State-of-the-art version of the Particle Swarm Optimisation (PSO) algorithm (SPSO-2011 and SPSO-2007 capable). hydroPSO can be used as a replacement of the 'optim' R function for (global) optimization of non-smooth and non-linear functions. However, the main focus of hydroPSO is the calibration of environmental and other real-world models that need to be executed from the system console. hydroPSO is model-independent, allowing the user to easily interface any computer simulation model with the calibration engine (PSO). hydroPSO communicates with the model through the model's own input and output files, without requiring access to the model's source code. Several PSO variants and controlling options are included to fine-tune the performance of the calibration engine to different calibration problems. An advanced sensitivity analysis function together with user-friendly plotting summaries facilitate the interpretation and assessment of the calibration results. hydroPSO is parallel-capable, to alleviate the computational burden of complex models with "long" execution time. Bugs reports/comments/questions are very welcomed (in English, Spanish or Italian). See Zambrano-Bigiarini and Rojas (2013)
0.4-0 10-Jun-2018 o Repository management moved from SVN to GIT, first using rforge and finally using Github. o Main page in Github list articles reporting the use of hydroPSO until May 31th 2018. o Package tested against R version 3.5.0 (2018-04-23) -- "Joy in Playing", following the request made by CRAN. o TRAVIS CI used to test source code against stable and development R versions. o zenodo used to manage DOI of different versions of the package
0.3-4-6 09-May-2016 o New field 'Additional_repositories' in the DESCRIPTION file, following the instructions given by CRAN. o Minor modification of the 'Description' field, following the instructions given by CRAN (avoiding redundancies). o CITATION file modified to allow working without package installed (following CRAN request). o DOI field added to CITATION file. o CRAN website removed from URL field in DESCRIPTION file (following CRAN request). o 'read_out' : observations (in 'obs' or 'Observations.txt') are allowed to be of class (Date", "POSIXct", "POSIXt"), not only 'Date' as before this enhancement. o 'plot_results': observations (obtained with 'read_results.txt') are allowed to be of class (Date", "POSIXct", "POSIXt"), not only 'Date' as before this enhancement. o 'hydroPSO' : -) Issue #4. hydroPSO is now able to optimise 1D functions, both R-defined functions or R-external model codes (Thanks to Azza Ahmed and @cjzilverberg!). -) Issue #3. hydroPSO now is able to correctly apply the 'method="wfips"' when 'topology="random"' (Thanks to @drknexus !). For a 2-dimensioanl problem wit 40 particles, it raised an error similar to: "Error in is.data.frame(x) : dims [product 2] do not match the length of object " -) minor bug fixed (Thanks to Aaron Jones !). This bug might have triggered when the user provided more than 1 parameter set as a first guess for the optimal solution (i.e, a matrix/data.frame), using the 'par' optional argument. Some of the parameters provided by the user might have been incorrectly compared against the 'lower' and 'upper' values.
