GA features update - Having multi-objective optimization (NSGA-II ) #2326
Conversation
…e iteration is completed.
…tests. Most of conflicts are resolved.
…enablingMinMaxList
| @@ -43,7 +43,7 @@ Note all install methods after "main" take | |||
| <scikit-learn>1.0</scikit-learn> | |||
| <pandas/> | |||
| <!-- Note most versions of xarray work, but some (such as 0.20) don't --> | |||
| <xarray>2023</xarray> | |||
| <xarray>2023.12</xarray> | |||
There was a problem hiding this comment.
Any reason to change to this specific version?
There was a problem hiding this comment.
Yes, @joshua-cogliati-inl suggested this as it was causing diffs in the tests
There was a problem hiding this comment.
FYI, you may check PR #2365 (review) for library changes and try to avoid conflicts.
dependencies.xml
Outdated
| <setuptools/> | ||
| <!-- This is because liblapack 3.9.0 build 21 is broken (and can probably be removed if there ever is a build 22). This can also be removed when scipy is updated to version 1.12 --> | ||
| <liblapack skip_check='True' os='linux' machine='x86_64'>3.9.0=20_linux64_openblas</liblapack> | ||
| <liblapack skip_check='True' os='windows' machine='x86_64'>3.9.0=20_win64_mkl</liblapack> | ||
| <liblapack skip_check='True' os='mac' machine='x86_64'>3.9.0=20_osx64_openblas</liblapack> | ||
| <setuptools>69</setuptools> <!-- ray 2.6 can't be installed with setuptools 70 --> |
There was a problem hiding this comment.
It seems there are some diverges between this branch and devel. The changes are already in devel branch, but still show in this pull request. @Jimmy-INL Could you try to rebase your branch with devel branch?
There was a problem hiding this comment.
This is Junyung's branch, I tried to fix the failing tests. I will check who will address it, but thanks a lot for the comments.
There was a problem hiding this comment.
I think we actually copied this from devel.
| self._canHandleMultiObjective = False # boolean indicator whether optimization is a sinlge-objective problem or a multi-objective problem | ||
|
|
There was a problem hiding this comment.
This line can be moved to base class with default of False, while set it to True in GA. This will also apply to GradientDescent, and Simulated Annealing class.
There was a problem hiding this comment.
Regardless of whether the base class has this line with default of False, the other algorithms will eventually have this line. saying True (if they support multi-objective optimization). Plus, I think user can feel less confused if we place self._canHandleMultiObjective = True (or False) in every class of optimizer.
| if len(self._minMax) != len(self._objectiveVar): | ||
| self.raiseAnError(IOError, 'The length of <type> in <Optimizers>-<GeneticAlgorithm>-<SamplerInit> and <objective> in <Optimizers>-<GeneticAlgorithm> must be of the same length!') | ||
| if list(set(self._minMax)-set(['min','max'])) != []: | ||
| self.raiseAnError(IOError, "<type> under <Optimizers>-<GeneticAlgorhtm> must be a either 'min' and/or 'max'") |
There was a problem hiding this comment.
The error messages in Optimizer class should not associated with Genetic Algorithm.
| else: | ||
| rlz[self._objectiveVar] *= -1 |
There was a problem hiding this comment.
It seems the self._objectiveVar will be a list, will the else condition happen in this case?
There was a problem hiding this comment.
self._objectiveVar is list only in GA. Other optimization algorithms do not treat this variable as a list. This can be simplified later once everyone defines self._objectiveVar as a list.
| if 'max' in self._minMax: | ||
| if not self._canHandleMultiObjective and len(self._objectiveVar) == 1: | ||
| rlz[self._objectiveVar[0]] *= -1 | ||
| elif type(self._objectiveVar) == list: | ||
| for i in range(len(self._objectiveVar)): | ||
| if self._minMax[i] == 'max': | ||
| rlz[self._objectiveVar[i]] *= -1 | ||
| else: | ||
| rlz[self._objectiveVar] *= -1 |
There was a problem hiding this comment.
