NEWS.md
New model tree feature: bbr now allows you to visualize any run_log() as an interactive model tree diagram via model_tree(). This function allows users to easily visualize and track the modeling process for a given project and display any information available in a run_log(), such as model summary information, configuration options, and more. See the vignette here (#673).
New run_nmtran() function for running NM-TRAN on a model object to validate its control stream for correct coding before submission. The NM-TRAN dataset (FDATA) and other NONMEM artifacts can be further inspected by keeping the run directory around (#705).
nm_data() now has a filter argument, allowing the user to filter data based on IGNORE LIST or ACCEPT LIST options defined in the $DATA record of a NONMEM control stream file (#711).
setup_bootstrap_run() now has a data argument, allowing users to provide their own starting dataset to resample from. This defaults to NULL, which will use the output from nm_data(.boot_run, filter = TRUE) (#707, #711).
setup_bootstrap_run() assumed a NUM column was present in the input dataset (the default .join_col in nm_join()). nm_join() was being used in the back-end to create the starting dataset to resample from; the intention being to only include subjects that entered the original problem. Given the nm_join() use, this also required that the original model had finished executing before bootstrapping. Rather than using nm_join() by default, we now use nm_data(.boot_run, filter = TRUE). This both fixes the aforementioned bug and removes the requirement that the base model (model being bootstrapped) has been executed. If the model has been executed, however, we will check the number of records to ensure the filtering was done correctly (#707, #711).New bootstrap feature: bbr now supports the creation, management, and summary of bootstrap runs. Runs are initialized and set up using new_bootstrap_run()
and setup_bootstrap_run(), respectively, and are tracked via the new bbi_nmboot_model type object. They can be submitted the same as any other model, after which they can be summarized via summarize_bootstrap_run() and “cleaned up” via cleanup_bootstrap_run(). (#671, #687, #701)
New simulation feature: bbr now supports simple simulations. add_simuation() will create and submit a new bbi_nmsim_model object, which is then attached to the existing bbi_nonmem_model object. Additional new functions are also exported, including nm_join_sim(), which serves to join the simulation table to the input data. (#687)
check_nonmem_finished() now returns FALSE if an output directory does not exist. (#693)
New get_model_status() helper for returning messages that indicate which model(s) have finished executing and which are incomplete. (#671, #693)
Add .delay argument to wait_for_nonmem() to account for variability in the amount of time required to wait before checking for the presence of an output directory (e.g., batch submissions). (#693)
Removed parenthesis from THETA names in initial_estimates(). The intent was to match the output from param_estimates() so initial and final estimates could be joined more easily. (#674)
When passed an nmbayes model (defined by the bbr.bayes package), nm_join() now gives a more informative error that points to bbr.bayes::nm_join_bayes(). (#662)
check_nonmem_finished() has been converted to a method so that tailored logic can be implemented for derived model types, such as the nmbayes model type in bbr.bayes. (#663)
The character methods for tail_lst() and tail_output() have been adjusted to work with custom model types where *.lst and OUTPUT files do not reside in the top-level output directory. (#661)
get_data_path() can now pull from the control stream file if the model has not yet been submitted. This allows functions like nm_data() to be used pre-model submission. Support for bbi_log_df methods and a new .check_exists argument (defaults to TRUE) have also been added. (#664)
New initial_estimates() function for extracting and formatting initial parameter estimates from a NONMEM control stream file. (#646)
New tweak_initial_estimates() function for tweaking or ‘jittering’ initial parameter estimates. Only $THETA records are supported for now. (#646)
With the latest data.table release (1.15.0), param_estimates_batch() failed for an edge case due to a change in the header detection heuristics. param_estimates_batch() now tells data.table::fread() to expect a header. (#648)
When passed .recurse = TRUE, run_log(), config_log(), and summary_log() included nested models, which was not by design. Any models under another model’s output directory are considered an implementation detail. (#643, #644, #645)
