Mandd/pareto front PP prestruct

doc/user_manual/postprocessor.tex

@@ -1863,50 +1863,37 @@ \subsubsection{RavenOutput}
 
 \subsubsection{ParetoFrontier}
 \label{ParetoFrontierPP}
-The \textbf{ParetoFrontier} post-processor is designed to identify the points lying on the Pareto Frontier in a cost-value space.
-This post-processor receives as input a \textbf{DataObject} (a PointSet only) which contains all data points in the cost-value space and it 
+The \textbf{ParetoFrontier} post-processor is designed to identify the points lying on the Pareto Frontier in a multi-dimensional trade-space.
+This post-processor receives as input a \textbf{DataObject} (a PointSet only) which contains all data points in the trade-space space and it 
 returns the subset of points lying in the Pareto Frontier as a PointSet.
 
 It is here assumed that each data point of the input PointSet is a realization of the system under consideration for a 
-specific configuration to which corresponds a cost and a value.
+specific configuration to which corresponds several objective variables (e.g., cost and value).
 
 %
 \ppType{ParetoFrontier}{ParetoFrontier}
 %
 \begin{itemize}
-  \item  \xmlNode{costID},\xmlDesc{string, required parameter}, ID of the input PointSet variable that is considered the cost variable. 
+  \item   \xmlNode{objective},\xmlDesc{string, required parameter}, ID of the objective variable that represents a dimension of the trade-space space. 
           The \xmlNode{costID} requires one identifying attribute:
           \begin{itemize}
-            \item \xmlAttr{inv}, \xmlDesc{boolean, required field}, False if costID represents a cost, True if costID represents a saving
-          \end{itemize}
-  \item  \xmlNode{valueID},\xmlDesc{string, required parameter}, ID of the input PointSet variable that is considered the value variable
-          The \xmlNode{valueID} requires one identifying attribute:
-          \begin{itemize}
-            \item \xmlAttr{inv}, \xmlDesc{boolean, required field}, False if valueID represents a value, True if valueID represents a loss of value
-          \end{itemize}
-  \item  \xmlNode{costLimit},\xmlDesc{float, optional parameter}, maximum value of the cost variable to be considered in the pareto frontier
-          The \xmlNode{costLimit} requires one identifying attribute:
-          \begin{itemize}
-            \item \xmlAttr{type}, \xmlDesc{string, required field}, this attribute can be either ``lower" or ``upper'', it indicates if the limit is an upper or lower limit
-          \end{itemize}
-  \item  \xmlNode{valueLimit},\xmlDesc{float, optional parameter}, minimum value of the value variable to be considered in the pareto frontier
-          The \xmlNode{valueLimit} requires one identifying attribute:
-          \begin{itemize}
-            \item \xmlAttr{type}, \xmlDesc{string, required field}, this attribute can be either ``lower" or ``upper'', it indicates if the limit is an upper or lower limit
+            \item \xmlAttr{goal}, \xmlDesc{string, required field}, Goal of the objective variable characteristic: minimzation (min) or maximization (max)
+            \item \xmlAttr{upperLimit}, \xmlDesc{string, optional field}, Desired upper limit of the objective variable for the points in the Pareto frontier
+            \item \xmlAttr{lowerLimit}, \xmlDesc{string, optional field}, Desired lower limit of the objective variable for the points in the Pareto frontier
           \end{itemize}
 \end{itemize}
 
 The following is an example where a set of realizations (the ``candidates'' PointSet) has been generated by changing two parameters 
-(var1 and var2) which produced two output variables (cost and value).
+(var1 and var2) which produced two output variables: cost (which it is desired to be minimized) and value (which it is desired to be maximized).
 The \textbf{ParetoFrontier} post-processor takes the ``candidates'' PointSet and populates a Point similar in structure 
 (the ``paretoPoints'' PointSet).
 
