"""
Case generators for Design-of-Experiments Driver.
"""
import csv
import os.path
import re
from collections import OrderedDict
import numpy as np
_LEVELS = 2 # default number of levels for pyDOE generators
[docs]
class DOEGenerator(object):
"""
Base class for a callable object that generates cases for a DOEDriver.
"""
def __call__(self, design_vars, model=None):
"""
Generate case.
Parameters
----------
design_vars : OrderedDict
Dictionary of design variables for which to generate values.
model : Group
The model containing the design variables (used by some subclasses).
Returns
-------
list
list of name, value tuples for the design variables.
"""
return []
[docs]
class ListGenerator(DOEGenerator):
"""
DOE case generator that reads cases from a provided list of DOE cases.
This DOE case generator will accept an existing data set in the form of
a list of DOE cases, each of which consists of a collection of name/value
pairs specifying values for design variables.
Parameters
----------
data : list
List of collections of name, value pairs for the design variables.
Attributes
----------
_data : list
List of collections of name, value pairs for the design variables.
"""
[docs]
def __init__(self, data=[]):
"""
Initialize the ListGenerator.
"""
super().__init__()
if not isinstance(data, list):
msg = "Invalid DOE case data, expected a list but got a {}."
raise RuntimeError(msg.format(data.__class__.__name__))
self._data = data
def __call__(self, design_vars, model=None):
"""
Generate case.
Parameters
----------
design_vars : OrderedDict
Dictionary of design variables for which to generate values.
model : Group
The model containing the design variables.
Yields
------
list
list of name, value tuples for the design variables.
"""
for case in self._data:
if not isinstance(case, list):
msg = "Invalid DOE case found, expecting a list of name/value pairs:\n{}"
raise RuntimeError(msg.format(case))
name_map = {}
for tup in case:
if not isinstance(tup, (tuple, list)) or len(tup) != 2:
msg = "Invalid DOE case found, expecting a list of name/value pairs:\n{}"
raise RuntimeError(msg.format(case))
name = tup[0]
if name in design_vars:
name_map[name] = name
elif model:
abs_name = model._resolver.any2abs(name, 'output')
if abs_name in design_vars:
name_map[name] = abs_name
# any names not found in name_map are invalid design vars
invalid_desvars = [name for name, _ in case if name not in name_map]
if invalid_desvars:
if len(invalid_desvars) > 1:
msg = "Invalid DOE case found, {} are not valid design variables:\n{}"
raise RuntimeError(msg.format(invalid_desvars, case))
else:
msg = "Invalid DOE case found, '{}' is not a valid design variable:\n{}"
raise RuntimeError(msg.format(invalid_desvars[0], case))
yield [(name_map[name], val) for name, val in case]
[docs]
class CSVGenerator(DOEGenerator):
"""
DOE case generator that reads cases from a CSV file.
This DOE case generator will accept an existing data set in the form of
a CSV file containing DOE cases. The CSV file should have one column per
design variable and the header row should have the names of the design
variables.
Parameters
----------
filename : str
The name of the file from which to read cases.
Attributes
----------
_filename : str
the name of the file from which to read cases
"""
[docs]
def __init__(self, filename):
"""
Initialize the CSVGenerator.
"""
super().__init__()
if not isinstance(filename, str):
raise RuntimeError("'{}' is not a valid file name.".format(filename))
if not os.path.isfile(filename):
raise RuntimeError("File not found: {}".format(filename))
self._filename = filename
def __call__(self, design_vars, model=None):
"""
Generate case.
Parameters
----------
design_vars : OrderedDict
Dictionary of design variables for which to generate values.
model : Group
The model containing the design variables.
Yields
------
list
list of name, value tuples for the design variables.
