😉

[trinker]

convert will make the DocumentTermMatrix I'm asking about converting from a DocumentTermMatrix to a dfm . I don't think convert will make a dfm as I believe it specifies that x must be a dfm.

This is what I want to try to accomplish...let's say I have been given the following DocumentTermMatrix called x by a street text analysis dealer:. How could I make it into a dfm? The as.dfm.DocumentTermMatrix(x) above will do this? You seem (https://github.com/quanteda/quanteda/blob/6e3469c98faca4bb03192854a36dd319f8034546/R/dfm-methods.R) to take pitty on people with data.frame, matrix, Matrix, and sparseMatrix data formats of counts of documents and terms... providing these souls with the as.dfm generic method.

library(tm); library(quanteda)
data("crude")

x <- DocumentTermMatrix(crude,
    control = list(
        weighting = function(x) weightTfIdf(x, normalize = FALSE),
        stopwords = TRUE
    )
)

convert(x, format = "tm")

Result of the as.dfm generic method on a DocumentTermMatrix:

as.dfm(x)
Error in as.dfm.default(x) : 
  as.dfm() only works on data.frame, dfm, dfmSparse, matrix, Matrix, textmodel_lsa objects.

Result of as.dfm after including the poorly written method I demoed above:

as.dfm.DocumentTermMatrix <- function(x, ...){

    as.dfm(
        Matrix::sparseMatrix(
            i = x$i, 
            j = dtm$j, 
            x = x$v, 
            dimnames = list(seq_len(nrow(x)), colnames(x))
        )
    )

}

as.dfm(x)

## Document-feature matrix of: 20 documents, 1,200 features (92.1% sparse).

I feel I didn't do a good job of explaining my request to convert to dfm not the other way around. Mostly I don't want to owe you a beer but if there is already a way to convert a documentTermMatrix to dfm I'm not that cheap and will buy promised beer. If there is not a way to make a dfm from a DocumentTermMatrix and you don't want to provide the as.dfm because you are trying to enforce an adherence to a specific standard I get that too.

[kbenoit]

OK now I understand! Happy to add as.dfm.DocumentTermMatrix() and as.dfm.TermDocumentMatrix() methods. Will do so asap, or feel free to start a PR and I can tidy it up.

You can still owe me a 🍺.

[trinker]

A 🍺 for a feature seems cheap. Done. Thanks :-)