I think these lines can be simplified since self._objectiveVar will be a list instead, am I right here?
There was a problem hiding this comment.
Again, same answer as above.
self._objectiveVar is list only in GA. Other optimization code does not treat this variable as a list. This can be simplified later once everyone defines self._objectiveVar as a list.
| if len(self._objectiveVar) == 1: # Single Objective Optimization | ||
| objValue = rlz[self._objectiveVar[0]] | ||
| if 'max' in self._minMax: | ||
| objValue *= -1 | ||
| toExport[self._objectiveVar[0]] = objValue | ||
| else: # Multi Objective Optimization | ||
| for i in range(len(self._objectiveVar)): | ||
| objValue = rlz[self._objectiveVar[i]] | ||
| if self._minMax[i] == 'max': | ||
| objValue *= -1 | ||
| toExport[self._objectiveVar[i]] = objValue |
There was a problem hiding this comment.
This if ... else ... can be combined, since self._objectiveVar is a list now, and you do not need to treat single and multi-objective separately.
There was a problem hiding this comment.
Correct! I removed if-else statement.
|
|
||
| # 5.1 @ n-1: fitnessCalculation(rlz) | ||
| # perform fitness calculation for newly obtained children (rlz) | ||
| if not self._canHandleMultiObjective or len(self._objectiveVar) == 1: # This is for a single-objective Optimization case. |
There was a problem hiding this comment.
is there reason why there is no else condition here?
There was a problem hiding this comment.
It seems we don't need "if not self._canHandleMultiObjective or...... "
I removed entire line of 782.
| @ In, g, xr.DataArray, the constraint evaluation function | ||
| @ In, info, dict, identifying information about the realization | ||
| """ | ||
| self.raiseADebug('*'*80) |
There was a problem hiding this comment.
I removed self.raiseADebug('*'*80)
| coords={'chromosome':np.arange(np.shape(objVal)[0]), | ||
| 'obj': self._objectiveVar}) | ||
| if self._writeSteps == 'every': | ||
| print("### rlz.sizes['RAVEN_sample_ID'] = {}".format(rlz.sizes['RAVEN_sample_ID'])) |
There was a problem hiding this comment.
convert print into debug statements?
| @@ -1030,10 +1594,14 @@ def _addToSolutionExport(self, traj, rlz, acceptable): | |||
| # meta variables | |||
| toAdd = {'age': 0 if self.popAge is None else self.popAge, | |||
| 'batchId': self.batchId, | |||
| 'fitness': rlz['fitness'], | |||
| # 'fitness': rlz['fitness'], | |||
There was a problem hiding this comment.
fitness values are no need to be placed in toAdd anymore. I removed the comment.
| @@ -42,7 +47,8 @@ def rouletteWheel(population,**kwargs): | |||
| """ | |||
| # Arguments | |||
| pop = population | |||
| fitness = kwargs['fitness'] | |||
| fitness = np.array([item for sublist in datasetToDataArray(kwargs['fitness'], list(kwargs['fitness'].keys())).data for item in sublist]) | |||
| # fitness = kwargs['fitness'].data | |||
| selectionProb = shiftedFitness/np.sum(shiftedFitness) # Share of the pie (rouletteWheel) | ||
| sumProb = selectionProb[counter] | ||
|
|
||
| while sumProb < roulettePointer : | ||
| while sumProb <= roulettePointer : |
There was a problem hiding this comment.
was this added to properly cover discrete cases?
| newFitness = xr.DataArray(newFitness, | ||
| dims=['chromosome'], | ||
| coords={'chromosome':np.arange(np.shape(newFitness)[0])}) | ||
| # newFitness = newFitness.to_dataset(name = list(kwargs['fitness'].keys())[0]) |
| @@ -375,15 +375,6 @@ | |||
| [../] | |||
| [../] | |||
|
|
|||
| [./GAwithEnsembleModelHDSMconvergence] | |||
There was a problem hiding this comment.
@Jimmy-INL I need your comments for this question.
There was a problem hiding this comment.
@JunyungKim @Jimmy-INL I have finished my first part of review. I will to finish the rest in the afternoon. Please let me know if you have any questions regarding my comments.
| @@ -43,7 +43,7 @@ Note all install methods after "main" take | |||
| <scikit-learn>1.0</scikit-learn> | |||
| <pandas/> | |||
| <!-- Note most versions of xarray work, but some (such as 0.20) don't --> | |||
| <xarray>2023</xarray> | |||
| <xarray>2023.12</xarray> | |||
There was a problem hiding this comment.
FYI, you may check PR #2365 (review) for library changes and try to avoid conflicts.
| # if not self._canHandleMultiObjective and len(self._objectiveVar) == 1: | ||
| # self._objectiveVar = self._objectiveVar[0] |
| optVal = rlz[self._objectiveVar] | ||
| # if not self._canHandleMultiObjective and len(self._objectiveVar) == 1: | ||
| # self._objectiveVar = self._objectiveVar[0] | ||
| if len(self._objectiveVar) > 1 and type(self._objectiveVar)==list: |
There was a problem hiding this comment.
Will self._objectiveVar have different type except list? if not, I think you do not need the second condition here.
| descr=r"""Name of the response variable (or ``objective function'') that should be optimized | ||
| (minimized or maximized).""")) | ||
| descr=r"""Name of the objective variable(s) (or ``objective function'') that should be optimized | ||
| (minimized or maximized). It can be a single string or a list of strings if it is a multi-objective problem. """)) |
There was a problem hiding this comment.
Please add something like: "Currently, only genetic algorithm supports multi-objective ...."
| @@ -103,7 +103,8 @@ def getInputSpecification(cls): | |||
| descr=r"""seed for random number generation. Note that by default RAVEN uses an internal seed, | |||
| so this seed must be changed to observe changed behavior. \default{RAVEN-determined}""") | |||
| minMaxEnum = InputTypes.makeEnumType('MinMax', 'MinMaxType', ['min', 'max']) | |||
| minMax = InputData.parameterInputFactory('type', contentType=minMaxEnum, | |||
| minMaxList = InputTypes.StringListType() | |||
| minMax = InputData.parameterInputFactory('type', contentType=minMaxList, | |||
There was a problem hiding this comment.
Please also update the descr to reflect the list of min, max values from users' input.
| <author>Mohammad Abdo</author> | ||
| <created>2024-02-18</created> | ||
| <classesTested/> | ||
| <description>NSGA-II min-max test </description> |
There was a problem hiding this comment.
Please add more descriptions.
tests/framework/Optimizers/GeneticAlgorithms/continuous/unconstrained/ZDT1.xml
Outdated
Show resolved
Hide resolved
| <author>Junyung Kim</author> | ||
| <created>2022-12-21</created> | ||
| <classesTested/> | ||
| <description>NSGA-II min-min test </description> |
There was a problem hiding this comment.
please add more descriptions
| <author>Mohammad Abdo</author> | ||
| <created>2024-04-27</created> | ||
| <classesTested/> | ||
| <description>NSGA-II min-min test </description> |
There was a problem hiding this comment.
please update the description
|
|
||
| # @author: Mohammad Abdo (@Jimmy-INL) | ||
|
|
||
| def evaluate(Inputs): |
There was a problem hiding this comment.
please add some docstring
|
Job Precheck on 014dae4 : invalidated by @joshua-cogliati-inl failed in fetch and branch |
|
Job Precheck on 014dae4 : invalidated by @joshua-cogliati-inl civet problems |
There was a problem hiding this comment.
These are my second part comments. Please let me know if you have any issues. In addition, Please try to make sure you can generate the manual for multi-objective optimization, and check the manual is complete. @Jimmy-INL @JunyungKim
| # Fitness | ||
| if self._survivorSelectionType not in ['ageBased','fitnessBased','rankNcrowdingBased']: | ||
| self.raiseAnError(IOError, f'Currently constrained Genetic Algorithms only support ageBased, fitnessBased, and rankNcrowdingBased as a survivorSelector, whereas provided survivorSelector is {self._survivorSelectionType}') | ||
| if len(self._objectiveVar) == 1 and self._survivorSelectionType == 'rankNcrowdingBased': |
There was a problem hiding this comment.
