def __init__(self, dat=None):
'''
ACOSSO interface constructor
'''
surrogate.__init__(self, dat)
self.minInputs = 2
self.maxInputs = 100
self.minOutputs = 1
self.maxOutputs = 1
self.ex = None
# Information about the method
self.methodDescription = \
"<html>"\
"<p>ACOSSO:</p>"\
"<p>The Adaptive COmponent Selection and Shrinkage"\
" Operator (ACOSSO) surface approximation was developed"\
" under the Smoothing Spline Analysis of Variance"\
" (SS-ANOVA) modeling framework. As it is a smoothing type"\
" method, ACOSSO works best when the underlying function"\
" is somewhat smooth. For functions which are known to have"\
" sharp changes or peaks, etc., other methods may be more"\
" appropriate Since it implicitly performs variable"\
" selection, ACOSSO can also work well when there are a"\
" large number of input variables. To facilitate the"\
" description of ACOSSO, we first review the univariate"\
" smoothing spline. The ACOSSO procedure also allows for"\
" categorical inputs.</p>"\
"</html>"
# acosso working directory
self.acossoDir = 'acosso'
# add options
self.options.add(
name="Data Filter",
default='all',
dtype=str,
desc="Filter for sample data, from flowsheet data",
validValues = ['All', 'None'])
self.options.add(
name="Use Flowsheet Data",
default='Yes',
dtype=str,
desc="Use data from FOQUS flowsheet or provide csv files",
validValues = ['Yes', 'No'])
self.options.add(
name="Input Data File",
default='acosso'+os.path.sep+'xdat.csv',
dtype=str,
desc="csv file containing data for model inputs")
self.options.add(
name="Output Data File",
default='acosso'+os.path.sep+'ydat.csv',
dtype=str,
desc="csv file containing data for model outputs")
#self.options.add(
# name="RScript path",
# default=\
# 'C:\\Program Files\\R\\R-3.1.2\\bin\\x64\\Rscript.exe',
# dtype=str,
# desc="Full path the RScript executable")
self.options.add(
name="Model File",
default='acosso_fit.rds',
dtype=str,
desc="ACOSSO output R data file")
self.options.add(
name="FOQUS Model (for UQ)",
default="acosso_surrogate_uq.py",
dtype=str,
desc=".py file for UQ analysis")
self.options.add(
name="FOQUS Model (for Flowsheet)",
default="acosso_surrogate_fs.py",
dtype=str,
desc=".py file flowsheet plugin, saved to user_plugins"\
" in the working directory")
self.options.add(
name="CV",
default='bic',
dtype=str,
validValues = ['5cv', 'gcv', 'bic'],
desc="Method used to calculate ACOSSO tuning parameter M")
self.options.add(
name="Order",
default=2,
dtype=int,
validValues = [1, 2],
desc="order of interactions to consider")
def __init__(self, dat=None):
'''
ALAMO interface constructor
'''
surrogate.__init__(self, dat)
self.name = "ALAMO"
self.ex = None
#still working on hanging indent for references
self.methodDescription = \
("<html>\n<head>"
".hangingindent {\n"
" margin-left: 22px ;\n"
" text-indent: -22px ;\n"
"}\n"
"</head>\n"
"<b>Automatic Learning of Algebraic Models for Optimization"
" (ALAMO)</b>"
"<p class=\"hangingindent\">Cozad, A., N. V. Sahinidis "
"and D. C. Miller, Learning surrogate models "
"for simulation‐based optimization, "
"AIChE Journal, 60, p. 2211–2227, 2014.</p></html>")
self.alamoDir = 'alamo'
self.inputCols = [
('XFACTOR', float, 1.0),
('EXTRAPXMIN', float, 0.0),
('EXTRAPXMAX', float, 1.0)]
self.outputCols = [
('MAXTERMS', int, -1),
('IGNORE', int, 0),
('TOLMEANERROR', float, 0.001),
('TOLRELMETRIC', float, 0.0001),
('ZMIN', float, 0.0),
('ZMAX', float, 1.0),
('CUSTOMCON', list, [])]
self.updateVarCols()
#include a Section called DATA Settings
#Check the ALAMOSettings_v2.xlsx with hints and new labels
self.options.add(
name="Initial Data Filter",
section="DATA Settings",
default='All',
dtype=str,
desc="Filter to be applied to the initial data set.",
hint="Data filters help the user to generate models based"
"on specific data for each variable.",
validValues = ['All', 'None'])
self.options.add(
name="Validation Data Filter",
section="DATA Settings",
default='None',
dtype=str,
desc="Data set used to compute model errors at the validation phase.",
hint="The number of data points in a preexisting validation data set"
"can be specified by the user",
validValues = ['All', 'None'])
#Maxtime
self.options.add(
name="MAXTIME",
section="DATA Settings",
default=2000,
desc="Maximum total execution time in seconds.",
hint="This time includes all steps of the algorithm, including "
"time to read problem, preprocess data, solve "
"optimization subproblems, and print results."
