dfr_browser: Create and launch a model browser

In agoldst/dfrtopics: Tools for exploring topic models of text

Create and launch a model browser

Description

Export model data and all supporting files needed to browse a model or models interactively using dfr-browser, then (optionally) open a web browser. It is also possible to browse a list of models.

Usage

dfr_browser(m, ...)

Arguments

m	mallet_model object from train_model or load_mallet_model, or a list of such models, or the results of align_topics; see Details.
...	passed on to export_browser_data, q.v., especially the parameters overwrite, n_scaled_words, info, proper, and permute
out_dir	directory for output. By default, files are saved to a temporary directory.
browse	if TRUE, launch web browser after export for viewing (requires servr package).
internalize	always set to FALSE. If TRUE, model data is in the browser home page rather than separate files, but this behavior is deprecated. See Details.
overwrite	if TRUE (default), overwrite any pre-existing files.
condition	dfr-browser displays topic proportions conditioned on (bins of) a chosen metadata variable (by default, publication date). Any variable in metadata(m) may be specified. A vector or list of such variables may also be supplied (see Details). To adjust the binning of continuous or time covariates, change the dfr-browser configuration via info or sub_info.
info	dfr-browser configuration object (see export_browser_data).
sub_info	list of lists of browser configurations (multiple-model display only). The elements of this list should have names corresponding to ids, and each of the elements of its elements should have names corresponding to condition. If NULL, default values are used.
metadata_header	if TRUE (FALSE is default), the exported metadata CSV will have a header row (not expected by dfr-browser by default, but see the example below)
ids	character vector. If multiple models are specified in m, the corresponding element of ids is used as a model ID in dfr-browser.
insert_alignment_table	if m is a topic_alignment object and this is TRUE, the "About" page of the generated browser will include a table of the topics in each model with aligned topics in the same row.

Details

If browse=T, the function attempts to start a local webserver and open a web browser pointing to the appropriate URL. This functionality requires the servr package.

For more control over the export, including the option to export data files only, if for example you have modified the HTML/CSS/JS of an existing dfr-browser, use export_browser_data (q.v.).

Browsing multiple models at once

dfr-browser can be configured to retrieve data from more than one model, or to show topic proportions in a single model conditioned on more than one metadata variable, or both. dfr_browser can generate the necessary data and configuration files. Passing dfr_browser a list of model objects will generate a single dfr-browser with a menu for swapping among the models. Passing a single mallet_model and a vector of variable names as condition will generate a dfr-browser with a menu for swapping among covariates (internally, dfr-browser treats these two possibilities the same). Passing both a list of models and a vector of variable names will generate all the model-covariate pairs. Passing a list of models and a list of the same length of variable names will use the corresponding element of condition as the chosen covariate options for the given model. Finally, in place of a list of models, you can supply the result of align_topics. In that case the topics in each model are relabeled according to their assigned cluster number. It is possible to achieve the same result more manually by passing a list of models and specifying a topic_ids vector for each model in the sub_info parameter.

Deprecated option

internalize=TRUE will cause all model data to be stored in the index.html file, whence it can be loaded by dfr-browser. This behavior is no longer useful; the model data should always be exported as separate files for asynchronous loading by dfr-browser. (In former days one could avoid having to launch a web server and simply point a browser at the index.html file. dfr-browser no longer works in this circumstance because it relies on a Web Worker, which modern web browsers will refuse to load from a file:/// URL.)

See Also

export_browser_data which does the work of exporting files, model_dfr_documents, train_model, topic_scaled_2d

Examples

## Not run: 
m <- model_dfr_documents("citations.CSV", "wordcounts",
    "stoplist.txt", n_topics=40)
# launch browser
dfr_browser(m)

# generate a second model and align its topics with the first for more
# convenient comparisons
m2 <- model_dfr_documents("citations.CSV", "wordcounts",
    "stoplist.txt", n_topics=40)
cl <- model_distances(list(m, m2), n_words=40) %>% align_topics()
dfr_browser(m2, permute=match(cl$clusters[[1]], cl$clusters[[2]])))
# or simply:
dfr_browser(cl)

# create a browser that allows you to toggle among 2 metadata variables
dfr_browser(m, condition=c("pubdate", "journaltitle"))

# create a browser for metadata in a non-DfR format; note info specification

dfr_browser(m, condition="something_else",
     metadata_header=T,
     info=list(VIS=list(
         metadata=list(type="base"),
         bib=list(type="base"),
         bib_view=list(major="all", minor="raw")
     )),
)

## End(Not run)

agoldst/dfrtopics documentation built on July 15, 2022, 4:13 p.m.