Based on the error message, should you use if len(self._objectiveVar) <= 2 instead
There was a problem hiding this comment.
I do not understand, I thought it works with multiobjective cases, even with two objrctives. At least that's what it should do. I will test it further.
| # TODO: @mandd, please explore the possibility to convert the logistic fitness into a constrained optimization fitness. | ||
| if 'Constraint' in self.assemblerObjects and self._fitnessType not in ['invLinear','feasibleFirst']: | ||
| self.raiseAnError(IOError, f'Currently constrained Genetic Algorithms only support invLinear and feasibleFirst fitnesses, whereas provided fitness is {self._fitnessType}') | ||
| if 'Constraint' in self.assemblerObjects and self._fitnessType not in ['invLinear','logistic', 'feasibleFirst']: |
There was a problem hiding this comment.
I would suggest to use inputSpecs to check the fitnessType instead.
| if len(self._objectiveVar) == 1: | ||
| convs = {} | ||
| for conv in self._convergenceCriteria: | ||
| fName = conv[:1].upper() + conv[1:] | ||
| # get function from lookup | ||
| f = getattr(self, f'_checkConv{fName}') | ||
| # check convergence function | ||
| okay = f(traj, new=new, old=old) | ||
| # store and update | ||
| convs[conv] = okay | ||
| else: | ||
| convs = {} | ||
| for conv in self._convergenceCriteria: | ||
| fName = conv[:1].upper() + conv[1:] | ||
| # get function from lookup | ||
| f = getattr(self, f'_checkConv{fName}') | ||
| # check convergence function | ||
| okay = f(traj, new=new, old=old) | ||
| # store and update | ||
| convs[conv] = okay |
There was a problem hiding this comment.
I do not see the different between if and else statements here. If so, I think you need to revert the changes.
| if len(self._optPointHistory[traj]) < 2: | ||
| return False |
There was a problem hiding this comment.
Combine these lines within the if statement and move them outside.
| obj1 = o1[self._objectiveVar[0]] | ||
| obj2 = o1[self._objectiveVar[1]] | ||
| converged = (obj1 == self._convergenceCriteria['objective'] and obj2 == self._convergenceCriteria['objective']) |
There was a problem hiding this comment.
These lines indicate the convergence only work for 2 objectives, I think you need to extend this part to handle multi-objectives optimizations with more than 2 objectives.
| 'rank': 0 if ((type(self._objectiveVar) == list and len(self._objectiveVar) == 1) or type(self._objectiveVar) == str) else rlz['rank'], | ||
| 'CD': 0 if ((type(self._objectiveVar) == list and len(self._objectiveVar) == 1) or type(self._objectiveVar) == str) else rlz['CD'], |
There was a problem hiding this comment.
Please try to simply these two lines, since the self._objectiveVar is a list.
|
Job Test Ubuntu 18 PIP on 730d7a4 : invalidated by @joshua-cogliati-inl cleared cache |
1 similar comment
|
Job Test Ubuntu 18 PIP on 730d7a4 : invalidated by @joshua-cogliati-inl cleared cache |
|
Job Test Ubuntu 18 PIP on 730d7a4 : invalidated by @joshua-cogliati-inl monitoring disk usage |
…x_ReBASE_05232024
Pull Request Description
What issue does this change request address? (Use "#" before the issue to link it, i.e., #42.)
#2069
What are the significant changes in functionality due to this change request?
For Change Control Board: Change Request Review
The following review must be completed by an authorized member of the Change Control Board.
<internalParallel>to True.raven/tests/framework/user_guideandraven/docs/workshop) have been changed, the associated documentation must be reviewed and assured the text matches the example.