inherit_param_estimates() calls model_summary() underneath and that failed when the parent model did not have .grd or .shk files. inherit_param_estimates() now has a .bbi_args argument (similar to nm_join()) that defaults to not looking for .shk and .grd files. (#638)inherit_param_estimates() function for setting the initial parameter estimates of a model using the final estimates of a previously executed model. (#623)nmrec for parsing control stream files, rather than regex/string manipulation. (#600, #603, #604, #606, #612)nm_join now tracks the origin of each column and stores it as an attribute of the return value (mostly important for bbr.bayes). (#617)update_model_id would replace the id with the based_on field, which was a relative file path rather than a model id. In other words, this function did not support child models created in sub-directories (or otherwise a different directory than the parent model). (#614)delete_models() previously ran into issues when using R 4.0 and an older version of tibble. (#625)All of the code changes in this release focus on enabling third-party packages to introduce new model types and were motivated by the soon-to-be-released bbr.bayes package. Aside from three new generics, these changes are internal and should not affect any user-facing behavior on the bbr side. (#543)
New open_model_file() function opens the model definition file (the control stream for NONMEM models) in RStudio or any other editor supported by utils::file.edit(). (#570)
The vignettes now include examples of calling run_log() with the .include argument (added in bbr 1.4.0). (#575)
delete_models() messages have been improved. (#577)
use_bbi() now allows a relative path and, on Windows, aborts if the path doesn’t end with “.exe”. (#579)
bbr.bbi_exe_mode now defaults to “local” when not on Linux (where the default remains “sge”). (#580)
Display a warning on Windows users if bbi version is below 3.2.2. (#581)
Various updates for compatibility with tidyr 1.3.0 and dplyr 1.1.0 (#572, #578).
model_summaries(), build_path_from_model(), and get_config_path() unintentionally signaled an error when the output directory of a model didn’t exist. (#573)
use_bbi() now does a better job of deciding when to message the user about further setup that’s needed. (#579)
submit_model() now has an .overwrite argument (similar to how new_model() and copy_model_from() have one). Previously, the user had to pass this through as .bbi_args = list(overwrite = TRUE) in order to overwrite output from a previous model execution. (#547)
Functions run_log(), summary_log(), and config_log() all previously defaulted to searching for models recursively in sub-directories. This has been changed so that now users will need to pass .recurse = TRUE to search recursively. The primary reason for the change is user feedback indicating that, more often than not, sub-directories under a model directory contain something like a bootstrap, which would consist of a large number of models that shouldn’t be included in the log table. Additionally, these sub-directories sometimes contain enough models (5000+) to make the *_log() call take quite some time to complete. (#492)
Previously, nm_file() (and, as a result, functions like nm_join() and nm_tables() that use nm_file()) was upper-casing all columns in the table that was loaded. This was not part of the original specification and ended up causing problems with some users’ downstream code, which expected the column names to remain intact. Going forward nm_file() does not modify any column names. (#564)
Previously, a user could pass a directory path to use_bbi() and have the executable installed inside that directory. That behavior was undocumented and, more importantly, led to problems when the same directory path was set as options("bbr.bbi_exe_path"), because that option needs to contain a path to the actual executable. This was fixed so that now use_bbi() explicitly accepts only a path to the desired location of the executable, erroring if it receives a path to an existing directory instead. (#552)
bbi 1.4.0 (specifically #514) made the change that users no longer have to specify parallel = TRUE to have submit_model() respect the number passed to threads. Unfortunately, it also introduced a bug where passing parallel = FALSE without passing threads causes an error. This has been corrected so that passing parallel = FALSE without passing threads will disable parallelization for that run. (#554)
The new .include argument of config_log(), run_log(), and summary_log() limits the result to the specified run names or tags. (#484, #526)
Models can now be starred. See add_star() and remove_star(). (#487)
