Skip to content
Snippets Groups Projects
Commit 4fb91b66 authored by Mauricio Zambrano-Bigiarini's avatar Mauricio Zambrano-Bigiarini
Browse files

lhoat.Rd: new and improved examples. DESCRIPTION and NEWS updated. Partof NEWS... 

lhoat.Rd: new and improved examples. DESCRIPTION and NEWS updated. Partof NEWS was moved into ChangeLog
parent dcb9479f
No related branches found
Tags exam-submission
No related merge requests found
Expand all Hide whitespace changes
Inline Side-by-side
ChangeLog
+ 196
− 0
View file @ 4fb91b66
0.2-0 29-Nov-2012
o 'hydroPSO' : -) SPSO 2011 capable, which includes: (check the latest \url{clerc.maurice.free.fr/pso/SPSO_descriptions.pdf})
* new default value for swarm size (npart=40),
* new equation for updating the velocity ( V[t+1] = w*V + Gr -x + alea_sphere(Gr, ||Gr-x||) ),
* new initialization of the velocity ( Vini = U(lower-Xini, upper-Xini) )
* new confinement of the velocity ( V[t+1] = -0.5 * V[t], when x[t+1] > x_max | x[t+1] < x_min )
* optional normalisation of parameter values
-) improved performance: ~ 33% faster for 32-bit machines and 38% faster for 64-bit machines.
Tested on 10-, 20- and 30-D benchmark functions with 'write2disk=FALSE'.
-) new 'normalise' parameter for the 'control' variable, in order to improve the performance when the search space is not
a hypercube
-) '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 'hydroPSO' only a character type was accepted.
-) now it handles models with sub-daily time step and with sub-daily observations (thanks to O. Rakovec !).
Suggested dependence on hydroGOF >= 0.3-5 and hydroTSM >= 0.3-6
-) new output file 'BestModel_out.txt', with the model outputs corresponding to the best parameter set.
Only available when 'fn.name=="hydromod"'
-) normalised swarm radius now it is computed using the median distance of all the particles the global best, instead of
using the maximum distance (as proposed in Evers and Ghalia 2009). This was done in order to make easier the
identification of the stagnation point for activating regrouping
-) when 'use.RG=TRUE', the default values for 'RG.thr', 'RG.r', and 'RG.miniter' were changed (related to the change in the
computation of swarm radius), from 1.1E-4, 0.8 and 5 to 1E-5, 2 and 100, respectively
-) argument 'method' now allows the following 3 new values: 'spso2007', 'spso2011' and 'canonical', which set the values of the PSO
engine to the ones of the SPSO 2007, SPSO 2011 and the canonical one, respectively.
-) argument 'method'. The old (default) value 'pso' was replaced by the new (default) value 'spso2011'
-) argument 'Vini.type' now allows 2 new values for setting the initial velocity of the particles according to the equation
specified in the SPSO 2011: 'random2011' and 'lhs2011', which use a uniform distribution or with a Latin-hypercube
sampling, respectively.
-) argument 'Vini.type'. The old values 'random' and 'lhs' were replaced by 'random2007' and 'lhs2007', in order to make
clear that it follows the equation described in SPSO 2007.
-) argument 'Vini.type', when missing its value depends on the value of the 'method' argument:
method == 'spso2007' => Vini.type='random2007'
method != 'spso2007' => Vini.type='random2011'
-) argument 'Xini.type', now its default value is 'random' instead of 'lhs', in order to make it compatible with the default
value for 'method' (method='spso2011')
-) argument 'npart', when missing its value depends on the value of the 'method' argument:
method == 'spso2007' => npart=10+2*[sqrt(n)]
method != 'spso2007' => npart=40
-) argument 'boundary.wall' now allows the following new values: 'absorbing2011', which set 'boundary.wall' to the
absorbing condition specified in the SPSO 2011
-) argument 'boundary.wall'. The old value 'absorbing' was replaced by 'absorbing2007', in order to make clear that is
follows the equation described in SPSO 2007.
-) argument 'boundary.wall', when missing its value depends on the value of the 'method' argument:
method == 'spso2007' => boundary.wall='absorbing2007'
method != 'spso2007' => boundary.wall='absorbing2011'
-) in the documentation, default values are now mentioned for each argument of 'control'.