"""
name_map = {}
with open(self._filename, 'r') as f:
# map header names to absolute names if necessary
names = re.sub(' ', '', f.readline()).strip().split(',')
for name in names:
if name in design_vars:
name_map[name] = name
elif model:
abs_name = model._resolver.any2abs(name, 'output')
if abs_name in design_vars:
name_map[name] = abs_name
# any names not found in name_map are invalid design vars
invalid_desvars = [name for name in names if name not in name_map]
if invalid_desvars:
if len(invalid_desvars) > 1:
msg = "Invalid DOE case file, {} are not valid design variables."
raise RuntimeError(msg.format(invalid_desvars))
else:
msg = "Invalid DOE case file, '{}' is not a valid design variable."
raise RuntimeError(msg.format(invalid_desvars[0]))
# read cases from file, parse values into numpy arrays
with open(self._filename, 'r') as f:
reader = csv.DictReader(f)
for row in reader:
case = [(name_map[name.strip()],
np.fromstring(re.sub(r'[\[\]]', '', row[name]), sep=' '))
for name in reader.fieldnames]
yield case
class _pyDOE_Generator(DOEGenerator):
"""
Base class for DOE case generators implementing methods from pyDOE3.
Parameters
----------
levels : int or dict, optional
The number of evenly spaced levels between each design variable
lower and upper bound. Dictionary input is supported by Full Factorial or
Generalized Subset Design.
Defaults to 2.
Attributes
----------
_levels : int or dict(str, int)
The number of evenly spaced levels between each design variable
lower and upper bound. Dictionary input is supported by Full Factorial or
Generalized Subset Design.
"""
def __init__(self, levels=_LEVELS):
"""
Initialize the _pyDOE_Generator.
"""
super().__init__()
self._levels = levels
self._sizes = None
def _get_dv_levels(self, name):
"""
Get the number of levels of a design variable.
If the name is not given, it looks for a "default" key in the dictionary. If this is also
missing, it uses the default number of levels (2).
Parameters
----------
name : str
Design variable name
Returns
-------
int
"""
levels = self._levels
if isinstance(levels, int):
return levels
else:
return levels.get(name, levels.get("default", _LEVELS))
def _get_all_levels(self):
"""Return the levels of all factors."""
sizes = self._sizes
if isinstance(self._levels, int): # All have the same number of levels
return [self._levels] * sum(self._sizes.values())
elif isinstance(self._levels, dict): # Different DVs have different number of levels
return sum([v * [self._get_dv_levels(k)] for k, v in sizes.items()], [])
else:
raise ValueError(f"Levels should be an int or dictionary, not '{type(self._levels)}'")
def __call__(self, design_vars, model=None):
"""
Generate case.
Parameters
----------
design_vars : OrderedDict
Dictionary of design variables for which to generate values.
model : Group
The model containing the design variables (not used).
Yields
------
list
list of name, value tuples for the design variables.
"""
self._sizes = OrderedDict([(name, _get_size(meta))
for name, meta in design_vars.items()])
size = sum(self._sizes.values())
doe = self._generate_design(size).astype('int')
# Maximum number of levels, or the default if the maximum is smaller than the default.
# This is to ensure that the array will be big enough even if some keys are missing
# from levels (defaulted).
levels_max = self._levels if isinstance(self._levels, int) else \
max(max(self._levels.values()), _LEVELS)
# Generate values for each level for each design variable
# over the range of that variable's lower to upper bound
# rows = vars (# rows/var = var size), cols = levels
values = np.empty((size, levels_max)) # Initialize array for the largest number of levels
values[:] = np.nan # and fill with NaNs.
row = 0
for name, meta in design_vars.items():
size = _get_size(meta)
for k in range(size):
lower = meta['lower']
if isinstance(lower, np.ndarray):
lower = lower[k]
upper = meta['upper']
if isinstance(upper, np.ndarray):
upper = upper[k]
levels = self._get_dv_levels(name)
values[row, 0:levels] = np.linspace(lower, upper, num=levels)
row += 1
# yield values for doe generated indices
for idxs in doe:
retval = []
row = 0
for name, meta in design_vars.items():
size_i = _get_size(meta)
val = np.empty(size_i)
for k in range(size_i):
idx = idxs[row + k]
val[k] = values[row + k][idx]
retval.append((name, val))
row += size_i
yield retval
def _generate_design(self, size):
"""
Generate DOE design.
Parameters
----------
size : int
The total size (sum of sizes) of all factors for the design.
Returns
-------
ndarray
The design matrix as a size x levels array of indices.