0.3-4 11-Abr-2014 o Package tested against R version 3.1.0 (2014-04-10) -- "Spring Dance". o 'lhoat' : -) when 'normalise=TRUE' the 'ParameterSets' output is now given in the real range of each parameter, and not in the normalised range [0, 1] as up to version 0.3.2. The 'lhoat-gof.txt' output file has always been written in the real range of each parameter when 'normalise=TRUE'. Thanks to Romain Lardy ! -) fixed error: 'Error in Thetas * (UPPER.ini - LOWER.ini) : non-conformable arrays', which was triggered when 'normalise=TRUE' & 'model.FUN="hydromod"'. -) now works correctly on Windows machines (when 'parallel="parallelWin"') (thanks to Dipangkar Kundu !) -) fixed error : ' Error in argfun(start + i - 1L) : object 'model.out.text.file' not found', which was triggered when 'parallel="parallelWin" AND 'write2disk=FALSE' -) 'parallel_logfile.txt' is now removed before start running this function (when 'parallel!="none"') -) 'parallel::makeCluster' was replaced by 'makeCluster' (and importFrom in NAMESPACE) -) 'parallel::makeForkCluster' was replaced by 'makeForkCluster' (and importFrom in NAMESPACE) -) 'parallel::detectCores' was replaced by 'detectCores' (and importFrom in NAMESPACE) -) 'parallel::clusterCall' was replaced by 'clusterCall' (and importFrom in NAMESPACE) -) 'parallel::clusterExport' was replaced by 'clusterExport' (and importFrom in NAMESPACE) -) 'parallel::mclapply' was replaced by 'mclapply' (and importFrom in NAMESPACE) -) 'parallel::parRapply' was replaced by 'parRapply' (and importFrom in NAMESPACE) -) 'parallel::clusterApply' was replaced by 'clusterApply' (and importFrom in NAMESPACE) -) 'parallel::stopCluster' was replaced by 'stopCluster' (and importFrom in NAMESPACE) -) 'formatC' calls were wrapped into a 'suppressWarnings' statement (to avoid unnecessary warnings, due to changes in R 3.0.2). o 'hydroPSO' : -) The 'Dimension' field is now correctly written into the 'PSO_logfile.txt' when 'parallel!=none' (in version 0.3-3 the number of cores/nodes was written instead of the dimension of the problem) -) 'parallel_logfile.txt' is now removed before start running this function (when 'parallel!="none"') -) now it shows the number of detected cores (when 'parallel!="none" and 'verbose=TRUE') -) 'parallel:::makeCluster' was replaced by 'makeCluster' (and importFrom in NAMESPACE) -) 'parallel:::makeForkCluster' was replaced by 'makeForkCluster' (and importFrom in NAMESPACE) -) 'formatC' calls were wrapped into a 'suppressWarnings' statement (to avoid unnecessary warnings, due to changes in R 3.0.2). o 'wquantile' : -) fixed error: 'Error en setTxtProgressBar(pbar, i) : object 'pbar' not found', which was triggered when 'verbose=FALSE'. Thanks to Antonio Piccolboni ! o 'plot_NparOF' : 'formatC' calls were wrapped into a 'suppressWarnings' statement (to avoid unnecessary warnings, due to changes in R 3.0.2). o 'verification' : 'formatC' calls were wrapped into a 'suppressWarnings' statement (to avoid unnecessary warnings, due to changes in R 3.0.2). o 'ModifyInputFile' : 'formatC' calls were wrapped into a 'suppressWarnings' statement (to avoid unnecessary warnings, due to changes in R 3.0.2). o 'plot_results' : 'x11()' was replaced by 'dev.new()' o 'plot_particles' : 'x11()' was replaced by 'dev.new()' o 'params2ecdf.default': 'x11()' was replaced by 'dev.new()' o 'read_out' : 'require(zoo)' was reaplaced by '!is.na( match("zoo", installed.packages()[,"Package"] ) )' o 'lhs' package was moved from 'Suggests' into 'Imports' in NAMESPACE file. o 'zoo' package removed from 'Suggests' (due to changes in CRAN policies) but it stays on 'Imports'
o 'hydroPSO' : fixed error introduced (not intentionally, of course !) in version 0.3-2:
Error in FUN(1:number[[1L]], ...) :
unused arguments (Particles = ...)
