This PsN version has been tested on the following systems: Windows 10 64-bit with Strawberry Perl 5.22.1 , gfortran 4.9.2 and NONMEM 7.2.0 and 7.3.0 Windows 7 64-bit with ActivePerl 5.20.1 (32 bit), gfortran 4.8.1 and NONMEM 7.3.0 Windows 7 64-bit with ActivePerl 5.16.3 (32 bit), gfortran 4.6.0 and NONMEM 7.3.0 Windows 7 64-bit with Strawberry Perl 5.18.2, gfortran 4.7.3 and NONMEM 7.2.0 and 7.3.0 Windows XP 32-bit with ActivePerl 5.16.3, gfortran 4.6.0 and NONMEM 7.1.0, 7.1.2 and 7.2.0 MacOS X 10.11.4 with Perl 5.18.2, gfortran 5.2.0 and NONMEM 7.3.0 MacOS X 10.6.6 with ActiveState Perl 5.16.3, gfortran 4.6.2 and NONMEM 7.3.0 MacOS X 10.6.6 with ActiveState Perl 5.16.3, ifort 12.0.0 and NONMEM 7.2.0 and 7.3.0 Ubuntu Linux 15.10 with Perl 5.20.2, gfortran 5.2.1 and NONMEM 7.2.0 and 7.3.0 Fedora 20 with Perl 5.18.2, gfortran 4.8.2 and NONMEM 7.2.0 Scientific Linux 6.3 with Perl 5.10.1, gfortran 4.4.7 and NONMEM 7.2.0 and 7.3.0 Scientific Linux 6.3 with Perl 5.10.1, gfortran 5.1.1 and NONMEM 7.3.0 CentOS 6.5 with Perl 5.10.1, gfortran 4.4.7 and NONMEM 7.1.0, 7.1.2 and 7.3.0 ************************ Installation ************************ Follow the installation instructions on http://psn.sourceforge.net/install.php ************************ Documentation ************************ All documentation is found in the doc subdirectory of the PsN installation directory. During installation it is also possible to have all documentation copied to a user-specified directory. All documentation can also be downloaded from the PsN website http://psn.sourceforge.net/ ************************ Dependencies ************************ - It is recommended to use Xpose version 4.5.3 or later for the -cwres functionality and the default execute, vpc and bootstrap R plots. - Each userguide lists the R packages required to obtain the diagnostic R plots. ************************ Known issues ************************ - Windows users need to be careful to choose a Perl version where the extra modules needed for PsN are available. For new Perl installations the PsN team recommends Strawberry Perl See the instructions on http://psn.sourceforge.net/install.php - Installation problem on some Windows systems where perl 'system' command does not work. Make sure perl module File::Copy::Recursive is installed *before* (re)running the installation script setup.pl ********************************************* Changes and bug fixes from PsN 4.4.8 to 4.6.0 ********************************************* ___General___ - Bug fix: Detection of non-unique parameter labels when reading a raw results file. - Bug fix: Clean up in all code to get rid of Perl warnings about uninitialized variables, numeric comparisons of non-numeric variables, deprecated functionality, and similar. - Refactored reading of NONMEM installation information from psn.conf - Bug fix: Rereading of raw_results file from disk: Replace NA entries with undef so that reanalysis correctly detects missing values. - Check that $THETA records do not contain unmatched parentheses, which would trigger an NMtran error. - Refactored PsN's internal debug message handling. ___common_options___ - New option -zip to automatically zip the m1 subdirectory. If PsN is started using option -directory, and the run directory contains a zipped m1, PsN will automatically unzip m1 before continuing the run, even if option -zip is not set. - New option -citations to print a list of references for each tool in BibTeX format. - The option -degree used for tweak_inits can now take degrees greater than 1. - Faster -help, -h, -version and some option error checking. - Now -help and -h can be run without valid NONMEM version defined in psn.conf. - Inside NM_run subdirectories, the selected (re)try files will be moved, instead of copied, to psn.mod, psn.lst etc. This is to save disk I/O. This means that the numbered (re)try files for the selected run will not be left in NM_run. - Documentation updates for threads option. - Remove .RData file after running an R script via rplots option. - Bug fix rplots option when model has no $OMEGA and/or no $SIGMA. - Refactored and modified procedure for submission and polling. Always check lowest run number first during polling. Add sleep 10000 microseconds after each monitor call except the first one, to slow down polling. Documentation in psn_configuration.pdf (job_polling_interval in psn.conf) - If a NONMEM run crashes, file