"""
pass
[docs]
class FullFactorialGenerator(_pyDOE_Generator):
"""
DOE case generator implementing the Full Factorial method.
Parameters
----------
levels : int or dict, optional
The number of evenly spaced levels between each design variable
lower and upper bound. Dictionary input is supported by Full Factorial or
Generalized Subset Design.
Defaults to 2.
Attributes
----------
_fullfact : function
The pyDOE3 full factorial function, lazily imported.
"""
[docs]
def __init__(self, levels=_LEVELS):
"""Initialize the FullFactorialGenerator."""
super().__init__(levels=levels)
try:
from pyDOE3 import fullfact
self._fullfact = fullfact
except ImportError:
raise RuntimeError(f"{self.__class__.__name__} requires the 'pyDOE3' package, "
"which can be installed with one of the following commands:\n"
" pip install openmdao[doe]\n"
" pip install pyDOE3")
def _generate_design(self, size):
"""
Generate a full factorial DOE design.
Parameters
----------
size : int
The number of factors for the design.
Returns
-------
ndarray
The design matrix as a size x levels array of indices.
"""
return self._fullfact(self._get_all_levels())
[docs]
class GeneralizedSubsetGenerator(_pyDOE_Generator):
"""
DOE case generator implementing the General Subset Design Factorial method.
Parameters
----------
levels : int or dict
The number of evenly spaced levels between each design variable
lower and upper bound. Defaults to 2.
reduction : int
Reduction factor (bigger than 1). Larger `reduction` means fewer
experiments in the design and more possible complementary designs.
n : int, optional
Number of complementary GSD-designs. The complementary
designs are balanced analogous to fold-over in two-level fractional
factorial designs.
Defaults to 1.
Attributes
----------
_reduction : int
Reduction factor (bigger than 1). Larger `reduction` means fewer
experiments in the design and more possible complementary designs.
_n : int, optional
Number of complementary GSD-designs. The complementary
designs are balanced analogous to fold-over in two-level fractional
factorial designs.
Defaults to 1.
_gsd : function
The pyDOE3 General Subset Design function, lazily imported.
"""
[docs]
def __init__(self, levels, reduction, n=1):
"""
Initialize the GeneralizedSubsetGenerator.
"""
super().__init__(levels=levels)
self._reduction = reduction
self._n = n
try:
from pyDOE3 import gsd
self._gsd = gsd
except ImportError:
raise RuntimeError(f"{self.__class__.__name__} requires the 'pyDOE3' package, "
"which can be installed with one of the following commands:\n"
" pip install openmdao[doe]\n"
" pip install pyDOE3")
def _generate_design(self, size):
"""
Generate a general subset DOE design.
Parameters
----------
size : int
The number of factors for the design.
Returns
-------
ndarray
The design matrix as a size x levels array of indices.
"""
return self._gsd(levels=self._get_all_levels(), reduction=self._reduction, n=self._n)
[docs]
class PlackettBurmanGenerator(_pyDOE_Generator):
"""
DOE case generator implementing the Plackett-Burman method.
Attributes
----------
_pbdesign : function
The pyDOE3 Plackett-Burman function, lazily imported.
"""
[docs]
def __init__(self):
"""
Initialize the PlackettBurmanGenerator.
"""
super().__init__(levels=2)
try:
from pyDOE3 import pbdesign
self._pbdesign = pbdesign
except ImportError:
raise RuntimeError(f"{self.__class__.__name__} requires the 'pyDOE3' package, "
"which can be installed with one of the following commands:\n"
" pip install openmdao[doe]\n"
" pip install pyDOE3")
def _generate_design(self, size):
"""
Generate a Plackett-Burman DOE design.
Parameters
----------
size : int
The number of factors for the design.
Returns
-------
ndarray
The design matrix as a size x levels array of indices.
"""
doe = self._pbdesign(size)
doe[doe < 0] = 0 # replace -1 with zero
return doe
[docs]
class BoxBehnkenGenerator(_pyDOE_Generator):
"""
DOE case generator implementing the Box-Behnken method.
Parameters
----------
center : int, optional
The number of center points to include (default = None).
Attributes
----------
_center : int
The number of center points to include.