-) the 'PSO_logfile.txt' output file now contains information about: 'best.update', 'random.update', 'Xini.type', 'Vini.type'
In addition, information about 'IW.type' and 'IW.exp' is now written only when 'length(IW.w) > 1'
-) output file 'BestParamPerParticle.txt' now has the GoF as first column (previously it was the last one)
-) fixed velocity equation for 'boundary.wall="reflecting"' :
ver <= 0.1-58 -> ver >= 0.2-0
v[t+1] : v[t] -> -v[t] (when x[t+1] > x_max | x[t+1] < x_min )
-) changed velocity equation for 'boundary.wall="damping"' :
ver <= 0.1-58 -> ver >= 0.2-0
v[t+1] : v[t] -> -v[t] (when x[t+1] > x_max | x[t+1] < x_min )
-) when the control argument 'out.with.fit.iter' (not used so far) is set to TRUE, the number of iterations returned
correspond to the effective number of iterations carried out, not the maximum defined by 'maxit'
-) observed values are now correctly written to disk for sub-daily models
-) the package now depends on R >= 2.13.0, not 2.10.0 as in previous releases, in order to be consistent with the
byte-compiler setting of the package
-) fixed some (very unlikely) error when 'IW.type="aiwf"'
o Running 'hydroPSO' >= 0.2-0 with default settings will produce DIFFERENT RESULTS from those obtained with 'hydroPSO' <= 0.1-58, due to the
following changes in default values:
ver <= 0.1-58 -> ver >= 0.2-0
-) npart : 10+2*[sqrt(n)] -> 40
-) Xini.type : 'lhs' -> 'random'
-) Vini.type : 'lhs2007' -> 'random2011'
-) boundary.wall: 'reflecting' -> 'absorbing2011'
-) TVc1.type : 'non-linear' -> 'linear' (no effect, because 'use.TVc1=FALSE' by default)
-) TVc2.type : 'non-linear' -> 'linear' (no effect, because 'use.TVc2=FALSE' by default)
-) TVlambda.type: 'non-linear' -> 'linear' (no effect, because 'use.TVlambda=FALSE' by default)
-) boundary.wall: 'reflecting' in hydroPSO ver <= 0.1-58 is not longer equivalent to 'reflecting' in
hydroPSO ver >= 0.1-58
o 'hydromod' : -) now it is able to handle sub-daily models
-) if the 'hydroGOF' package is not available, a simple correlation plot is produced between observations and
the best simulation
-) improved documentation
-) removed default value for 'exe.name' (previously './swat.out')
o 'test_functions' : -) new benchmark functions:
'schwefel' : Schwefel function
'ssphere' : shifted Sphere (CEC 2005),
'srosenbrock' : shifted Rosenbrock (CEC 2005),
'srastrigin' : shifted Rastrigin (CEC 2005),
'sgriewank' : shifted Griewank (CEC 2005),
'sackley' : shifted Ackley (CEC 2005),
'sschwefel1_2': shifted Schwefel's Problem 1.2 (CEC 2005)
-) added equations, description and references for each test function.
o 'plot_convergence': -) the label 'Gbest' was replaced by "Global Optimum' in the title of the plot and in the label of the 'y' axis, in
order to make it more intuitive to people non-familiar with PSO
o 'read_particles' : -) new argument 'do.pairs' (by default 'do.pairs=FALSE'), to control if the correlation matrix among parameters has to be plotted or not
(up to hydroPSO <=0.1.58 it was always plotted)
o 'plot_particles' : -) new argument 'do.pairs' (by default 'do.pairs=FALSE'), to control if the correlation matrix among parameters has to be plotted or not
(up to hydroPSO <=0.1.58 it was always plotted)
-) arguments 'MinMax' and 'beh.thr' are now passed to 'plot_NparOF'
-) arguments 'gofs', 'MinMax' and 'beh.thr' are now passed to 'params2ecdf'
o 'plot_results' : -) new argument 'do.pairs' (by default 'do.pairs=FALSE'), to control if the correlation matrix among parameters has to be plotted or not
(up to hydroPSO <=0.1.58 it was always plotted)
-) default value for 'alpha' argument was changed from 1 to 0.65
o 'plot_out' : -) argument 'MinMax' is not required any more!
