# Copyright 2016 Quan Pan
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
# Author: Quan Pan <quanpan302@hotmail.com>
# License: Apache License, Version 2.0
# Create: 2016-12-02
"""
Base classes for all estimators.
Class definition for Individual, the base class for all surrogate models.
"""
import sys
from collections import Sequence
from operator import mul, truediv
[docs]class Individual(object):
"""A Individual
:ref:`Fitness`
:param estimator: physical based model
"""
[docs] def __init__(self, estimator, variable=None, constraint=None, weights=()):
"""__init__
:param estimator:
:param variable:
:param constraint:
:param weights:
:return:
"""
# needed
self.estimator = estimator
if variable is None:
self.variable = {}
else:
self.variable = variable
# if objective is None:
# self.objective = {}
# else:
# self.objective = objective
if constraint is None:
self.constraint = {}
else:
self.constraint = constraint
# not yet decided
self.fitness = Fitness(estimator(variable), weights)
# self.rank = None
# self.distance = None
# self.strategy = set()
# self.solution = set()
# self.feature = None
# self.dominate = None
# deap
# NDIM = 5
# MU = 12
# pop = toolbox.population(n=MU)
#
# # Evaluate the individuals with an invalid fitness
# invalid_ind = [ind for ind in pop if not ind.fitness.valid]
# fitnesses = toolbox.map(toolbox.evaluate, invalid_ind)
#
# for ind, fit in zip(invalid_ind, fitnesses):
# ind.fitness.values = fit
#
# # debug
# 10 = {individual} array('d', [0.87, 0.74, 0.16, 0.67, 0.42])
# .fitness
# .crowding_dist = {float} inf
# .valid = {bool} True
# .values = {tuple} (0.99, 8.46)
# .weights = {tuple} (-1.0, -1.0)
# .wvalues = {tuple} (-0.99, -8.46)
# .itemsize = {int} 8
# .typecode = {str} 'd'
[docs] def getVar(self, i):
"""The fitness is a measure of quality of a solution. If *values* are
provided as a tuple, the fitness is initalized using those values,
otherwise it is empty (or invalid).
:param i: index of variable
.. code-block:: python
if not (isinstance(i, int) and i >= 0):
raise ValueError("Variable index must be an integer >= 0 .")
.. note::
Note
"""
if not (isinstance(i, int) and i >= 0):
raise ValueError("Variable index must be an integer >= 0 .")
return self._variables[i]
# def __len__(self):
# return len(self.variable)
# def __repr__(self):
# return repr((self.variable, self.objective, self.constraint, self.fitness))
[docs]class Fitness(object):
"""The fitness is a measure of quality of a solution. If *values* are
provided as a tuple, the fitness is initalized using those values,
otherwise it is empty (or invalid).
:param values: The initial values of the fitness as a tuple, optional.
Fitnesses may be compared using the ``>``, ``<``, ``>=``, ``<=``, ``==``,
``!=``. The comparison of those operators is made lexicographically.
Maximization and minimization are taken care off by a multiplication
between the :attr:`weights` and the fitness :attr:`values`. The comparison
can be made between fitnesses of different size, if the fitnesses are
equal until the extra elements, the longer fitness will be superior to the
shorter.
Different types of fitnesses.
.. note::
When comparing fitness values that are **minimized**, ``a > b`` will
return :data:`True` if *a* is **smaller** than *b*.
"""
weights = None
"""
The weights are used in the fitness comparison. They are shared among
all fitnesses of the same type. When subclassing :class:`Fitness`, the
weights must be defined as a tuple where each element is associated to an
objective.
A negative weight element corresponds to: (-1.0, -1.0)
``the minimization of the associated objective.``
A positive weight element corresponds to: (1.0, 1.0)
``the maximization of the associated objective.``
.. note::
If weights is not defined during subclassing, the following error will
occur at instantiation of a subclass fitness object:
``TypeError: Can't instantiate abstract <class Fitness[...]> with
abstract attribute weights.``
"""
wvalues = ()
"""Contains the weighted values of the fitness, the multiplication with the
weights is made when the values are set via the property :attr:`values`.
Multiplication is made on setting of the values for efficiency.
Generally it is unnecessary to manipulate wvalues as it is an internal
attribute of the fitness used in the comparison operators.
"""
[docs] def __init__(self, values=(), weights=()):
if weights is None:
raise TypeError("Can't instantiate abstract %r with abstract "
"attribute weights." % (self.__class__))
self.weights = weights
if not isinstance(self.weights, Sequence):
raise TypeError("Attribute weights of %r must be a sequence."