)
self.options.add(
name="MINPOINTS",
section="DATA Settings",
default=0,
dtype=int,
desc="Convergence is assessed only if the simulator is able to compute the "
"output variables for at least MINPOINTS out the data points requested by ALAMO.",
hint="A reduced number of MINPOINTS may reduce the computational time to get a model,"
"but also reducess the accuracy of the model. MINPOINTS must be a positive integer.")
self.options.add(
name="NSAMPLE",
section="Data Settings",
default=0,
dtype=int,
desc="Number of data points to be generated by sampling before any model is built.",
hint="These points will be used for model builing along with the NDATA points specified by the user. NSAMPLE must be a nonnegative integer.")
self.options.add(
name="MAXSIM",
section="Data Settings",
default=5,
dtype=int,
desc=" Maximum number of successive simulator failures allowed before quit",
hint="MAXSIM must be a nonnegative integer")
#Sampler
self.samplers={
"None":0,
"Random":1,
"SNOBFIT":2}
self.options.add(
name="SAMPLER",
section="DATA Settings",
default="None",
desc="Adaptive sampling method to be used. If adaptive sampling is used, also set MAXITER below.",
hint="Adaptive sampling method to be used by ALAMO when more data points are needed by the model"
"If random is used a simulator must be provided by the user. If SNOBFIT is used a simulator must be"
" by the user and MATLAB must be istalled.",
validValues = sorted(list(self.samplers.keys()),
key=lambda k: self.samplers[k]))
self.options.add(
name="MAXITER",
section="DATA Settings",
default=1,
desc="Maximum number of ALAMO iterations, 1 = no adaptive sampling, 0 = no limit",
hint="The maximum number of ALMAO iterations for adaptive sampling, 1 = no adaptive sampling, 0 = no limit")
self.options.add(
name="PRESET",
section="DATA Settings",
default=-111111,
desc="Value to be used if the simulation fails.",
hint="This value must be carefully chosen to be an otherwise not realizable value for the output variables")
#Add the Section: Model Settings
self.options.add(
name="MONOMIALPOWER",
section="Model Settings",
default=[1,2,3],
dtype=list,
desc="Vector of monomial powers considered in basis "
"functions. Use an empty vector for none.",
hint="Exponential terms allowed in the algebraic model as basis functions.")
self.options.add(
name="MULTI2POWER",
section="Model Settings",
default=[1],
desc="Vector of powers to be considered for pairwise "
"combinations in basis functions. Empty vector "
"for none.",
hint="Pairwise combination of powers allowed in the algebraic model.")
self.options.add(
name="MULTI3POWER",
section="Model Settings",
default=[],
desc="Vector of three variables combinations of powers to be consiedered as basis functions.",
hint= "Empty vector for none [].")
self.options.add(
name="RATIOPOWER",
section="Model Settings",
default=[],
desc="Vector of ratio combinations of powers to be considered in the basis functions.",
hint="Ratio combinations of powers are empty as default.")
self.options.add(
name="EXPFCNS",
section="Model Settings",
default=True,
desc="Use or not of exponential functions as basis functions in the model.")
self.options.add(
name="LOGFCNS",
section="Model Settings",
default=True,
desc="Logarithimic functions are considered as basis "
"functions if true; otherwise, they are not considered.")
self.options.add(
name="SINFCNS",
section="Model Settings",
default=False,
desc="Sine functions are considered as basis functions if "
"true; otherwise, they are not considered.")
self.options.add(
name="COSFCNS",
section="Model Settings",
default=False,
desc="Cosine functions are considered as basis functions if "
"true; otherwise, they are not considered.")
self.options.add(
name="CUSTOMBAS",
section="Model Settings",
default = [],
dtype = list,
desc = "A list of user-supplied custom basis functions can "
"be provided by the user. ",
hint="The parser is not case sensitive and allows for any "
"Fortran functional expression in terms of the XLABELS. "
"In addition, the symbol ^ may be used to denote power "
"operations instead of **.")