-) argument 'sim' may also be 'integer' (before it had to be 'numeric', which does not allow 'integer' objects !)
o 'plot_NparOF' : -) new argument 'beh.thr' to allow selecting only behavioral parameter sets
-) argument 'MinMax' now accepts a NULL value (default)
-) default value for 'alpha' argument was changed from 1 to 0.65
o 'plot_2parOF' : -) default value for 'alpha' argument was changed from 1 to 0.65
o 'params2ecdf' : -) new arguments 'gofs', 'MinMax' and 'beh.thr'
o new function 'pest2hydroPSO' for importing PEST input files into hydroPSO
o new function 'hydroPSO2pest' for exporting hydroPSO input files to PEST
0.1-58 14-Sep-2012
o 'hydroPSO' : -) 'random.update' is now ONLY used when 'best.update="async". In hydroPSO 0.1-57 'random.update' was set to TRUE
by default, independent of the 'best.update="sync" value.
-) argument values are now effectively checked for: 'MinMax', 'Xini.type', 'Vini.type', 'best.update', 'boundary.wall',
'topology', 'IW.type', 'TVc1.type', 'TVc2.type', 'TVlambda.type'.
-) more and better examples
o 'test_functions': -) name of the "rastrigrin" function was changed to its correct name of "rastrigin", and modified in all the examples.
The old (and wrong) name 'rastrigrin' is kept only for backwards compatibility.
-) The definition of the 'schafferF6' function was corrected (changed) from:
0.5 + ( ( sin( sqrt( sum( x^2 ) ) ) )^2 - 0.5) / ( ( 1 + 0.01*sum(x^2) )^2 )
to:
0.5 + ( ( sin( sqrt( sum( x^2 ) ) ) )^2 - 0.5) / ( ( 1 + 0.001*sum(x^2) )^2 )
0.1-57 29-Jun-2012
o 'hydroPSO' : -) added '...' parameter. It is only used when 'fn' is different from "hydromod". This is only done for 'optim' compatibility.
-) fixed small bug related to improper PSO evolution when 'best.update'=="async" (NOT the default option !)
AND (topology!="gbest" | method!="pso")
-) new 'random.update' parameter for the 'control' variable, in order to allow random update of personal/global best
-) default value of the inertia weight 'IW' was changed from (linearly decreasing) IW.w= c(1.2, 0.4) to 'IW.w=1/(2*log(2))'
-) default value for the time-varying c1 was changed from (non-linearly decreasing) 'TVc1.rng= c(2.2, 1.8)' to
'TVc1.rng= c(1.28, 1.05)'. However, by default 'use.TVc1= FALSE' still.
-) default value for the time-varying c2 was changed from (non-linearly increasing) 'TVc1.rng= c(1.8, 2.2)' to
'TVc2.rng= c(1.05, 1.28)'. However, by default 'use.TVc2= FALSE' still.
-) default value for the time-varying lambda was changed from (non-linearly decreasing) 'TVlambda.rng= c(0.9, 0.1)' to
'TVlambda.rng= c(1, 0.25)'. However, by default 'use.TVlambda= FALSE' still.
-) default value for the exponent of the time-varying lambda was changed from 'TVlambda.exp= 1.7' to 'TVlambda.exp= 1'.
However, by default 'use.TVlambda= FALSE' still.
-) source code was tidy up
o 'plot_out' : -) Fixed error that appeared when the 'plot_out' function was called from an R script running directly on the OS console
(not within the R editor), with the following error message:
"Error in plot2(x = sim, y = obs, plot.type = "single", main = paste("Daily", :
could not find function "hasArg"
Calls: plot_results -> plot_out -> ggof -> plot2"
o 'plot_results': -) Same as immediately above (because 'plot_results' internally calls 'plot_out')
0.1-56 14-Jun-2012
o 'hydroPSO' : -) much less memory consumption for large values of 'maxit' and/or 'npart' (thanks to P. Smith for reporting it !)