% self.__class__)
if len(values) > 0:
self.values = values
def getValues(self):
return tuple(map(truediv, self.wvalues, self.weights))
def setValues(self, values):
try:
self.wvalues = tuple(map(mul, values, self.weights))
except TypeError:
_, _, traceback = sys.exc_info()
raise TypeError, ("Both weights and assigned values must be a "
"sequence of numbers when assigning to values of "
"%r. Currently assigning value(s) %r of %r to a fitness with "
"weights %s."
% (self.__class__, values, type(values), self.weights)), traceback
def delValues(self):
self.wvalues = ()
values = property(getValues, setValues, delValues,
("Fitness values. Use directly ``individual.fitness.values = values`` "
"in order to set the fitness and ``del individual.fitness.values`` "
"in order to clear (invalidate) the fitness. The (unweighted) fitness "
"can be directly accessed via ``individual.fitness.values``."))
[docs] def dominates(self, other, obj=slice(None)):
"""Return true if each objective of *self* is not strictly worse than
the corresponding objective of *other* and at least one objective is
strictly better.
:param obj: Slice indicating on which objectives the domination is
tested. The default value is `slice(None)`, representing
every objectives.
"""
not_equal = False
for self_wvalue, other_wvalue in zip(self.wvalues[obj], other.wvalues[obj]):
if self_wvalue > other_wvalue:
not_equal = True
elif self_wvalue < other_wvalue:
return False
return not_equal
@property
def valid(self):
"""Assess if a fitness is valid or not."""
return len(self.wvalues) != 0
[docs] def __hash__(self):
"""hash
:return:
"""
return hash(self.wvalues)
[docs] def __gt__(self, other):
return not self.__le__(other)
[docs] def __ge__(self, other):
return not self.__lt__(other)
[docs] def __le__(self, other):
return self.wvalues <= other.wvalues
[docs] def __lt__(self, other):
return self.wvalues < other.wvalues
[docs] def __eq__(self, other):
return self.wvalues == other.wvalues
[docs] def __ne__(self, other):
return not self.__eq__(other)
# TODO 20161212 pass memeory address instead of values
# offspring = [deepcopy(ind) for ind in offspring]
# def __deepcopy__(self, memo):
# """Replace the basic deepcopy function with a faster one.
#
# It assumes that the elements in the :attr:`values` tuple are
# immutable and the fitness does not contain any other object
# than :attr:`values` and :attr:`weights`.
# """
# copy_ = self.__class__()
# copy_.wvalues = self.wvalues
# return copy_
[docs] def __str__(self):
"""Return the values of the Fitness object."""
return str(self.values if self.valid else tuple())
[docs] def __repr__(self):
"""Return the Python code to build a copy of the object."""
return "%s.%s(%r)" % (self.__module__, self.__class__.__name__,
self.values if self.valid else tuple())
[docs]class SurrogateModel(object):
"""A class for surrogate models.
:ref:`Individual`
"""
def __init__(self):
self.trained = False
[docs] def fit(self, x, y):
"""fit ML model
:param x: training dataset
:param y: training dataset
:return: void
"""
self.trained = True
def predict(self, x):
if not self.trained:
msg = "{0} has not been trained, so no prediction can be made." \
.format(type(self).__name__)
raise RuntimeError(msg)
def predict_proba(self, x):
if not self.trained:
msg = "{0} has not been trained, so no prediction can be made." \
.format(type(self).__name__)
raise RuntimeError(msg)
# raise NotImplementedError('predict_proba not implemented.')
def linearize(self, x):
msg = "{0} has not defined a jacobian method." \
.format(type(self).__name__)
raise RuntimeError(msg)
[docs]class MultiFiSurrogateModel(SurrogateModel):
"""Base class for surrogate models using multi-fiddelity training data
:param SurrogateModel: Object
"""
[docs] def fit(self, x, y):
"""fit ML model
:param x: training dataset
:param y: training dataset
:return: void
"""
super(MultiFiSurrogateModel, self).train(x, y)
self.train_multifi([x], [y])
def train_multifi(self, x, y):
"""Trains the surrogate model, based on the given
multi-fidelity training data.
x: list of (m samples, n inputs) ndarrays
Values representing the multi-fidelity training case inputs.
y: list of ndarray
output training values which corresponds to the multi-fidelity
training case input given by x.
"""