self.modelers = {
"BIC":1,
"Mallow's Cp":2,
"AICc":3,
"HQC":4,
"MSE": 5,
"Convex Penalty": 6,
"RIC": 7}
self.options.add(
name="MODELER",
section="Model Settings",
default="BIC",
dtype=str,
desc="Fitness metric to be used for model building.",
hint="Possible values are Bayesian information criterion, "
"Mallow's Cp, the corrected Akaike's information "
"criterion, the Hannan-Quinn information criterion, "
"mean square error, and a convex penalty consisting of "
"the sum of square errors and a term penalizing model "
"size.",
validValues = sorted(list(self.modelers.keys()),
key=lambda k: self.modelers[k]))
self.options.add(
name="CONVPEN",
section="Model Settings",
default=10.0,
desc="Convex penalty term used if Convex penalty is selected.",
hint="When MODELER is set to \"Convex Penalty,\" a convex "
"penalty consisting of the sum of square errors and a "
"term penalizing model size is used for model building. "
"In this case, the size of the model is weighted "
"by CONPEN.")
self.options.add(
name="REGULARIZER",
section="Model Settings",
default=False,
desc="Regularization method is used to reduce the number of "
"potential basis functions before optimization.",
hint="If true, "
"regularization with the LASSO is used.")
self.options.add(
name="SCALEZ",
section="Model Settings",
default=False,
desc="If used, the variables are scaled prior to the optimization problem is solved.",
hint="The problem is solved using a mathematical programming solver. Usually, scaling"
" the variables may help the optimization.")
#Add the Section: SOLVER Settings
self.options.add(
name="GAMS",
section="Solver Settings",
default="gams",
desc="GAMS path is needed. GAMS is the software used to solve the optimization problems.",
hint="The executable path is expected here or the user must declare GAMS.exe in the environment path.")
self.options.add(
name="GAMSSOLVER",
section="Solver Settings",
default="BARON",
desc="Name of preferred GAMS solver for solving ALAMO's mixed-integer quadratic subproblems.",
hint=" Special features "
"have been implemented in ALAMO and BARON that "
"make BARON the preferred selection for this option. "
"However, any mixed-integer quadratic programming solver "
"available under GAMS can be used. ")
self.options.add(
name="SOLVEMIP",
section="Solver Settings",
default=False,
dtype = bool,
desc="GAMS will be used to solve ALAMO's MIPs/MIQPs if checked",
hint="")
self.funform = {
"Fortran":1,
"GAMS":2,
"BARON":3,
"C":4}
self.options.add(
name="FUNFORM",
section="Solver Settings",
default="Fortran",
desc="Format for printing basis functions and models found by ALAMO. Fortran must be selected to generate FOQUS UQ and flowsheet models.",
hint="",
validValues = sorted(list(self.funform.keys()), key=lambda k: self.funform[k]))
self.options.add(
name="MIPOPTCA",
section="Solver Settings",
default=0.05,
desc="Absolute convergence tolerance for mixed-integer "
"optimization problems.",
hint="This must be a nonnegative "
"scalar.")
self.options.add(
name="MIPOPTCR",
section="Solver Settings",
default=0.0001,
desc="Relative convergence tolerance for mixed-integer "
"optimization problems.",
hint="This must be a nonnegative "
"scalar.")
self.options.add(
name="LINEARERROR",
section="Solver Settings",
default=False,
desc="If true, a linear objective is used when solving "
"the mixed-integer optimization problems.",
hint="If used the Quadratic objective function is replaced by a linear objective function.")
self.options.add(
name="CONREG",
section="Solver Settings",
default=False,
desc="Specify whether a constraint regression is used or not.",
hint="If CONREG is true, bounds on output variables are "
"enforced.")
self.options.add(
name="CRNCUSTOM",
section="Solver Settings",
default=False,
desc="If true, constraints need to be entered in variables tab.")
self.options.add(
name="CRNINITIAL",
section="Solver Settings",
default=0,
dtype=int,
desc="Number of random bounding points at which constraints "
"are sampled initially.",
hint="CRNINITIAL must be a "
"nonnegative integer.")
self.options.add(
name="CRNMAXITER",
section="Solver Settings",
default=10,
dtype=int,
desc="Maximum allowed constrained regressions iterations.",
hint="Constraints are enforced on additional points during "
"each iteration. CRMAXITER must be a positive integer.")
self.options.add(
name="CRNVIOL",
section="Solver Settings",
default=100,
dtype=int,
desc="Number of bounding points added per round per bound "
"in each iteration.",
hint="CRNVIOL must be a positive integer.")