PRDERR will be kept in NM_run, to facilitate debugging. ___NONMEM output parsing___ - If a NONMEM run is automatically killed, for example due to a job time limit on a cluster, the file OUTPUT with iteration results will not be appended to the lst-file by the nmfe script. If PsN option -handle_crashes or -handle_msfo is set, PsN will parse OUTPUT if the lst-file appears to not have had OUTPUT appended, and initiate a crash restart if OUTPUT contains iteration information from a started but not completed NONMEM run. - Skip reading of phi-files during regular output parsing. - Documentation (in common_options) on why PsN does not use NONMEM-computed shrinkage. - Refactored parsing of ext-files. - Refactored handling of time stamps in lst-file. - Bug fix: Reading of NONMEM lst-file when there were no unfixed parameters in the model, or when $MSFI was used, i.e. no $THETA/$OMEGA/$SIGMA in the model file. - Bug fix: handling of NONMEM termination message when no parameters to estimate but maxeval > 0. - Bug fix: Reading signification digits from lst-files with SIG. DIGITS: NaN ___Cluster/Grid___ - Bug fix: torque job submission regexp only take job ID number part of qsub response, not entire string - If, after the monitor command indicates that the cluster job is done, the NM_run subdirectory has no sign of NMtran even starting, not even any stderr output, this can be a sign of file sync problems. PsN will initiate a crash restart in this situation if option -handle_crashes is set and -crash_restarts>0. - Updated submit bash script on slurm: use local file names when checking that input control stream exists. Use printenv | sed in bash script to print job hostname to file in NM_run. ___Test scripts___ - Adapt test scripts to NMQual, if option -nmqual set in psn.conf. - Refactored tests to not trigger Perl warnings. - Changed test scripts to not create new run directory with same name as a directory that was just removed. - All test scripts can be run using -nm_version passed on via prove :: -nm_version ___psn_clean___ - Bug fix handle .ctl extension of control stream in NM_run when option nmqual is used instead of nmfe. - Cleanup of additional output files for NM7.3. - New option -rdata, on per default for levels 2-4. Cleanup of .RData file in main run directory. ___sumo___ - Updated userguide. ___bootstrap___ - Updated rplots for diagnostic output. - Documentation with instructions on how to use covmat script to obtain an empirical covariance matrix based on a bootstrap raw results file. - Bug fix: Summary calculations in bootstrap_results.csv were wrong if original model was filtered out as a result of any of the -skip options. - Bug fix: Summary calculations in bootstrap_results.csv were missing if original model was not estimated or had no .lst-file when using option -no-run_base_model - Bug fix: Summary calculations for standard errors were wrong if only a subset of non-skipped samples had successful covariance steps. - Bug fix: If option -dofv is used then set delta-ofv undefined if original model or sample had undefined ofv. - Bug fix: option -summarize after bootstrap run with option -dofv. - Bug fix: More robust search for ID in IGNORE statements to warn about renumbering of individuals during bootstrapping of data sets. - Bug fix: Filter out crashed runs (ofv and estimates NA) when doing summary calculations when those runs have not already been filtered out by any -skip option. Before the bug fix estimates from such runs were included as zeros in summary computations. - Bug fix: On 32bit windows udistr(0.5), which should be 0, is wrong. Set special case for udistr(0.5). - Bug fix: Detecting IGNORE or ACCEPT on ID when alternative NONMEM syntax without equal sign, e.g. IGNORE(ID.EQ.5), is used. - Bug fix: Skip end-of-line comments when checking code for usage of ID ___vpc/npc___ - Option -sim_table may now be either a single file or a comma-separated list of files. - New option -in_filter, to be used in combination with -rawres_input to filter out certain lines from the rawres file. - More documentation for Xpose plotting and stratification. - Better error handling when zipping a large table file. - Updated error messages. ___frem___ - Program has been completely rewritten. See frem_userguide.pdf for documentation. ___mcmp___ - Print ofv of full and reduced models to rplots preamble. If lst of either model is not available, the sum of iofv