The files 'BestParamPerIter.txt', 'PbestPerIter.txt', 'LocalBestPerIter.txt', 'Velocities.txt' are now written at the end of
each iteration, and not at the end of the main algorithm (so the corresponding local variables were removed)
-) the convergence of the algorithm controlled by 'reltol' is now defined by :
abs(reltol) <= abs(best.pbest[iter] - best.pbest[iter-1]),
instead of:
reltol <= [ gbest(iter) / ( gbest(iter-1) +reltol ) ]
-) 'PbestPerIter.txt' output file: now it has an additional first column labelled "Iter", with the iteration number
-) 'LocalBestPartPerIter.txt' output file: now it has an additional first column labelled "Iter", with the iteration number
-) 'BestParamPerIter.txt' output file: now it has an additional first column labelled "Iter", with the iteration number.
The column with the goodness-of-fit measure ('GoF') was changed from the last position to the second one
(just after the new 'Iter' column)
-) minor change: 'gbest.fit.prior' is now initialized in 'gbest.fit' instead of 'gbest.fit + 10*reltol'
o 'plot_results': -) 'drty.out' is now checked for existence before reading the results.
o Specific version numbers were added for the following suggested packages:
hydroGOF(>=0.3-3), hydroTSM(>= 0.3-0), zoo(>=1.7-2), xts(>=0.8-2)
0.1-55 11-May-2012
o 'hydromod' : -) 'sim' and 'obs' are now checked for time attributes when 'gof.Ini' and/or 'gof.Fin' is/are used
(thanks to M. Leh for reporting !)
o 'hydroPSO' : -) fixed bug originated when 'gof.Ini' and/or 'gof.Fin' was/were not given as parameters for .model.FUN.args.
(thanks to M. Leh for reporting !)
This bug originated the following error during the run of the model with the best parameter set:
"Error in as.Date.default(model.FUN.args[["gof.Ini"]]) : do not know how to convert 'model.FUN.args[["gof.Ini"]]' to class "Date" "
o 'plot_results' : the default value for 'ftype' was changed from "dm" to "o", to avoid problems with time series with a resolution
coarser than daily
o 'plot_out' : the default value for 'ftype' was changed from "dm" to "o", to avoid problems with time series with a resolution
coarser than daily
*) 01-Apr-2012 : 0.1-54
-) 'params2ecdf' : minor bugfix for png generation ('file' argument changed by 'filename')
-) 'plot_GofPerParticle' : minor bugfix for png generation ('file' argument changed by 'filename')
......
DESCRIPTION
+ 2
− 2
View file @ 4fb91b66
Package: hydroPSO
Type: Package
Title: Particle Swarm Optimisation, with focus on Environmental Models
Version: 0.3-0-6
Date: 2013-05-09
Version: 0.3-1
Date: 2013-05-10
Author: Mauricio Zambrano-Bigiarini [aut, cre], Rodrigo Rojas [ctb]
Authors@R: c(person("Mauricio", "Zambrano-Bigiarini", email="mzb.devel@gmail.com", role=c("aut","cre")), person("Rodrigo", "Rojas", email="Rodrigo.RojasMujica@gmail.com", role=c("ctb")) )
Maintainer: Mauricio Zambrano-Bigiarini <mzb.devel@gmail.com>
......
NEWS
+ 11
− 200
View file @ 4fb91b66
This diff is collapsed.
R/lhoat.R
+ 3
− 3
View file @ 4fb91b66
......@@ -108,7 +108,7 @@ lhoat <- function(
if ( trunc(N) != N ) stop( "Invalid argument: 'N' must be integer" )
# Checking that '0 < f < 1'
if ( (f <= 0) | (f >= 1) ) stop( "Invalid argument: 'f' must be in [0,1]" )
if ( (f <= 0) | (f >= 1) ) stop( "Invalid argument: 'f' must be in ]0, 1[" )
# 'hydromod' checkings
......@@ -116,7 +116,7 @@ lhoat <- function(
# Checking that 'param.ranges' really exists
if ( !file.exists( param.ranges ) )
stop( paste("Invalid argument: The file '", param.ranges, "' doesn't exist !", sep="") )
stop( paste("Invalid argument: The file '", param.ranges, "' does not exist !", sep="") )
# checking that 'model.FUN' is a valid function
if ( is.null(model.FUN) ) {
......@@ -466,7 +466,7 @@ lhoat <- function(
if ( fn.name == "hydromod" ) {
model.FUN.args <- modifyList(model.FUN.args, list(param.values=Theta.New))
hydromod.out <- do.call(model.FUN, as.list(model.FUN.args))
} else hydromod.out <- do.call(fn, list(Theta.Zero))
} else hydromod.out <- do.call(fn, list(Theta.New))
##########################################################################
# 7) Extracting simulated values #
......
