NEWS.md

In ML4LHS/wizard: Windowed Summarization for Autoregressive Data

wizard 0.0.0.9000

• Added a NEWS.md file to track changes to the package.

wizard 0.0.0.9001 (2020-10-04)

• Changed variable argument to variables inside both wiz_add_predictors() and wiz_add_outcomes(). This was done to reflect the fact that the variables argument can accept a vector of variable names.
• Added check_size_only argument to wiz_add_predictors() and wiz_add_outcomes()
• Renamed wiz_categorical_to_numeric() to wiz_dummy_code() to better reflect what the function does. wiz_dummy_code() can now be used to dummy code specific variables, which is useful if you only want to dummy code subset a specific set of variables or if you want to dummy code a categorical variable that is currently coded with numbers (such as 1 meaning high and 2 meaning low).
• Important bug fix to encoding of time to allow for non-time-stamps
• Added check_size_only argument to wiz_add_predictors() to allow you to anticipate the size of the output object/file without actually running the calcualtions.

wizard 0.0.0.9002 (2020-10-07)

• Added create_folder argument to wiz_frame() to allow for automatic creation of the output_folder if it does not already exist. This is useful for automated creation of directories to override the need for user confirmation prior to directory creation.
• Added several checks when creating a wiz_frame() to ensure that there is not missingness or duplication in the data that would be expected to lead to failure of one of the downstream tasks.
• Added save_wiz_frame argument to wiz_frame() and wiz_dummy_code(). If set to TRUE, this saves the wiz_frame object to the specified output_folder with the file_name wiz_frame.rds. Note that this will overwrite prior versions of the file.
• Added log_file option to wiz_add_predictors() and wiz_add_outcomes() to create and append to a log file, which is auto-titled wiz_log.txt and saved in the output_folder.
• Added wiz_add_baseline_predictors() function to add baseline predictors with an option to specify an offset (e.g., up to 1 hour prior to admission)

wizard 0.0.0.9003 (2020-10-08)

• All windows are now calculated simultaneously for each step, which greatly reduces the number of parallel jobs (and the time and memory taken by the allocating memory step).
• Removed time column from wiz_add_baseline_predictors() so that it is treated as a type of "fixed data", and moved its logic into the wiz_add_predictors() function.

wizard 0.0.0.9004 (2020-10-09)

• Added optional max_length argument to wiz_frame() that limits the maximum time or sequence length for each id.
• Bug fix to incorrect calculation of outcomes (this error was introduced in 0.0.0.9003 due to substantial refactoring of code)
• Fixed LOCF imputation bug: now, imputation only occurs within windows in the lookback period (and not beyond)
• Fixed missingness leading to length stat of NA. Now, every stat's missingness is computed dynamically (e.g., length becomes 0)
• Fixed implicit missingness issue due to some IDs and times ids not having certain variables left after filtering
• Changed defaults so that save_wiz_frame and log_file are TRUE. This makes it more consistent with output_file = TRUE in that saving output and logs to file are the default.

wizard 0.0.0.9005 (2020-10-11)

• Created batch_size option in wiz_frame() to chunk processing into batches. A batch of 1000 means 1000 patients will be processed at once.
• Moved processing code to wizard_internal.R and converted all other functions into wrappers
• Moved implicit missingness and LOCF imputation into the parallel jobs to reduce memory footprint after row-binding parallel jobs
• Bug fix to error introduced in wizard 0.0.0.9004 resulting extra time steps for individuals during "handling implicit missingness" step.
• Note: wiz_combine() does not support batches yet but will in a subsequent version.

wizard 0.0.0.9006 (2020-10-27)

• Added wiz_add_growing_predictors() for building cumulatively growing windows beginning at time zero
• Changed batch_size parameter to chunk_size in wiz_frame()
• Changed names of files to indicate baseline, growing, or rolling to indicate the type of variables contained within it.
• Made some changes to the character_language_model.Rmd vignette to generate a large rolling dataset and test the growing window functionality. This vignette needs to be cleaned up for educational use.

wizard 0.0.0.9007 (2020-11-09)

• Updated wiz_combine() to support chunked files and changed interface so that supplying a vector of files is optional.
• Removed wiz_add_predictors_streaming functions

wizard 0.1.0 (2021-02-22)

• Fixed bug with the temporal_data_of_interest being empty due to highly sparse variables
• Also added check for max_step_times_per_id being empty (though this should not occur)

wizard 0.2.0 (2021-02-22)

• Moved wiz_calc out to a separate function to greatly reduce the memory footprint for parallel processing

wizard 0.2.1 (2021-02-24)

• Perform a final check in wiz_add_predictors_internal to ensure that the output_frame is not empty. If it is, then return a message rather than simply writing an empty output file to disk, which can result in an error when combining this with other files due to mismatches in data types for the temporal_id column.

ML4LHS/wizard documentation built on May 31, 2021, 1:24 a.m.