_bbdesign : function
The pyDOE3 BoxBehnken function, lazily imported.
"""
[docs]
def __init__(self, center=None):
"""
Initialize the BoxBehnkenGenerator.
"""
super().__init__(levels=3)
self._center = center
try:
from pyDOE3 import bbdesign
self._bbdesign = bbdesign
except ImportError:
raise RuntimeError(f"{self.__class__.__name__} requires the 'pyDOE3' package, "
"which can be installed with one of the following commands:\n"
" pip install openmdao[doe]\n"
" pip install pyDOE3")
def _generate_design(self, size):
"""
Generate a Box-Behnken DOE design.
Parameters
----------
size : int
The number of factors for the design.
Returns
-------
ndarray
The design matrix as a size x levels array of indices.
"""
if size < 3:
raise RuntimeError("Total size of design variables is %d,"
"but must be at least 3 when using %s. " %
(size, self.__class__.__name__))
doe = self._bbdesign(size, center=self._center)
return doe + 1 # replace [-1, 0, 1] with [0, 1, 2]
[docs]
class LatinHypercubeGenerator(DOEGenerator):
"""
DOE case generator implementing Latin hypercube method via pyDOE3.
Parameters
----------
samples : int, optional
The number of samples to generate for each factor (Defaults to n).
criterion : str, optional
Allowable values are "center" or "c", "maximin" or "m",
"centermaximin" or "cm", and "correlation" or "corr". If no value
given, the design is simply randomized.
iterations : int, optional
The number of iterations in the maximin and correlations algorithms
(Defaults to 5).
seed : int, optional
Random seed to use if design is randomized. Defaults to None.
Attributes
----------
_samples : int
The number of evenly spaced levels between each design variable
lower and upper bound.
_criterion : str
the pyDOE criterion to use.
_iterations : int
The number of iterations to use for maximin and correlations algorithms.
_seed : int or None
Random seed.
_lhs : function
The pyDOE3 latin hypercube sampling function, lazily imported.
"""
# supported pyDOE criterion names.
_supported_criterion = [
"center", "c",
"maximin", "m",
"centermaximin", "cm",
"correlation", "corr",
None
]
[docs]
def __init__(self, samples=None, criterion=None, iterations=5, seed=None):
"""
Initialize the LatinHypercubeGenerator.
See : https://pythonhosted.org/pyDOE/randomized.html
"""
super().__init__()
try:
from pyDOE3 import lhs
self._lhs = lhs
except ImportError:
raise RuntimeError(f"{self.__class__.__name__} requires the 'pyDOE3' package, "
"which can be installed with one of the following commands:\n"
" pip install openmdao[doe]\n"
" pip install pyDOE3")
if criterion not in self._supported_criterion:
raise ValueError("Invalid criterion '%s' specified for %s. "
"Must be one of %s." %
(criterion, self.__class__.__name__,
self._supported_criterion))
self._samples = samples
self._criterion = criterion
self._iterations = iterations
self._seed = seed
def __call__(self, design_vars, model=None):
"""
Generate case.
Parameters
----------
design_vars : OrderedDict
Dictionary of design variables for which to generate values.
model : Group
The model containing the design variables (not used).
Yields
------
list
list of name, value tuples for the design variables.
"""
if self._seed is not None:
np.random.seed(self._seed)
size = sum([meta['size'] for meta in design_vars.values()])
if self._samples is None:
self._samples = size
# generate design
doe = self._lhs(size, samples=self._samples,
criterion=self._criterion,
iterations=self._iterations,
random_state=self._seed)
# yield desvar values for doe samples
for row in doe:
retval = []
col = 0
for name, meta in design_vars.items():
size = meta['size']
sample = row[col:col + size]
lower = meta['lower']
if not isinstance(lower, np.ndarray):
lower = lower * np.ones(size)
upper = meta['upper']
if not isinstance(upper, np.ndarray):
upper = upper * np.ones(size)
val = lower + sample * (upper - lower)
retval.append((name, val))
col += size
yield retval
def _get_size(dct):
# Returns global size of the variable if it is distributed, size otherwise.
return dct['global_size'] if dct['distributed'] else dct['size']
