This PsN version has been tested on the following systems: - MS Windows XP with ActiveState Perl 5.12.2 and NM 7.2.0. - MS Windows 7 64bit with ActiveState Perl 5.16.3 and NM 7.2.0. - Mac OS X 10.6.6 with Perl 5.10.0 and NM 7.2.0. - Linux Ubuntu 13.04 with Perl 5.14.2 and NM 7.2.0. - Scientific Linux release 6.3 with Perl 5.10.1 and NM 7.2.0 - Linux CentOS 5.9 with Perl 5.8.8 and NM 6.1, 6.2, 7.1.0, 7.1.2, 7.2.0 and 7.3.0 beta-26AUG2013. ************************ 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 ************************ - Xpose must, if used, be version 4.2.1 or later to handle PsN-3.5.3 or later output format. It is recommended to use the latest Xpose release from http://xpose.sf.net ************************ Known issues ************************ - Windows users ***who are to install PsN for the first time*** need to be careful to choose a Perl version where the extra modules needed for PsN are available. See the instructions on http://psn.sourceforge.net/install.php Windows users ***who are already running an older version of PsN*** do not need to upgrade the Perl version. - 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 - On Windows PsN has only been tested with ActiveState Perl. Strawberry Perl is not recommended. ********************************************* Changes and bug fixes from PsN 3.6.2 to 3.7.5 ********************************************* ___General___ - New program linearize for linearization of a model. Use command linearize -h for help. - New program randtest for randomization testing. See randtest_userguide.pdf for details. - Major refactoring of all code. All programs have been cleaned up. Removed debug module and replaced with Perl core module Carp.pm. Removed modules extra_data, xrsl_file. Removed option adaptive, and related code. Removed all support for automatic wrapping of data files, including scripts create_cont_data, single_valued_columns and unwrap_data, since recent versions of NONMEM can handle any number of columns in $INPUT and $TABLE when $SIZES is set. Removed usage of external Run3 module. - New format of all userguides to increase readability. - More error checking in installation script, detect and handle configuration errors in Perl installation. - Easier diagnose of errors when -clean >=3. If PsN detects an error with the run that the user needs to check for in the lst-file, PsN will not remove the NM_run subdirectory with the lst-file even if clean >=3. - Bug fix: When options -verbose and -clean=3 are used in combination do not try to read output from file that has been cleaned. - Bug fix: Fixed the output object error that would give message "function can only be called where output object parsed successfully" in e.g. scm. ___Cluster/grid interaction___ - All grid/cluster submissions are now made using Perl backtick operator or perl 'system' command, not the run3 sub-routine of the external Run3 module as in earlier PsN versions. - Running on sge: Removed : (colon) from regular expression in sge queue monitoring code so that PsN will handle qstat response from Univa Grid Engine 8 (a fork of SGE 6.2u5). - Running on sge: If -run_on_sge_nmfe is set then print qsub command to file qsubcommand in NM_run. - Running on slurm: The option -slurm_project has been renamed to slurm_account, to match the slurm documentation. The option maps to -A as in earlier PsN versions. - Running on slurm: A new option -slurm_partition, which maps to -p. - Running on slurm: Print sbatch command to file sbatchcommand in NM_run. - Running on slurm: Bug fix: PsN-3.7.6 will pass on -nmfe_options and -parafile to the nmfe script when option -run_on_slurm is set ___NONMEM interaction___ - New option -check_nmtran, set by default. PsN will run NMtran on models before forking on a local computer or submitting to a grid/cluster. This saves time in case of simple control stream syntax errors, since the user does not have to wait for the run to get a slot in the queue to get the error message. The check will only be done if the path to a NONMEM installation directory has been set in psn.conf, rather than the path to an executable script. This functionality can be disabled with -no-check_nmtran. PsN will only run the check on the first model in a set of models that are essentially copies of the same original model, e.g. in a bootstrap or sse. - In $PK $PRED $ERROR do not print a line break after $, to avoid a bug in NM7.3 beta - Use Perl backtick operator or Perl 'system' command to invoke nmfe script instead of run3 routine from external Run3 module. - Bug fix: When the setting of the NONMEM installation directory is incorrect in psn.conf and PsN tries to diagnose the error, PsN will distinguish between incorrect setting of a directory and an executable file. - Bug fix: Do not treat the line 0OBJ. FUNCT. IS NOT DEFINED as a sure sign of run failure if found in the lst-file. ___sse___ - New column in raw_results file that identifes the simulated dataset used for fitting. - More informative error messages and improved detection of input errors. - Shorter file names for estimation files, to give a better overview in cluster monitoring tools (qstat, squeue...). - Store simulation raw_results file in main run directory so that it will be saved even when -clean=3. - Additional variants of bias and rmse in sse_results.csv. - More flexible and robust -rawres_input functionality to allow raw_results files from previous sse runs to be used as input. Also make format rules for the input file more flexible, so that quotes are only necessary around header cells