 \textbf{Example:}
 \begin{lstlisting}[style=XML,morekeywords={anAttribute},caption=ParetoFrontier input example (no expand)., label=lst:ParetoFrontier_PP_InputExample]
   <Models>
     <PostProcessor name="paretoPP" subType="ParetoFrontier">
-      <costID>cost</costID>
-      <valueID>value</valueID>
+      <objective goal='min' upperLimit='0.5'>cost</objective>
+      <objective goal='max' lowerLimit='0.5'>value</objective>
     </PostProcessor>   
   </Models>
 
@@ -1930,7 +1917,7 @@ \subsubsection{ParetoFrontier}
   </DataObjects>
 \end{lstlisting}
 
-\nb it is possible to specify upper and lower limits for the cost and value variable respectively.
+\nb it is possible to specify both upper and lower limits for each objective variable.
 When one or both of these limits are specified, then the pareto frontier is filtered such that all pareto frontier points that
 satisfy those limits are preserved.
 

framework/Models/PostProcessors/ParetoFrontierPostProcessor.py

@@ -25,6 +25,7 @@
 from .PostProcessor import PostProcessor
 from utils import utils
 from utils import InputData, InputTypes
+from utils import frontUtils
 import Runners
 #Internal Modules End-----------------------------------------------------------
 
@@ -47,6 +48,7 @@ def __init__(self, runInfoDict):
     self.validDataType = ['PointSet'] # The list of accepted types of DataObject
     self.outputMultipleRealizations = True # True indicate multiple realizations are returned
 
+
   @classmethod
   def getInputSpecification(cls):
     """
@@ -58,48 +60,31 @@ class cls.
     """
     inputSpecification = super(ParetoFrontier, cls).getInputSpecification()
 
-    costIDInput = InputData.parameterInputFactory("costID", contentType=InputTypes.StringType)
-    costIDInput.addParam("inv", InputTypes.BoolType, True)
-    inputSpecification.addSub(costIDInput)
-
-    valueIDInput = InputData.parameterInputFactory("valueID", contentType=InputTypes.StringType)
-    valueIDInput.addParam("inv", InputTypes.BoolType, True)
-    inputSpecification.addSub(valueIDInput)
-
-    costLimitInput = InputData.parameterInputFactory("costLimit", contentType=InputTypes.FloatType)
-    costLimitInput.addParam("type", InputTypes.StringType, True)
-    inputSpecification.addSub(costLimitInput)
+    objDataType = InputTypes.makeEnumType("objective", "objectiveType", ['min','max'])
 
-    valueLimitInput = InputData.parameterInputFactory("valueLimit", contentType=InputTypes.FloatType)
-    valueLimitInput.addParam("type", InputTypes.StringType, True)
-    inputSpecification.addSub(valueLimitInput)
+    objective = InputData.parameterInputFactory('objective', contentType=InputTypes.StringType)
+    objective.addParam('goal',       param_type=objDataType,           required=True)
+    objective.addParam('upperLimit', param_type=InputTypes.FloatType,  required=False)
+    objective.addParam('lowerLimit', param_type=InputTypes.FloatType,  required=False)
+    inputSpecification.addSub(objective)
 
     return inputSpecification
 
+
   def _handleInput(self, paramInput):
     """
       Function to handle the parsed paramInput for this class.
       @ In, paramInput, ParameterInput, the already-parsed input.
       @ Out, None
     """
-    PostProcessor._handleInput(self, paramInput)
-    costID  = paramInput.findFirst('costID')
-    self.costID  = costID.value
-    self.invCost = costID.parameterValues['inv']
-
-    valueID = paramInput.findFirst('valueID')
-    self.valueID = valueID.value
-    self.invValue = valueID.parameterValues['inv']
+    self.objectives = {}
 
-    costLimit  = paramInput.findFirst('costLimit')
-    if costLimit is not None:
-      self.costLimit  = costLimit.value
-      self.costLimitType = costLimit.parameterValues['type']
-
-    valueLimit = paramInput.findFirst('valueLimit')
-    if valueLimit is not None:
-      self.valueLimit = valueLimit.value
-      self.valueLimitType = valueLimit.parameterValues['type']
+    for child in paramInput.subparts:
+      if child.getName() == 'objective':
+        self.objectives[child.value]={}
+        self.objectives[child.value]['goal']       = child.parameterValues['goal']
+        self.objectives[child.value]['upperLimit'] = child.parameterValues.get('upperLimit')
+        self.objectives[child.value]['lowerLimit'] = child.parameterValues.get('lowerLimit')
 
 
   def inputToInternal(self, currentInp):
@@ -119,6 +104,7 @@ def inputToInternal(self, currentInp):
                                  .format(self.name, currentInp.type))
     return currentInp
 
+
   def run(self, inputIn):
     """
       This method executes the postprocessor action.
@@ -128,40 +114,24 @@ def run(self, inputIn):
     inData = self.inputToInternal(inputIn)
     data = inData.asDataset()
 
-    if self.invCost:
-      data[self.costID] = (-1.) * data[self.costID]
-    if self.invValue:
-      data[self.valueID] = (-1.) * data[self.valueID]
-
-    sortedData = data.sortby(self.costID)
-    coordinates = np.zeros(1,dtype=int)
-    for index,elem in enumerate(sortedData[self.costID].values):
-      if (index>1) and (sortedData[self.valueID].values[index]>sortedData[self.valueID].values[coordinates[-1]]):
-        # the point at index is part of the pareto frontier
-        coordinates = np.append(coordinates,index)
-
-    selection = sortedData.isel(RAVEN_sample_ID=coordinates)
-
-    if self.invCost:
-      selection[self.costID] = (-1.) * selection[self.costID]
-    if self.invValue:
-      selection[self.valueID] = (-1.) * selection[self.valueID]
-
-    if self.valueLimit is not None:
-      if self.valueLimitType=="upper":
-        selection = selection.where(selection[self.valueID]<=self.valueLimit)
-      else:
-        selection = selection.where(selection[self.valueID]>=self.valueLimit)
-    if self.costLimit is not None:
-      if self.costLimitType=="upper":
-        selection = selection.where(selection[self.costID]<=self.costLimit)
-      else:
-        selection = selection.where(selection[self.costID]>=self.costLimit)
+    dataTemp = data[list(self.objectives.keys())]
+    for index,obj in enumerate(self.objectives.keys()):
+      if self.objectives[obj]['goal']=='max':
+        dataTemp[obj] = (-1.) * dataTemp[obj]
+
+    paretoFrontMask = frontUtils.nonDominatedFrontier(np.transpose(dataTemp.to_array().values), returnMask=False)
+    selection = data.isel(RAVEN_sample_ID=np.array(paretoFrontMask))
+
+    for obj in self.objectives.keys():
+      if self.objectives[obj]['upperLimit']:
+        selection = selection.where(selection[obj]<=self.objectives[obj]['upperLimit'])
+      if self.objectives[obj]['lowerLimit']:
+        selection = selection.where(selection[obj]>=self.objectives[obj]['lowerLimit'])
 
     filteredParetoFrontier = selection.to_array().values
     paretoFrontierData = np.transpose(filteredParetoFrontier)
     paretoFrontierDict = {}
-    for index,varID in enumerate(sortedData.data_vars):
+    for index,varID in enumerate(data.data_vars):
       paretoFrontierDict[varID] = paretoFrontierData[:,index]
     paretoFrontierDict = {'data':paretoFrontierDict, 'dims':{}}
     return paretoFrontierDict

framework/utils/frontUtils.py

@@ -22,7 +22,7 @@
 
 
 
-def nonDominatedFrontier(data, returnMask):
+def nonDominatedFrontier(data, returnMask, minMask=None):
   """
     This method is designed to identify the set of non-dominated points (nEfficientPoints)
 
@@ -38,11 +38,21 @@ def nonDominatedFrontier(data, returnMask):
 
     @ In, data, np.array, data matrix (nPoints, nCosts) containing the data points
     @ In, returnMask, bool, type of data to be returned: indices (False) or True/False mask (True)
+    @ Out, minMask, np.array, array (nCosts,1) of boolean values: True (dimension need to be minimized), False (dimension need to be maximized)
     @ Out, isEfficientMask , np.array, data matrix (nPoints,1), array  of boolean values if returnMask=True
     @ Out, isEfficient, np.array, data matrix (nEfficientPoints,1), integer array of indexes if returnMask=False
 
     Reference: the following code has been adapted from https://stackoverflow.com/questions/32791911/fast-calculation-of-pareto-front-in-python
   """
+  if minMask is None:
+    pass
+  elif minMask is not None and minMask.shape[0] != data.shape[1]:
+    raise IOError("nonDominatedFrontier method: Data features do not match minMask dimensions: data has shape " + str(data.shape) + " while minMask has shape " + str(minMask.shape))
+  else:
+    for index,elem in np.ndenumerate(minMask):
+      if not elem:
+        data[:,index] = -1. * data[:,index]
+
   isEfficient = np.arange(data.shape[0])
   nPoints = data.shape[0]
   nextPointIndex = 0

tests/framework/PostProcessors/ParetoFrontierPostProcessor/test_paretoFrontier.xml

@@ -21,8 +21,8 @@
 
   <Models>
     <PostProcessor name="paretoPP" subType="ParetoFrontier">
-      <costID inv="False" >cost</costID>
-      <valueID inv="False" >value</valueID>
+      <objective goal='min'>cost</objective>
+      <objective goal='max'>value</objective>
     </PostProcessor>   
   </Models>
 

tests/framework/PostProcessors/ParetoFrontierPostProcessor/test_paretoFrontierInverted.xml

@@ -21,8 +21,8 @@
 
   <Models>
     <PostProcessor name="paretoPP" subType="ParetoFrontier">
-      <costID  inv="True" >out2</costID>
-      <valueID inv="True" >out1</valueID>
+      <objective goal='max'>out2</objective>
+      <objective goal='min'>out1</objective>
     </PostProcessor>   
   </Models>
 

tests/framework/PostProcessors/ParetoFrontierPostProcessor/test_paretoFrontierWithBothLimits.xml

@@ -22,10 +22,8 @@
 
   <Models>
     <PostProcessor name="paretoPP" subType="ParetoFrontier">
-      <costID inv="False" >cost</costID>
-      <valueID inv="False" >value</valueID>
-      <valueLimit type="lower">0.5</valueLimit>
-      <costLimit type="upper">0.5</costLimit>
+      <objective goal='min' upperLimit='0.5'>cost</objective>
+      <objective goal='max' lowerLimit='0.5'>value</objective>
     </PostProcessor>   
   </Models>
 

tests/framework/PostProcessors/ParetoFrontierPostProcessor/test_paretoFrontierWithCostLimit.xml

@@ -22,9 +22,8 @@
 
   <Models>
     <PostProcessor name="paretoPP" subType="ParetoFrontier">
-      <costID inv="False" >cost</costID>
-      <valueID inv="False" >value</valueID>
-      <costLimit type="upper">0.5</costLimit>
+      <objective goal='min' upperLimit='0.5'>cost</objective>
+      <objective goal='max'>value</objective>
     </PostProcessor>   
   </Models>
 

tests/framework/PostProcessors/ParetoFrontierPostProcessor/test_paretoFrontierWithValueLimit.xml

@@ -22,9 +22,8 @@
 
   <Models>
     <PostProcessor name="paretoPP" subType="ParetoFrontier">
-      <costID inv="False" >cost</costID>
-      <valueID inv="False" >value</valueID>
-      <valueLimit type="lower">0.5</valueLimit>
+      <objective goal='min'>cost</objective>
+      <objective goal='max' lowerLimit='0.5'>value</objective>
     </PostProcessor>   
   </Models>
 

tests/framework/PostProcessors/ParetoFrontierPostProcessor/tests

@@ -2,27 +2,27 @@
  [./ParetoFrontier]
   type = 'RavenFramework'
   input = 'test_paretoFrontier.xml'
-  csv = 'ParetoFrontier/PrintPareto.csv'
+  UnorderedCsv = 'ParetoFrontier/PrintPareto.csv'
  [../]
  [./ParetoFrontierWithCostLimit]
   type = 'RavenFramework'
   input = 'test_paretoFrontierWithCostLimit.xml'
-  csv = 'ParetoFrontier/PrintParetoCostLimit.csv'
+  UnorderedCsv = 'ParetoFrontier/PrintParetoCostLimit.csv'
  [../]
  [./ParetoFrontierWithValueLimit]
   type = 'RavenFramework'
   input = 'test_paretoFrontierWithValueLimit.xml'
-  csv = 'ParetoFrontier/PrintParetoValueLimit.csv'
+  UnorderedCsv = 'ParetoFrontier/PrintParetoValueLimit.csv'
  [../]
  [./ParetoFrontierWithBothLimits]
   type = 'RavenFramework'
   input = 'test_paretoFrontierWithBothLimits.xml'
-  csv = 'ParetoFrontier/PrintParetoBothLimits.csv'
+  UnorderedCsv = 'ParetoFrontier/PrintParetoBothLimits.csv'
  [../]
  [./ParetoFrontierInverted]
   type = 'RavenFramework'
   input = 'test_paretoFrontierInverted.xml'
-  csv = 'ParetoFrontier/PrintParetoInverted.csv'
+  UnorderedCsv = 'ParetoFrontier/PrintParetoInverted.csv'
  [../]
 []