New convenience functions get_omega(), get_sigma(), and get_theta() return labeled values, with the OMEGA and SIGMA values expanded to a full matrix. (#515)
param_estimates gained an .alpha argument that’s useful for identifying the ETAs flagged by the eta_pval_significant heuristic. (#497)
nm_join() now aborts if the join column has duplicate values, suggesting to the caller that the table format probably needs to be widened to prevent NONMEM from truncating the values. (#533)
If the caller specifies a threads value above one in .bbi_args but does not specify parallel, parallel = TRUE is added so that the threads value is in effect. (#514)
New functions check_nonmem_finished() and wait_for_nonmem() enable checking and waiting on models submitted to the grid. (#480)
Added test_threads() for benchmarking simulation run times with various threads values. (#473, #489, #519, #542)
Added check_run_times() for checking the estimation times of model runs. (#473, #489, #511)
Added delete_models() for removing all model files associated with specified model tags. (#473)
Updated parallel tips and tricks vignette to reference new test_threads() function and related helpers. (#503)
bbi encodes “unspecified” and NA values as -999999999. bbr now maps all occurrences of this value to NA when creating a model summary object. (#524)
When bbi v3.2.0 or later is available, model_summaries() now uses bbi’s concurrency rather than calling model_summary() on each model, leading to a speed up when many models are passed. (#527)
model_diff() learned to print a message rather than call diffobj::diffFile() when there are no changes to avoid confusingly displaying the entire file. (#522)
use_bbi now creates leading directories if needed. (#499)
new_model now ignores the extension of the supplied path. (#510)
The output of print_bbi_args has been reworked to make it easier to digest. (#537)
nm_join() did not reliably sort the resulting data frame when passed .superset = TRUE. (#508)
nm_file() and friends now detect duplicate column names and make them unique. (#530, #539)
Unlike submit_model(), submit_models() didn’t abort when bbi.yaml was missing. (#496)
In combination with a change in bbi v3.2.0, model_summary() can now handle .lst files that have NaN objective function values. (#506)
$TAB syntax in nm_table_files(). (#466)inst/validation/ for creating validation documents with mrgvalidate. (#469)param_estimates_compare() for comparing the result of param_estimates_batch() to a single model (or, more generally, for comparing a set of parameter estimates). (#457)param_estimates_batch() now transforms the parameter names that come from bbi nonmem params ..., replacing _ with , to match the format that is used elsewhere and expected by downstream tools. (#457)use_bbi() printed the wrong current release when an older version was requested. (#144)
use_bbi() and bbi_version() crashed if the bbi executable path contained spaces. (#409)
model_summary() failed with an unclear error message when a caller accidentally placed more than one .lst file in the model directory. (#449)
bbi_help() had a longstanding regression that prevented it from emitting any output. (#447)
bbi nonmem summary used to fail with an out-of-bounds error when fed ONLYSIM output, but, as of bbi v3.1.1, it sets the “only_sim” field. print.bbi_nonmem_summary() and param_estimates.bbi_nonmem_summary() have been updated to check for the new field. (#443)
This release adds a number of helper functions, primarily for use with NONMEM models.
Added nm_file(), nm_grd(), nm_tab(), nm_par_tab(), and nm_data() for reading in NONMEM files more easily. (#426)
Added nm_join(), nm_tables(), and nm_table_files() for reading in NONMEM tables more easily. Notably, nm_join() can be used to get a single tibble containing your NONMEM input data joined against all of your table outputs. (#429 and #430)
param_estimates_batch() for extracting a tibble of parameter estimates from a batch of NONMEM runs. Especially useful for large batches of runs created by something like a bootstrap. (#386)
cov_cor() and check_cor_threshold() for pulling in covariance and correlation matrices from NONMEM .cov and .cor files. (#414)
Passing the .new_model argument to copy_model_from() is now optional. By default, it now tries to increment to the next available integer in the destination directory (the directory containing the parent model). (#424)
update_model_id() for updating mentions of the parent model in the child model’s control stream. (#417)
Per guidance in bbr 1.0.0 release, replace_tags(), replace_bbi_args(), replace_based_on(), add_decisions(), and replace_decisions() have been removed.