that contain a comma. - Bug fix: Set correct label for DV in sse_results.csv - Bug fix: When simulating with uncertainty, compute the correct bias and rmse, using the individual simulation parameter values as reference. - Bug fix: When simulating with uncertainty using $PRIOR, collect the parameter values used for simulations so that they can be used as reference when computing results. - Bug fix: Set numbered names of table files in simulation model when using -add_models and original run had option -no-estimate_simulation set. - Bug fix: Print full PsN version name in sse_results.csv. ___scm___ - Bug fix: Handling of -missing_data_token has been completely rewritten to handle other tokens than -99. In earlier PsN versions the data statistics (median, max, min) would not be computed correctly if the covariate column had missing data and -missing_data_token was anything else than -99. - Automatic setting of $SIZES if NM version is 7.2 or later and number of needed items in $INPUT or $TABLE exceeds 50. - Automatic setting of $SIZES LTH if more thatn 40 THETA needed in a candidate model. - Bug fix: scm will ignore any header in the data file, which does not need to have a header at all. The scm program will only look at $INPUT of model file when searching for covariates. - Bug fix: When -linearize is set, use adaptive FORMAT statement when default FORMAT in $TABLE would trigger bug in NONMEM that data records can be at most 999 characters long. - Bug fix: When -linearize is set, if MCETA is set in the input model then this MCETA setting will be copied to the linearized model. - Bug fix: When computing the median of a covariate, use only one value per individual. ___vpc/npc___ - Automatic binning for vpc, used by default. New option -auto_bin controls the auto-binning, see vpc_npc_userguide.pdf for details. The binning algorithm is described in vpc_binning.pdf. - Ten times faster reading and formatting of DV data after rewrite of parsing algorithm. - vpc will compute mean and confidence intervals for the mean. Use latest version of Xpose to visualize. - New option -refstrat, see userguide for details. If set then vpc will compute delta-mean, i.e. difference in mean between current stratum and reference stratum set with -refstrat option. Use latest version of Xpose to visualize. - New option -fine_pi to compute more percentiles, see userguide. - Changed run directory structure so that NM_run1 and NM_run2 are found in a modelfit_dir subdirectory, and the lst-files from the simulation are copied to m1 instead of the top level of the run directory. - vpc will now accept a list of variables for option -stratify_on. The first variable in the list will be used for stratification in the current run, and the rest will be added to table output so that those variables can be used for stratification in later restarts without needing to rerun simulations. - More flexible and robust -rawres_input functionality to allow raw_results files from previous sse runs to be used as input. Also make format rules for the input file more flexible, so that quotes are only necessary around header cells that contain a comma. - Bug fix: Make sure the label for the simulation file in vpc_results.csv is correct when flip_comments is not used. - Bug fix: Do not try to update initial estimates from an existing lst-file unless it has an estimation step that was run. - Bug fix: Ignore comment lines when doing input checking on control stream. - Bug fix: Make sure full PsN version number is written to vpc_results.csv - Bug fix: More clear error message when option -lower_bound needs to be set/adjusted because a PRED value below the lower bound was found. ___bootstrap___ - Code for automatic generation of R scripts for bootstrap results is no longer maintained and may not work. Functionality for handling bootstrap results is incorporated into latest version of Xpose. - New option -dofv: Run a MAXEVAL=0 run with each set of bootstrap parameter estimates on original data, and compute delta-ofv compared to original estimates. Related option -mceta, which maps to $EST MCETA, allowed for NM7.3. - bootstrap will now only accept a column name from $INPUT as value for option stratify_on. Column number not accepted. - Bug fix: The bootstrap program can be run with option -directory set to a global path when option -bca is set. - Bug fix: Ignore comment lines when checking input model for synonym use. ___RunRecord___ - Bug fix: Improved reading of data file name and parsing of documentation comments. - Bug fix: No empty headers printed for non-existing sigma or omega. ___psn_clean___ Bug fix: Remove simulation_dir subdirectories completely if option -level is >= 3. ___update_inits___ The program will now handle renumbering of file names with .csv-extension in $TABLE. ___extended_grid___ Bug fix: Better detection of missing files and more clear error messages if NONMEM runs fail. Bug fix: Always use headers in $INPUT instead of any header in the data file. ___sumo___ Bug fix: Set 'WARNING' flag for Covariance step if lst-file says S or R matrix is singular. ___nonpb___ Bug fix: When -nonpb_version=2 program will set sample size for bootstrap call. ___mcmp___ Bug fix: Only print undropped columns in $TABLE of simulation model. If using full model as simulation model then exit with message saying dropping columns in $INPUT is not supported when full model is used as simulation model. ___lasso___ Bug fix: If the input model has thetas with estimates smaller than the cutoff, lasso will no longer cut these thetas and destroy the model.