computed from phi-files is used. - Output: A parametric power estimation curve will be included in the extended rplots. - Bug fix: Automatically ensure that phi-file is copied back to m1, even if user did not request it via -nm_output. ___scm___ - Refactored code for filtering data on IGNORE/ACCEPT - Refactored code for file path handling - Bug fix computation of covariate theta bounds for linear and hockey-stick parameterization when median of covariate is negative. - Refactored covariate code generation and computation of covariate theta bounds and initial values, new tests. - Bug fix setting user-selected ofv changes for forward and backward searches when option -gof=ofv is used. See scm_userguide.pdf for examples. - Documentation on why user must encode priors with prior-specific records $THETAP etc. - Documentation updated for option -time_varying, and how means and medians are computed when option -time_varying is not set. - Updated config file examples to use -base_ofv instead of -base_criteria_values. ___boot_scm___ - Refactored code for filtering data on IGNORE/ACCEPT - Bug fix: Use common options, e.g. -nm_version, that are set on command-line. - Bug fix: If valid states not set in scm config file then use default set of valid states. ___ebe_npde___ - Renamed script to simeval. See below. ___simeval___ (was ebe_npde) - New output files, including diagnostic R plots. Some intermediate output files removed. See details in simeval_userguide.pdf - Refactored output computations. - Computation of npde for IWRES and CWRES. - Separate decorrelation of iiv ebe:s, and iov ebe:s per occasion - New option -n_simulation_models, allowing more than one sample per NONMEM control stream - New option -idv, default TIME. - New option -extra_variables, to append columns to NONMEM tables for evaluation of original and simulated data. - Additional default columns in table output: PRED, IPRED if defined, idv, DV, MDV if available - simeval will automatically create vpc plots of DV vs PRED and CWRES vs idv (TIME). - Bug-fix: Ensure that phi-files are copied back to m1 even if -nm_output does not include phi. ___update_inits___ - The option -degree can now take degrees greater than 1. - New functionality for automatic Cholesky reparameterization of OMEGA and SIGMA, and automatic inverse reparameterization. Reparameterization can be done with or without bounded thetas. - update_inits no longer requires a source (model or lst-file) to read parameter estimates from, it is possible to update other model properties anyway. - Do not update tag based_on if output and input models are the same - New option -silent, to suppress all log messages. - Refactored code for file path handling - Bug fix alias update for update_inits. Script update is now an identical copy of update_inits. ___execute___ - Refactored input checking. ___parallel_retries___ - new options -keep_table_files and -keep_nm_output to keep output from not chosen retry models in main rundirectory even if option -clean >= 2. ___nca___ - Cleanup table files in NM_run subdirectories. - Automatically zip table files. Note: Latest ncappc R package handles zipped table files. - Print warning if no simulation table file was generated. ___covmat___ - New option -require_numeric_ofv, only use parameter vectors where ofv column has a value. - Changed ordering of columns in output matrix, use same order as NONMEM. ___sir___ - Major redesign of procedure, see sir_userguide.pdf for a complete description. Now sir runs an iterative procedure with Box-Cox transformation for more efficient sampling of assymetric distributions, recentering when input model found to be in local minimum, and adjustment of sampling and resampling to compensate for failed samples. - More alternatives for proposal density when covariance matrix unavailable: Automatic creation of proposal density based on guesses of parameter rse:s or parameter variances, or based on a raw_results file that will be automatically extended to full rank if the number of parameter vectors is too small. - Extended resume/restart functionality - Automatic rplots with basic and extended diagnostic graphs - Default settings for -samples and -resamples - New option -cap_resampling for resampling with limited replacement. - Individual inflation theta, omega and sigma per parameter using options -theta_inflation etc. Removed old option -inflation. - New format for raw_results