Deprecated check_nonmem_table_output(), check_grd(), and check_ext() and replaced them with nm_file() variants. These functions will warn about this for two more releases and then begin to error for two more releases before being removed altogether. (#426)
Deprecated plot_nonmem_table_df(), plot_grd(), and plot_ext() in an effort to more tightly define the scope of bbr. These functions will warn about this for two more releases and then begin to error for two more releases before being removed altogether. (#426)
collapse_to_string() (#393)Fixed a bug where submitting multiple models in a loop with submit_model(.mode = "local", .wait = FALSE) would cause the models to never finish because the bbi processes would get killed by processx when the R objects were garbage collected. (#390)
bbr now checks for a valid configuration file before calling out to bbi to avoid the situation where .wait = FALSE and the “no config file” error from bbi is swallowed. Note, this check is skipped if .dry_run = TRUE. (#390)
Added options(bbr.bbi_exe_mode) for globally setting .mode argument to submit_model() and submit_models(). (#377)
Added .bbi_args argument to bbi_init() for passing through defaults to be stored in the created bbi.yaml file. (#378)
submit_models() (e.g. for bootstrapping) would hang indefinitely. This had something to do with a bug in processx. It was fixed (in bbr) by routing the stdout and stderr to a temp file and then reading from it when necessary, instead of relying on processx to poll the process. (#374)Added tags_diff() function for comparing the tags between different models. (#337)
Added model_diff() function for comparing the model files between different models. (#342)
Added check_up_to_date() function for checking whether the model file(s) and data file(s) associated with a model have changed since the model was run. (#338)
Added more documentation about the heuristics returned from model_summary.bbi_nonmem_model() (#343)
param_estimates() now correctly errors when a Bayesian method is used but is not the final method. (#344)bbi_model parent class to bbi_nonmem_model and bbi_nonmem_summary objects. Many of the helpers in get-path-from-object.R now dispatch on this class. This had been discussed in the past but was primarily done now in preparation for beginning development for Stan modeling, which will create bbi_stan_model and bbi_stan_summary objects that will also inherit from this parent class. (#332)This release is fairly small in terms of changes, but it increments to a new major release version primarily because of the name change which happened in bbr 0.12.0. The most significant change, from a user perspective, is to the default behavior of where bbr looks for bbi on the system. This change is described in issue #321 and a bit more detail is given below.
The 1.0.0 release also represents a stable feature set of basic NONMEM-related functionality. While there will be more features and development relevant to NONMEM in the future, for the immediate future we are shifting our attention towards building similar functionality to support Stan modeling with bbr.
The minimum compatible version of bbi is increased to 3.0.2. This is because there was a breaking change in bbi 3.0.2 which changed all references to "Patients" in the bbi nonmem summary output to "Patients" (discussed further below). Also because there was a bug where bbi could not parse the summary output from NONMEM 7.5 and this bug was fixed in bbi 3.0.1.
options("bbr.bbi_exe_path"), which tells bbr where to look for a bbi installation, now defaults to "bbi". This means that, by default, bbr will look for a bbi installation in the user’s $PATH. options("bbr.bbi_exe_path") can still be set manually by the user and, in fact, we encourage users to set this to an absolute path in their .Rprofile for their project because this explicitly guarantees the correct bbi installation is being used. (#322)
The use_bbi() installer function first tries to install to whatever path is set in options("bbr.bbi_exe_path"), falling back to the bbi in my $PATH (accessed via Sys.which("bbi")) and then an OS-dependent default, in that order. See ?use_bbi() for more details. (#322)
Added print method for bbi_nonmem_model object. Similar to the bbi_nonmem_summary object, the bbi_nonmem_model object should print nicely in the console, and also look good in .Rmd chunks with the option results = 'asis'. (#307)
Added get_data_path() helper function to extract the absolute path to the input data file from bbi_nonmem_model and bbi_nonmem_summary objects. (#314)
Added build_path_from_model() helper function to extract the absolute path to various output files from bbi_nonmem_model and bbi_nonmem_summary objects. (#314)
The output from model_summary() (and model_summaries() and summary_log()) will now refer to individuals in the data set as "Subjects" instead of "Patients", in accordance with the terminology widely used in scientific and medical literature. (#320)
Per guidance in rbabylon 0.10.0 release, replace_tags(), replace_bbi_args(), replace_based_on(), add_decisions(), and replace_decisions() will now error instead of warn about their impending deprecation. These functions will be removed entirely in two more releases.
We are no longer checking in either the .Rprofile or anything in the renv folder. As a result, the development workflow has changed slightly. This change is reflected in the README. (#307 and #308)
Added an option to suppress the minimum bbi version constraint. This is intended only for developers who want to try out development (unreleased) version of bbi while developing on bbr. (#305)
Our Drone CI system now uses a .drone.yml instead of .drone.jsonette. We have also switched the containers that we use for testing in CI to smaller containers which are more specialized for the purpose. (#309)
This package has been renamed to from rbabylon to bbr and the accompanying command-line tool has been renamed from babylon to bbi (which was already its alias, used throughout the package). Any mentions of babylon and rbabylon throughout the package have been renamed accordingly. Mentions of either in the older parts of this NEWS.md document were left as is for historical purposes.