man/lhoat.Rd
+ 82
− 14
View file @ 4fb91b66
......@@ -111,26 +111,94 @@ Mauricio Zambrano-Bigiarini, \email{mzb.devel@gmail.com}
}
\examples{
# Number of dimensions to be optimised
nparam <- 5
# Running the Latin-Hypercube One-factor-At-a-Time sensitivity analysis.
# Dummy case, because for this test function all the parameters have the
# same sensitivity
##################################################
# Example 1: Linear model (n=3) #
##################################################
# Distributions for the three parameters, are all uniform in the intervals
# [0.5, 1.5], [1.5, 4.5], and [4.5,13.5], respectively.
# 1.1) defining the dimension of the parameter space
nparam <- 3
# 1.2) defining the model
linear <- function(x) x[1] + x[2] + x[3]
# 1.3) Running the LH-OAT sensitivity analysis for the 'linear' test function
# The model is linear and since x[3] has the largest mean value, it should
# be the most important factor.
set.seed(123)
lhoat(
fn=linear,
lower=c(0.5, 1.5, 4.5),
upper=c(1.5, 4.5, 13.5),
control=list(N=100, f=0.1, write2disk=FALSE, verbose=FALSE)
)
\dontrun{
##################################################
# Example 2: non-linear monotonic model (n=2) #
##################################################
# A uniform distribution in the interval [0, 5] was assigned to both parameters.
# This makes the second factor more important than x[1]
# 2.1) defining the dimension of the parameter space
nparam <- 2
# 2.2) defining the model
non.linear <- function(x) x[1] + x[2]^4
# 2.3) Running the LH-OAT sensitivity analysis for the 'non.linear' test function
# The model is linear and since x[3] has the largest mean value, it should
# be the most important factor.
setwd("~")
set.seed(123)
lhoat(
fn=sphere,
lower=rep(-100,nparam),
upper=rep(100,nparam),
control=list(N=10, f=0.1, write2disk=TRUE)
fn=non.linear,
lower=rep(0, nparam),
upper=rep(5, nparam),
control=list(N=1000, f=0.1, write2disk=TRUE, verbose=FALSE)
)
# Plotting dotty plots
read_params(file="LH_OAT-gof.txt", header=TRUE, skip=0, param.cols=2:(nparam+1),
of.col=1, of.name="sphere", ptype="dottyplot")
}
# 2.4) reading ALL the parameter sets used in LH-OAT, and plotting dotty plots
params <- read_params(file="LH_OAT/LH_OAT-gof.txt", header=TRUE, skip=0,
param.cols=2:(nparam+1), of.col=1, of.name="non.linear",
ptype="dottyplot")
##################################################
# Example 3: non-monotonic model (ishigmai, n=3) #
##################################################
# All three input factors have uniform distributions in the range [-pi, pi].
# 3.1) defining the dimension of the parameter space
nparam <- 3
# 3.2) defining the model
ishigami <- function(x, a=7, b=0.1) {
sin(x[1]) + a*(sin(x[2]))^2 + b*(x[3]^4)*sin(x[1])
}
# 3.3) Running the LH-OAT sensitivity analysis for the 'lineal' test function
# first order analytical sensitivity indices for the Ishigami function are:
# S1=0.3138, S2=0.4424, S3=0.0000. Therefore, the first order sensitivity
# indices indicate that factor x[2] is more important than factor x[1].
setwd("~")
set.seed(123)
lhoat(
fn=ishigami,
lower=rep(-pi, nparam),
upper=rep(pi, nparam),
control=list(N=5000, f=0.1, write2disk=TRUE, verbose=FALSE)
)
# 3.4) reading ALL the parameter sets used in LH-OAT, and plotting dotty plots
params <- read_params(file="LH_OAT/LH_OAT-gof.txt", header=TRUE, skip=0,
param.cols=2:(nparam+1), of.col=1, of.name="non.linear",
ptype="dottyplot")
}
}
% Add one or more standard keywords, see file 'KEYWORDS' in the
% R documentation directory.
......
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment