Changelog

0.5.10

Date:: April 09, 2024

Features

Added a save and load method to class DOELib.Design
Added a random_seed optional argument to DOELib.Design
Added a get_design method to load an instance DOELib.Design intialized to the provided data

0.5.9

Date:: April 22, 2023

Fixes

Fix compatibility with python versions 3.8, 3.9, 3.10, and 3.11

0.5.8

Date:: April 18, 2023

Fixes

Fix the compute_volume method to avoid float object TypeError on call to math.factorial

0.5.7

Date:: March 04, 2023

Features

compute_absorption() method of MarginNetwork now has an optional method argument for specifying different root finding algorithms
Margins can now be allocated using different margin allocation strategies for each margin node using the allocate_margins method
Added technical documentation explaining the theory of the Margin Value Method

Fixes

Removed the update of Distribution histograms self.value_dist = self.values on every __call__ of scalar parameters to improve performance

Incompatible changes

forward() method of MarginNetwork now requires the argument strategy to be a list of strategies one for each Decision instance

0.5.6

Date:: June 29, 2022

Features

The method allocate_margins() allows the user to specify a margin allocation strategy such as min_excess or manual
The Decision class now stores selection_values in order to allow the use to probe the frequency of each decision made

Fixes

Fixed case when errors occur in MarginNode objects due to value_dist property, when attempting to generate histogram from samples
random() method of Specification now calls the random() method of Distribution implicitly
MarginNetwork instances now store Specification instances when using the save and load methods
MarginNetwork instances now store deterioration_vector``and ``utilization_matrix
Fixed the train_surrogate, train_inverse, and train_performance_surrogate methods to cap data at 100 * dim for KRG models

Incompatible changes

forward() method of MarginNetwork now requires two optional arguments allocate_margin, and strategy for all Decision instances
__call__() method of Decision now requires two arguments allocate_margin, and strategy
The method allocate_margins() must now be called before every forward() call of a MarginNetwork instance
It is recommended to use self.input_spec instead of those defined globally outside forward() and randomize() implementations

0.5.5

Date:: April 15, 2022

Fixes

Behaviour objects now support the train_inverse method (works only if the train_surrogate method has completed successfully) for training a surrogate model that inverts decided variables to any variable of choice (given by the key argument) and intermediate parameters (given by the intermediate attribute)
Behaviour objects now support the inv_call which can be user defined (must define the intermediate and inverted attributes) or from the train_inverse method
if use_estimate is False, the MarginNetwork must have equal number of decision and margin nodes associated with each other (1:1 ratio)
The user-implemented forward method of the MarginNetwork object must include the outputs optional argument if using inv_call anywhere in the method
Added error handling for train_surrogate function by rejecting NaN values during sampling of the blackbox

Incompatible changes

use_estimate argument of compute_impact method now relates to whether a surrogate is used or not for impact calculation
train_surrogate argument of Behaviour class now needs the arguments n_i, n_p, n_dv, and n_tt for each of the output parameters
output of a decision node is given by the attribute output_value instead of decided_value

0.5.4

Date:: April 10, 2022

Fixes

Removed example scripts since they are more for research than for utility
Modified impact on performance calculation compute_impact to cap performance parameters by their maximum and minimum values
Fixed impact on performance calculation compute_impact to utilized target thresholds and decided values instead of excess values from surrogate
Added visualization tests

0.5.3

Date:: March 29, 2022

Fixes

removed dependency on py

0.5.2

Date:: March 29, 2022

Features

First public release of mvmlib

0.5.1

Date:: March 27, 2022

Incompatible changes

Rename the library to mvmlib and the main module to mvm

0.5.0

Date:: March 14, 2022

Incompatible changes

The init_decisions() must be called before the first forward() and after every randomize() or changing the input specifications to store the universe of decided values
if the argument num_threads of train_performance_surrogate(), compute_absorption(), or compute_decisions() is greater then 1 then Jupyter notebook cannot be used with ur script
The argument value of InputSpec initializer must be a float or an integer only. A Distribution object must be passed to the optional argument distribution

Features

The Decision class can now support multiple decided values using the n_nodes optional argument. User must supple equal length of target thresholds
Can pass additional arguments as kwargs to behaviour model for Decision class during the __call__ method
The method train_performance_surrogate() of the class MarginNetwork``now supports different surrogate models specified using the ``sm_type argument
The method train_performance_surrogate() of the class MarginNetwork now supports parallel processing specified using the num_threads argument
The method compute_absorption() of the class MarginNetwork now supports parallel processing of the decision universe (which changes every time the input specifications self.spec_vector are iterated during absorption computation) specified using the num_threads argument
The method compute_decisions() of the class Decision now supports parallel processing specified using the num_threads argument

Fixes

More efficient compute_absorption() method by lumping threshold limit and specification limit calculations
Fixed train_performance_surrogate() method to properly handle ordinal type variables such as INT
Fixed compute_absorption() method to properly handle input specifications that have negative nominal values
Fixed compute_absorption() method to change nan values to zero
Added __deepcopy__ directives to all classes

0.4.8

Date:: February 20, 2022

Incompatible changes

Added mandatory argument variable_type to InputSpec and DesignParam classes during initialization

Features

Add the Decision class for defining decision nodes and off-the-shelf components
Add the integer and continuous type variables for InputSpec and DesignParam classes

0.4.7

Date:: February 17, 2022

Incompatible changes

AbsorptionMatrix class is removed, instead call MarginNetwork.absoprtion_matrix.value to retrieve absorption values
ImpactMatrix class is removed, instead call MarginNetwork.impact_matrix.value to retrieve impact values
AbsorptionMatrix.deteriorations attribute is removed, instead call MarginNetwork.deterioration_vector.value to retrieve deterioration values
All random sampling functions use the .random() method to draw samples. Cannot use the __call__ operator anymore
MarginNetwork method reset_outputs() not takes a single optional argument to reset by N samples
dist method reset_outputs() not takes a single optional argument to reset by N samples
Distribution object method .random() (previously __call__) now returns a 1D np.ndarray for one dimensional pdfs
Rename methods to comply with PEP 582 standard compute_mvp, get_array, set_func
Rename classes to comply with PEP 582 standard GuassianFunc, UniformFunc, TriangularFunc, FuzzySet, FuzzyFunc, FuzzySystem, FuzzyRule

Features

Separate absorption, deterioration, and impact matrics into separate MarginNetwork attributes
Use a Factory design parameter for defining matrix and vector caches used during stochastic simulation of MarginNetwork

0.4.6

Date:: January 14, 2022

Features

Add utilization calculation as part of the compute_absorption method
Add utilization storage to AbsorptionMatrix class
Add compute_MVP method to MarginNetwork class to show margin value map
Add nearest method to dmLib to allow calculation of the distance metric for the MVP

Fixes

Adapt train_performance_surrogate, view_perf, and compute_impact to include scaling functionality when training Kriging model
Add input specifications samples as input to performance surrogate in train_performance_surrogate to accommodate variability input specifications

Incompatible changes

InputSpec now requires the argument universe upon initialization

0.4.5

Date:: January 06, 2022

Features

Add distribution type uniformFunc for multivariate uniform distributions

Fixes

Fix MarginNode.value property to retrieve the last available sample after calling the reset(N) method
Fix value_dist property of Performance and MarginNode classes to construct histogram of samples and then initialize a Distribution class from them
Force absorption computing to ignore 0 deteriorations by outputting a np.nan
Make absorption computation sign independent
Add relevant tests for absorption and deterioration computation
Simplified length calculation procedure in strut_design.py example by using analytical expression instead of fsolve in B1 model

0.4.4

Date:: December 20, 2021

Features

Add ability to selectively choose how to randomize the MAN by redefining the randomize method of MarginNetwork
Selectively choose when to reset the outputs of the MAN using the reset_outputs method
Can retrieve design parameters, input specs, excess, target thresholds, decided values, and performances using the properties design_vector, spec_vector, excess_vector, dv_vector, tt_vector, and perf_vector, respectively
Add output storage class for a Margin Analysis Network (MAN) AbsorptionMatrix which stores absorption and deterioration
Add method compute_absorption to compute an observation of the change absorption capability matrix and deterioration vector
Add view(), view_cdf(), view_det(), view_det_cdf() methods to AbsorptionMatrix class by inheritance from VisualizeDist

Incompatible changes

instances of InputSpec should be called using the .value property just like DesignParam and FixedParam
train_performance_surrogate argument ext_samples now takes training points of (excess, performance) instead of (decided_value, performance)

