Releases

pyDarwin-Certara 2.0

What’s New

  • Added NLME Engine support, including Omega Block Search.

  • Introduced {darwin_cmd} alias for Linux Grid runs.

  • Introduced search_info command.

  • Introduced estimations of number of models and remaining time. The former is shown in the beginning of the search as well as in the search_info output (except for Exhaustive search), the latter – in the beginning of every but first iteration. Those may be not too accurate since they assume a certain amount of Downhill Search iterations (if it is requested) and that a model run time is on average the same (which is not always the case due to duplicates that sometimes make up a large part of the iteration).

  • Added tokens consistency check before running the search.

  • Added NONMEM license expiration error to the visible errors.

  • Updated and renamed columns in results.csv.

  • Introduced new model statuses:

    • Clone – a sibling with the same genotype

    • Twin – a sibling with the same phenotype

    • Restored – a model run picked from the cache, when the cache was loaded form a file and the model was picked for the first time

    • Cache – a model run picked from the cache in all other cases

  • Improved Omega Structure Search, e.g.:

    • added individual omega structure search

    • increased number of possible omega block patterns

    • reduced number of model runs by adding the omega structure to the model phenotype, detecting twins, and looking for duplicates in model cache by phenotype

  • Introduced key models retention: best models from every iteration can be saved with all the necessary output in a separate folder for further analysis.

  • Changed Final Downhill and 2-bit search iteration names.

  • Removed unnecessary 0 generation from Genetic Algorithm.

  • Reorganized examples folder, added NONMEM and NLME subfolders.

  • If random_seed is not set in the options file it will be initialized with a random number.

  • working_dir must be an absolute path.

Issues Fixed

  • Removed commas and new lines from translation and runtime messages so they won’t break CSV structure anymore.

  • Corrected calculation of ‘Number of unique models to best model’.

pyDarwin-Certara 1.1.1

What’s New

  • The condition_number penalty is now added to the fitness value for the case when the covariance step is successful and condition_number > 1000. Previously, the condition_number penalty was added to fitness value only for the case when covariance was unsuccessful or not requested.

Issues Fixed

  • An issue was corrected where the correlation penalty does not get added to the resulting fitness value for cases when it should.

pyDarwin-Certara 1.1

What’s New

Issues Fixed

  • An issue has been fixed that could cause the search to fail if the covariance step ($COV) was not specified in template.txt.

pyDarwin-Certara 1.0

Initial release of pyDarwin-Certara offers the Python module darwin to perform a search over a candidate model space using one of the following machine learning algorithms:

Users can alternatively select the Exhaustive Search (EX) to execute all potential candidate models in a given search space without machine learning optimization.

Two primary functions to execute the search have been made available:

python -m darwin.run_search <template_path> <tokens_path> <options_path>

To execute, call the darwin.run_search function and provide the paths to the following files as arguments:

  1. Template file (e.g., template.txt) - basic shell for NONMEM control files

  2. Tokens file (e.g., tokens.json) - json file describing the dimensions of the search space and the options in each dimension

  3. Options file (e.g., options.json) - json file describing algorithm, run options, and post-run penalty code configurations.

Alternatively, you may execute the darwin.run_search_in_folder function, specifying the path to the folder containing the template.txt, tokens.json, and options.json files as a single argument:

python -m darwin.run_search_in_folder <folder_path>