News

3.3.0

New features

  • bbi nonmem run now cleans up empty WK_* files. (#306)

3.2.3

Bug fixes

  • bbi nonmem summary omitted the problem_text key from its JSON output when a model's $PROBLEM was empty, breaking the expectation of bbr. (#299)

  • bbi nonmem summary did not permit leading whitespace before a $SIM directive despite that being valid NONMEM syntax. (#301)

  • bbi nonmmem run and bbi nonmem summary did not ignore commented $DATA lines. (#301)

  • bbi init now warns the user if there is a collision when generating identifiers under the "nonmem" entry of bbi.yaml. (#304)

  • bbi init now support NONMEM installation directories that include a period in their name. (#304)

3.2.2

Bug fixes

  • The bbi nonmem summary check for incomplete .lst files (added in v3.2.0) has been reworked to use an approach that is less susceptible to misfires. (#294)

  • bbi nonmem run local now invokes its shell script by calling bash directly rather than relying on the shebang for compatibility with Windows and with systems that have bash installed to a different spot. (#296)

  • The logic used by bbi init to detect the NONMEM directory was broken on Windows. (#296)

  • The bash scripts generated by bbi nonmem run now quote tokens when needed. (#297)

3.2.1

Bug fixes

  • In v3.2.0, bbi nonmem summary gained a check for incomplete .lst files. This check was inappropriately triggered by models that specify an estimation record with METHOD=CHAIN. (#289)

3.2.0

New features and changes

  • -maxlim=2 is now passed to nmfe by default. To disable passing -maxlim to nmfe (the old default), set the value to 0 via the --maxlim option of bbi nonmem or with a bbi.yaml entry:

    nmfe_options:
  maxlim: 0

    Note that the maxlim value set in an existing bbi.yaml will be respected, and that, for compatibility reasons, 100 is treated the same as 0. (#279)

  • To make bbi.yaml more portable across users, bbi init no longer writes the resolved path of the bbi binary in bbi.yaml. Instead it's resolved when generating the SGE submission script. Note that any bbi_binary values in existing bbi.yaml files will still be honored. (#278)

  • As of v7.5, NONMEM generates an FDATA.csv file. bbi now gives this the same clean-up treatment as FDATA. (#282)

  • The "Errors" field in the bbi nonmem summary --json output is now an array of integers that identify which, if any, items in the "Results" array were unable to be summarized. Each item in the "Results" array now includes a "success" field, and, for failures, an "error_msg" field. (#272)

  • bbi nonmem summary --json no longer omits elapsed time values (estimation_time, covariance_time, and cpu_time) when they are zero. (#270, #274)

Bug fixes

  • bbi nonmem summary (#272, #281)

    • gained more safeguards to catch incomplete .lst files.
    • tries harder to channel underlying failures into error messages that are propagated to JSON consumers.
    • reliably exits with a non-zero status on failure.

  • The .lst parser didn't consider that "$PROB" could also be spelled as "$PROBLEM", leading to "LEM" sneaking into rendered problem descriptions. (#283)

  • The .lst parser failed when it encountered NaN objective function values. (#268)

  • Commands that support a --json flag did not consistently relay failures to encode the JSON output. (#268)

  • Several issues around parsing elapsed time metadata (estimation_time, covariance_time, and postprocess_time) have been fixed. (#270)

  • bbi nonmem was supposed to accept --prdefault and --tprdefault options and relay those to nmfe as -prdefault and -tprdefault, but the options on bbi's side weren't exposed. (#280)

  • Calling bbi nonmem reclean with no positional arguments triggered an indexing error rather than showing the help message. (#276)

3.1.1

  • Teach bbi nonmem summary how to handle ONLYSIM runs. (#224)

3.1.0

Primarily, this release contains a refactored test suite which fixes a number of broken unit tests and incorporates integration tests formerly in the (now archived) bbitest repo into the core bbi repo.

Additionally, beginning with this release, only 64-bit binaries will be built by goreleaser.

New features

  • bbi nonmem params has been added. This takes a --dir argument and will output a table with final parameter estimates for all models in that directory. The output is in .csv format, and can be parsed to a tibble by bbr::param_estimates_batch().

3.0.3

  • Fixed a bug where condition_number would come back as 1.0 in situations where it shouldn't. (#214)

  • Added eigenvalue_issues heuristic, triggered when any of the of the eigenvalues <= 0, or if NONMEM needs to force them to be positive. (#216)

3.0.2

The only change in this release is that the bbi nonmem summary output 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.

3.0.1

Primarily a patch to fix a bug when parsing output from NONMEM 7.5 runs (#205). Also ran gofmt and cleaned up some old files and documentation that had gotten stale and was no longer in use or accurate (#206).