o 'pest2hydroPSO' : -) typos fixed in the (automatically created) 'hydroPSO-Rscript.R' file (Thanks to Ramadan Abdel A. !):
*) line 38: ending comma removed
*) line 57: added ending ', sep=""', to avoid white spaces in the filename obs.fname <- paste(model.drty, "/PSO.in/", obs.fname, sep="")
-) file 'ParamFiles.txt' (automatically created) uses the correct row number for each input file. Up to hydroPSO <= 0.3-2 the row number for each parameter had a shift of 1 line (e.g., it used '5' when the correct number was 4) o NAMESPACE: functions 'read.ParameterRanges' and 'rLHS' are now exported
0.3-2 29-May-2013 o 'lhoat' : -) parallel capable -) much faster than in previous hydroPSO versions, even without using the new 'parallel' option (this function was completely re-written, in order to make it parallel capable) -) new 'normalise' parameter for the 'control' argument, in order to improve the sampling design when the search space is not a hypercube -) new 'RelativeImportance.Norm' column at the end of the output ranking. The values in 'RelativeImportance.Norm' provide normalised values of the relative importance of each parameter. The sum of all the 'RelativeImportance.Norm' must be 1. (Thanks to Rui Esteves for suggesting this feature !) -) 'control' argument now allows the following new components: 'REPORT', 'parallel', 'par.nnodes', 'par.pkgs', 'normalise' -) new argument '...' to allow function arguments that are kept constant during the sensitivity analysis. o 'pest2hydroPSO' : -) now it handles correctly cases where the fifth row of 'pst.fname' uses tab instead of white spaces as separator (Thanks to Ramadan Abdel A. !) o 'hydroPSO' : -) minor bug fixed. When the goodness-of-fit value of a given parameter set is not finite, the parameter values are now correctly written into 'BestParamPerIter.txt' -) when 'fn.name!="hydromod"' AND 'normalise=TRUE' (i.e., an R function and NOT an external model run from the system console) the function 'fn' is now evaluated with the correct (non normalised) parameter values (up to hydroPSO <= 0.3-1-1 'fn' was evaluated with the (non-correct) normalised values when 'fn.name!="hydromod"' AND 'normalise=TRUE')
0.3-1-1 10-May-2013 o Suggested package 'multicore' (superseded) was replaced by 'parallel' o 'hydroPSO: functions from the 'multicore' package were replaced by equivalent from the 'parallel' package (detectCores, mclapply)
o new 'SWAT2005.zip' file, with and updated tutorial for calibrating SWAT 2005, compatible with hydroPSO >= 0.3-0
The old zip file was renamed to 'SWAT2005_old.zip', which is compatible with hydroPSO <= 0.1-58 The new 'SWAT2005.zip' file now includes a new 'swat2005.out' binary file, in order to allow running SWAT2005 in GNU/Linux machines o new 'MF2005.zip' file, with and updated tutorial for calibrating MODFLOW 2005, compatible with hydroPSO >= 0.3-0
The old zip file was renamed to 'MF2005_old.zip', which is compatible with hydroPSO <= 0.1-58
o 'lhoat' : -) fixed bug that led to all parameters have the same Ranking (equal to 0) when 'fn!="hydromod"' (Thanks to Rui Esteves !) -) fixed bug that led to the following error message "object 'model.out.text.fname' not found" when 'write2disk=FALSE' -) 'fn' argument now can be any R function or a character. In the latter case, it can be "hydromod" or the name of a valid R function. In previous versions of 'lhoat' only a character type was accepted (Thanks to Rui Esteves !) o 'hydroPSO' : -) if a 'parallel' directory was not successfully removed during a previous run, it is removed before running the algorithm in parallel mode again. o 'ReadPlot_results': updated description of the 'dp3D.names' argument
o 'plot_out' : 'main' argument is not overwritten to "Observed vs 'best' Simulation" when this argument is provided and 'ptype=="ts"' o 'plot_results' : 'main' argument is now (internally) passed to the 'plot_out' function. o 'plot_particles' : 'main' argument is now omitted for the 3D dotty plots ('plot_NparOF' function), in order to avoid confusing (and sometimes unnecessary) titles when 'main' is too long
o 'plot_NparOF' : 'main' argument was disabled, in order to avoid confusing titles when 'main is a long text
0.3-0 19-Dec-2012 o 'hydroPSO' : -) parallel capable (OS-independent) * new argument 'parallel' to define how to parallelise the evaluation of the objective function * new 'control' argument 'par.nnodes' to define the number of cores/nodes to be used in a multi-core machine or network cluster * new 'control' argument 'par.pkgs' (only needed for Windows machines) with the list of package that need to be loaded on each core/node -) Default number of particles for 'method=spso2007' was changed from 'ceiling(10+2sqrt(n))' to 'floor(10+2sqrt(n))', following Clerc 2012
o New vignette (14/Dec/2012) o 'read_out' : -) new argument 'nsim' used to specify the number of model outputs. It is useful when the model to be calibrated returns NA instead of the simulated values for some particles (e.g., MODFLOW) o 'read_results': -) See 'read_out' above o 'plot_results': -) See 'read_out' above
24-Jun-2011 May-2011 Apr-2011 29-Dec-2010 23-Dec-2010 Dec-2010 Nov-2010 2009-Oct2010: hydroPSO hibernation :( ~Jul-2008: hydroPSO beginning...