self.options.add(
name="CRNTRIALS",
section="Solver Settings",
default=100,
dtype=int,
desc="Number of random trial bounding points per round of "
"constrained regression.",
hint="CRNTRIALS must be a positive "
"integer.")
self.options.add(
name="CRTOL",
section="Solver Settings",
default=1e-3,
dtype=int,
desc="Tolerance within which custom constraints must be satisfied. Real greater than 1e-5 is expected",
hint="Bound and custom constraints will be satisfied within an absolute tolerance equal to CRTOL")
#Add the Section: Advanced Settings
self.options.add(
name="Input File",
section="Advanced Settings",
default="alamo.alm",
dtype=str,
desc="File name for ALAMO input file.")
self.options.add(
name="FOQUS Model (for UQ)",
section="Advanced Settings",
default="alamo_surrogate_uq.py",
dtype=str,
desc=".py file for UQ analysis.")
self.options.add(
name="FOQUS Model (for Flowsheet)",
section="Advanced Settings",
default="alamo_surrogate_fs.py",
dtype=str,
desc=".py file flowsheet plugin, saved to user_plugins"\
" in the working directory.")
self.inputVarButtons = (
('Set XFACTOR from range', self.autoXFact),)
self.outputVarButtons = (
('Set ZMIN and ZMAX from current data filter', self.autoZMin),)
def __init__(self, dat=None):
'''
BSS-ANOVA interface constructor
'''
surrogate.__init__(self, dat)
self.ex = None
# Information about the method
self.methodDescription = \
("<html>"
"<p>BSS-ANOVA:</p>"
"<p>The Bayesian Smoothing Spline ANOVA (BSS-ANOVA) is"
" essentially a Bayesian version of ACOSSO. It is"
" Gaussian Process (GP) model with a non-conventional"
" covariance function that borrows its form from SS-ANOVA."
" It tackles the high dimensionality (of inputs) on two"
" fronts: (i) variable selection to eliminate"
" uninformative variables from the model and (ii)"
" restricting the level of interactions involved among"
" the variables in the model. This is done through a"
" fully Bayesian approach which can also allow for"
" categorical input variables with relative ease."
" Since it is closely related to ACOSSO, it generally"
" works well in similar settings as ACOSSO.</p>"
"</html>")
# bssanova working directory
self.bssanovaDir = 'bssanova'
# add options
self.options.add(name="Data Filter",
default='all',
dtype=str,
desc="Filter for sample data, from flowsheet data",
validValues=['All', 'None'])
self.options.add(
name="Use Flowsheet Data",
default='Yes',
dtype=str,
desc="Use data from FOQUS flowsheet or provide csv files",
validValues=['Yes', 'No'])
self.options.add(name="Input Data File",
default='bssanova' + os.path.sep + 'xdat.csv',
dtype=str,
desc="csv file containing data for model inputs")
self.options.add(name="Output Data File",
default='bssanova' + os.path.sep + 'ydat.csv',
dtype=str,
desc="csv file containing data for model outputs")
#self.options.add(
# name="RScript path",
# default=\
# 'C:\\Program Files\\R\\R-3.1.2\\bin\\x64\\Rscript.exe',
# dtype=str,
# desc="Full path the RScript executable")
self.options.add(name="Model File",
default='bssanova_fit.rds',
dtype=str,
desc="BSS-ANOVA output R data file")
self.options.add(name="FOQUS Model (for UQ)",
default="bssanova_surrogate_uq.py",
dtype=str,
desc=".py file for UQ analysis")
self.options.add(
name="FOQUS Model (for Flowsheet)",
default="bssanova_surrogate_fs.py",
dtype=str,
desc=".py file flowsheet plugin, saved to user_plugins"\
" in the working directory")
self.options.add(name="Burn In",
default=0,
dtype=int,
desc="Number of data point to ignore from begining")
self.options.add(name="MCMC Iterations",
default=1000,
dtype=int,
desc="Total number of MCMC iterations")
self.options.add(name="Record Interval",
default=1,
dtype=int,
desc="How often to record a sample")
self.options.add(
name="nterms",
default=25,
dtype=int,
desc="number of eigenvalues to use to approximate BSSA-NOVA"\
"covariance")
self.options.add(name="order",
default=2,
dtype=int,
validValues=[1, 2],
desc="order of interactions to consider covariance")
self.options.add(
name="priorprob",
default=0.5,
dtype=float,
desc="prior probability that a given component is"\
" uninformative")