0.4.3

Date:: December 18, 2021

Features

Add VisualizeDist class to uncertaintyLib.py module
Add output storage class for a Margin Analysis Network (MAN) Performance
Add output storage class for a Margin Analysis Network (MAN) ImpactMatrix
Add method train_performance_surrogate which uses the library SMT to estimate threshold performances
Add method compute_impact to compute an observation of the Impact on Performance matrix
Add method view_perf to MarginNetwork class to visualize 2D projections of performance surrogate models
Add view() and view_cdf() methods to Performance and ImpactMatrix classes by inheritance from VisualizeDist
Design class can now take array_like values for argument nsamples if using doe_type='full_fact'

Incompatible changes

move compute_cdf() method from class MarginNode to module level method in uncertaintyLib.py module
use property .values instead of excess to retrieve observations of excess from MarginNode object
Added dependency on SMT
view() and view_cdf() methods now take optional arguments folder, file, img_format, instead of just savefile
Argument type of Design initialization changed to doe_type to avoid overloading python object type

0.4.2

Date:: December 17, 2021

Features

Add building block for a Margin Analysis Network (MAN) as a class object InputSpec
Add building block for a Margin Analysis Network (MAN) as a class object FixedParam
Add building block for a Margin Analysis Network (MAN) as a class object DesignParam
Add building block for a Margin Analysis Network (MAN) as a class object Behaviour
Behaviour __call__ method must be redefined by the user
Add MarginNetwork class object that must be inherited and redefined by user
Add ability to call MarginNetwork.forward() in a Monte Carlo setting

0.4.1

Date:: December 15, 2021

Incompatible changes

MarginNode class object is now called using MarginNode(decided_value,threshold), where decided_value and threshold are vectors of equal length sampled from their respective functions

0.4.0

Date:: October 26, 2021

Features

Add building block for a Margin Analysis Network (MAN) as a class object MarginNode
Add ability to call MarginNode() using a set of requirement observations and design parameters in a Monte Carlo setting
Add ability to view MarginNode excess pdf and cdf using MarginNode.view() and MarginNode.view_cdf() methods

Fixes

Transfer class object labels to plot axes for fuzzySystem.view(), Distribution.view(), and gaussianFunc.view()

0.3.0

Date:: October 23, 2021

Features

Add support for defining arbitrary probability densities using raw density values Distribution(p)
Add support for random sampling from instance of Distribution by calling it
Add support for sampling from Gaussian distribution gaussianFunc by calling it directly
Add support for viewing samples from defined distribution using the .view() method for Distribution and gaussianFunc instances
Add support for viewing aggregate function after computing using .view() method for fuzzySystem after using .compute() method

Incompatible changes

Must manually reset fuzzySystem instance after .compute() to clear aggregate function

Fixes

Fixed problem with fuzzySystem.output_activation` not being calculated properly using element-wise operations
Add PDF_examples.py script
Improve existing tests test_fuzzyInference_N
Add new tests test_gaussian_pdf_rvs and test_arbitrary_pdf_rvs
Update documentation conf.py to include class docstring from __init__

0.2.1

Date:: October 14, 2021

Features

Add support for calculating probability density of multivariate Gaussian at a given Mahalanobis distance gaussianFunc.compute_density_r

Incompatible changes

Rename the method gaussianFunc.multivariateGaussian to gaussianFunc.compute_density_r

0.2.0

Date:: October 14, 2021

Features

Add support for multi-dimensional arrays or floats for triangularFunc.interp, fuzzyRule.apply, fuzzySet.interp, and fuzzySystem.compute
Update example TRS_example.py and documentation example to use these functionalities
Add support for directly plotting triangularFunc using triangularFunc.view()

Incompatible changes

Simplify API to directly import triangularFunc, fuzzyRule, fuzzySet, fuzzySystem, Design, and gaussianFunc

0.1.0

Date:: October 9, 2021

Features

Introduce fuzzyLib, DOELib, and uncertaintyLib, and fuzzySystem.compute
Introduce fuzzy inference using dmLib.fuzzyLib.fuzzySystem.fuzzySystem.compute() for a dict of floats
Add example TRS_example.py and documentation example to use these functionalities