file: First line is always center vector, which can be different from input model if input model was at local minimum. New column resampling order. - Refactored code for checking that sampled parameter vectors have positive definite $OMEGA and $SIGMA blocks. - Automatic detection of Cholesky reparameterized $OMEGA/$SIGMA blocks and checks that sampled parameter vectors give positive definite inverse parameterized blocks. - Print summary of causes of sample rejections (based on boundary violations, non-positive definite $OMEGA/$SIGMA blocks etc) to file. - If many samples are rejected, sir can modify sampled parameter vectors to force non-positive definite $OMEGA/$SIGMA blocks to positive definite. See details in sir_userguide.pdf. ___precond___ - New option -copy_data. Used in the same way as for execute - Cleanup of error messages and information about unidentifiability - Option -always is changed to dummy option. Functionality is always on, the option is only kept for backwards compatibility. - New option -notalways to only precondition after a failed covariance step. - Bug fix handling of case when S matrix is unobtainable. ___benchmark___ - A new tool for systematic comparisons of ofv, parameter estimates and run times across different combinations of NONMEM record settings, THETA initial estimates and NONMEM/compiler versions. See details in benchmark_userguide.pdf. ___psn___ - The option -nm_versions now also print the path to the psn.conf that was used, and, if defined in psn.conf, a description of each NM version. ___nmoutput2so___ - Bug fix: misspelling of Parameter in SE and RSE table - Bug fix: Don't create covariance_step_warnings and covariance_step_successful warnings if $COV not specified - Bug fix: Don't give message minimization_successful after simulation - Add Bootstrap/PrecisionEstimates. Currently with one single alpha. - Add bootstrap parameters not on var/cov scale. Also add a warning listing these parameters. - Add description to TimeStamp explaining that unit is hours. - Add more PsN result files to RawResults. Also set specific oids for easier retrieval ___crossval___ - The crossval script now prints version_and_option_info.txt to the run directory. - Bug fix setting of run subdirectory names to non-default to avoid rapid removal and recreation of subdirectories with identical names, as this increases the risk of file sync problems on a cluster/grid. ___randtest___ - New rplots template for diagnostic plots. - Bug fix: If estimation of a sample fails then set delta-ofv for that sample to undef and exclude it from diagnostic plots. - Handle restart of partially complete randtest with some samples already run, only start remaining samples. ___lasso___ - Refactoring of lasso model code generation and computation of lasso coefficients. - Bug fix: Computations of summary statistics for categorical covariates, centering of categorical covariates and covariates with piece-wise linear relations in final model. - New option -log_scale. Include covariates on log scale instead of normal scale. Effect is added to LNTV instead of multiplied with TV. - New option -no-normalize, to turn off normalization of covariates by their standard deviation. By default normalization is performed. - New option -adaptive, to run additional lasso iterations with coefficients modified by lasso coefficient from previous iteration. - New option -external_data, to create model for external evaluation of adaptive lasso results. - New option -adjusted, to run adjusted adaptive lasso where coefficients in first iteration of adaptive lasso are computed from standard errors of full model. - New option -al_coefficients, for use in combination with -adjusted. Input lasso coefficients on command-line instead of running covariance step of full model to obtain them. - New output file lasso_coefficients.csv with lasso coefficients and convergence information. - Added lasso coefficients to raw_results file. - New option -run_final_model, default not set. Decides whether final model on normal scale should be run. - Removed option -lst_file - Bug fix setting of run subdirectory names to non-default to avoid rapid removal and recreation of subdirectories with identical names, as this increases the risk of file sync problems on a cluster/grid. __rawresults___ - New option -silent. ___se_of_eta___ - Program is removed, because the program output is given by NONMEM in phi-file.