Bug fix

  • Fixing the semantic coherence function (#220)

Minor changes

  • Updating the paper information
  • Deleting the save() function (thank you @AMindToThink for repoting this error in #214)

Minor changes

  • Adding options to set the hyperparameters eta.
  • Documentation updates to pass CRAN checks.

Major changes

  • Migrating to C++17 to follow the new CRAN check. We edited shuffled_indexes() that internally used std::random_shuffle(). This change does not guarantee backward compatibility across all platforms.
  • Using the package cli instead of the base R message() and warning() functions.
  • A new feature to resume the iteration.
  • Stopped support for the label model in keyATM() (it was an experimental feature).

Minor changes

Bug fix

  • A bug fix in summary.keyATM_docs().

Major changes

Minor changes

  • Pure R text loading to address issues related to UTF-8 encoding in Windows (#189)

Minor changes

Bug fix

Major changes

  • Implementation of Polya-Gamma covariate keyATM.
  • Use future.apply instead of parallel (no backward compatibility if you use the init_parallel option).
  • The keyATM_read() function returns a list of objects (e.g., text and document index).
  • An option to store document names in a quanteda dfm object. The keep_docnames option in the keyATM_read() function (thank you Morgan ‘Les’ DeBusk-Lane for the suggestion!).
  • An option to split a dfm to choose keywords with an unsupervised topic model.

Bug fix

  • Using just first 58 speeches of inaugural corpus in test (thank you Ken Benoit for catching this!).

Major changes

  • Changes related to release of dplyr 1.0.1.

Major changes

  • Use the Highest Density Interval as a default (method = "hdi") in plot.strata_doctopic(), plot_timetrend(), and plot_pi(). The previous version uses the Equal-tailed Interval (method = "eti").
  • Add read_keywords for reading dictionary files (e.g. YAML, LIWC).
  • Add the predict() function for the covariate keyATM (thank you Sanja Hajdinjak for the suggestion!).
  • Detailed options for standardization in the covariate keyATM:
    • The standardize option in model_settings argument of the keyATM() function now takes one of "all", "none", or "non-factor" (default).
    • "all" standardizes all covariates (except the intercept), "none" does not standardize any covariates, and "non-factor" standardizes non-factor covariates.
    • In previous versions, this option takes either TRUE (default, standardizing all covariates) or FALSE.
  • A bug fix in the by_strata_DocTopic() function.
  • The output of the keyATM() includes the index of documents used for fitting (this will be useful if the input includes documents with zero length).
  • Add a progress_bar option in the keyATM_read() function (thank you Jae Yeon Kim for the suggestion!).

Bug fix

  • Fix checking time index input (thank you Jae Yeon Kim for pointing out this issue!).

Major changes

  • Updates for dplyr 1.0.0.
  • Update tests.

Major changes

  • Temporary update test-Initialization.R to deal with some errors.

Major changes

Bug fix

  • weightedLDA() without specifying the number of iterations (Chung-hong Chan independently reported this bug, thank you!).
  • Log-likelihood of dynamic models.
  • Saving figures
  • Topic labels when there is no keyword topic.
  • summary.strata_doctopic(): the last topic is removed when the number of no-keyword topic is 0 (thank you Emma Ebowe for pointing out this issue!).

Major changes

  • The first CRAN version.
  • Organize functions into a package.
  • Add keyATM Label.
  • Replace hashmap with fastmap.
  • Thank you Santiago Olivella for finding several bugs!

Major changes

  • We have a new syntax (this version does not support objects made in older keyATM).
  • Faster read functions.
  • Memory efficiency.

Major changes

  • Add keyATM Dynamic.

Major changes

  • Add keyATM Covariate.

Major changes

  • Faster estimation.

Major changes

  • This is the first stable version.

Major changes

  • This version implements weighted model.

Major changes

  • This is the first release of keyATM.
  • It includes the Base model and the first version of the Covariate model.