3.0.0

Rename to bbi

There are no functional changes in this release, but it represents the official rename from babylon to bbi. Previously bbi was used as an alias for the CLI to babylon and by default the built binary was saved as bbi. Going forward it is the name of the entire app. The accompanying R package will now be known as bbr.

2.3.1

Bug fixes

  • Increased buffer size for reading lines from the .ext file to 256k. (#191)

  • Changed checks for Bayesian and Non-gradient estimation methods to partial string matching so that more estimation methods are correctly matched. (#191)

  • Changed lower-diagonal lookup to a dynamic function instead of a lookup table. This effectively means there is no longer a limit on the number of ETA's that can be parsed from a model. (#189)

  • Fixed bug where model files with a period in the name (i.e. 1000.1.ctl) were being parsed incorrectly. (#192)

2.3.0

All user-facing changes for this release only effect the bbi nonmem summary command and its output. There are also some changes to the CI configuration to facilitate rendering updated validation documents with babylontest and goProjectValidator.

Additions and changes

  • bbi nonmem summary will no longer parse the .cov and .cor files

    • summary --json output will no longer have covariance_theta or correlation_theta elements
    • --no-cov-file and --no-cor-file flags are deprecated

  • bbi nonmem covcor command now returns the json output that was formerly contained in the covariance_theta or correlation_theta elements. The --json flag is not needed for this command. It always returns json.

  • The following fields were added or modified in the bbi nonmem summary --json output:

    • ofv -- Previously included the objective function value for only the last estimation method. Now includes all estimation methods.
    • condition_number -- The condition number. This will only be in the .lst file in certain cases (when scientists ask for it in the $COV block PRINT=E). There will be one for each estimation method. If any of them are large, the large_condition_number boolean under run_heuristics will be true.
    • eta_pval_significant -- Added to run_heuristics. true if any shrinkage_details.pval < 0.05. Will be false if no shrinkage file is present.
    • PRDERR -- Added to run_heuristics. Indicates whether a PRDERR file is present in the output directory.

Bug fixes

  • Previously, the .shk file was not added to the files_used section of the bbi nonmem summary --json output, even when it was used. This has been fixed.

2.1.3

Do not use for processing - there is an extraneous log statement that prints out when asking for --json which will print an extra info log making it unsuitable for piping into subsequent tools that expect a clean json output. This is fixed in 2.1.4

2.1.1

Post Execution Hooks and CI Build

This release adds a post-execution hook to Babylon. The --post_work_executable flag is used to tell Babylon the binary to any script or binary you wish to run after execution. Before executing it, Babylon will set a series of environment variables:

  • BABYLON_MODEL_PATH: Fully qualified path to the directory in which the original model was located
  • BABYLON_MODEL: Full name (file + ext) of model. file.mod
  • BABYLON_MODEL_FILENAME: First part of the filename. IE if the model is file.mod, this would be file
  • BABYLON_MODEL_EXT: Extension of the filename. IE if model is file.mod, this will be mod
  • BABYLON_OUTPUT_DIR: The directory into which the model was placed and ran
  • BABYLON_SUCCESSFUL: Whether or not an error occurred. TRUE if completed without error. Error if errors were encountered.
  • BABYLON_ERROR: If an error occurred, any error generated (Default is an empty string)

During the execution process, this is all rendered down into a post_processing.sh script in the OutputDir. This allows for easy troubleshooting.

What if the script you're executing actually needs more environmentally than the above? The --additional_post_work_envs flag handles that. Here you can specify in comma-separated ENV style additional values to set into the environment:

--additional_post_work_envs "THIS=THAT,THESE=THOSE,BABYLON_ADDITIONAL=1"

The above would set the following additional environmental variables before executing your requested script /binary:

  • THIS=THAT
  • THESE=THOSE
  • BABYLON_ADDITIONAL=1

One thing you may notice after doing this is that only BABYLON_ADDITIONAL shows up in the post_processing.sh script. This is on purpose. We only write environment values that are prefixed with BABYLON to the post_processing.sh script. Using this pattern, you can feel free to add additional environment variables for security (Such as auth tokens for Slack) without worrying about them being written to the file, or add any number of other variables that get added to the file for reproducibility.

0.3.0

This release represents an expansion of the capabilities of the bbq server to handle both ping/version endpoints to introspect information about a running server.

This also represents an expansion in the information tracked for a model. Error handling has been made slightly more robust, with errors bubbling up to store in the database with an ERROR status for failed run.

For execution, the ability to specify a custom directory has been implemented through the --runDir flag via the CLI or through the ProposedRunDir run setting through the server structure.

0.2.0

This version adds a --oneEst flag to the cli when doing bbi run. Given a model has already been run, and an estimation directory is present, the model will be skipped. This can be quite useful when running a directory of models, then subsequently adding/removing some models, while leaving the estimation records for others. When trying to re-run the models, the entire directory can be run, and --oneEst will skip all models that already have been estimated.