def manageAttrs(self):
extensionPoolPercentage = .2
self.maxAttrsCount = self.maxAvgAttrsPerTree*len(self.treesPool)
if self.writeLog:
self.logWriter.writeLine( '\nManaging attrs. Initial attrs count = ',self.currentAttrsCount,'/',self.maxAttrsCount)
self.logWriter.timeMarker('manageAttrs',False)
deletionSequence = helpers.proportionalrandom(self.treesPool,
lambda x: x.lenStatements()/(x.fitness*x.numNodes),
max(0,int(self.currentAttrsCount- self.maxAttrsCount*(1-extensionPoolPercentage)) ))
if self.writeLog:
self.logWriter.writeLine('Deletion sequence length = ', len(deletionSequence))
_s=0
while len(deletionSequence) !=0:
count = 0
elem = deletionSequence[-1]
while deletionSequence[-1] == elem:
deletionSequence.pop()
count+=1
if len(deletionSequence) ==0:
break;
outcome, delSequence = elem.removeWorstStates(count)
if outcome:
self.currentAttrsCount -= count #if an entire tree is removed, the same is done inside removeTree method.
else:
self.removeTree(elem)
_s+=1
if self.writeLog:
self.logWriter.writeLine( _s,' trees removed due to zero attrs')
extensionSequence = helpers.proportionalrandom(self.treesPool,lambda x: x.fitness*x.numNodes / x.lenStatements(),max(self.maxAttrsCount-self.currentAttrsCount, 0))
if self.writeLog:
self.logWriter.writeLine('Extension sequence length = ', len(extensionSequence))
while len(extensionSequence)!=0:
count=0
elem = extensionSequence[-1]
while extensionSequence[-1] == elem:
extensionSequence.pop()
count+=1
if len(extensionSequence) ==0:
break;
trueCount = elem.expandBestStates(count)
self.currentAttrsCount+=trueCount
if self.writeLog:
self.logWriter.timeMarker('manageAttrs')
self.logWriter.writeLine( 'OnAfterSubStep attrs count = ',self.currentAttrsCount,'/',self.maxAttrsCount)
self.logWriter.writeLine( 'End of attrs management.\n')
def processSamples(self,samples, sumAdditionRates = 5.):
"""processes a list of samples"""
for sample in samples:
self.updateWithSample(sample,sumAdditionRates/len(samples))
self.samplesPool[sample] = set()
treeseq = helpers.proportionalrandom(self.treesPool,
lambda x:x.fitness/len(x.samples),
int(round(sumAdditionRates*self.maxSamplesCount)))
while len(treeseq) !=0:
count = 0
elem = treeseq[-1]
while treeseq[-1] == elem:
treeseq.pop()
count+=1
if len(treeseq) ==0:
break;
count = min(count, len(samples))
self.currentSamplesCount += count
for samp in helpers.proportionalrandomNoRep(samples,lambda x:1,count):#WHY THE F*****G F**K DOES lambda x:<sum of fitnesses of trees failed> result in drops?!
elem.addSample(samp)
self.samplesPool[samp].add(elem)
for sample in samples:
if len(self.samplesPool[sample])==0:
del self.samplesPool[sample]
def processSamples(self,samples, sumAdditionRates = 5.):
"""processes a list of samples"""
for sample in samples:
self.updateWithSample(sample,sumAdditionRates/len(samples))
assert sample not in self.samplesPool
self.samplesPool[sample] = set()
#if you strongly want to reprocess existing samples, you have at least 4 options:
# 1) make sure that the sample is not added to the trees that have it already
# 2) use lists instead of sets to store samples
# 3) assign sample weights
# 4) simply make a shallow copy of repeated samples right before this loop
treeseq = helpers.proportionalrandom(self.treesPool,
lambda x:x.fitness/len(x.samples),
int(round(sumAdditionRates*self.maxSamplesCount)))
while len(treeseq) !=0:
count = 0
elem = treeseq[-1]
while treeseq[-1] == elem:
treeseq.pop()
count+=1
if len(treeseq) ==0:
break;
count = min(count, len(samples))
self.currentSamplesCount += count
for samp in helpers.proportionalrandomNoRep(samples,lambda x:1,count):#WHY THE F*****G F**K DOES lambda x:<sum of fitnesses of trees failed> result in drops?!
assert elem.addSample(samp)
self.samplesPool[samp].add(elem)
for sample in samples:
if len(self.samplesPool[sample])==0:
del self.samplesPool[sample]
def generateChromo(self,numNodes,numSamples,numStatements,fitness= 1):
newCh = chromo(self,
set(helpers.proportionalrandom(self.samplesPool.keys(),lambda x:1,min(numSamples,len(self.samplesPool)))) ,
self.boolStatements,
self.numStatements,
1)
newCh.removeWorstStates(max([newCh.lenStatements() - numStatements,0]))
for i in range(numNodes):
if not newCh.expandBestNode():
break
if newCh.numNodes ==0:
return False
newCh.fitness = fitness
return newCh
def manageSamples(self):
'''samples removal from trees (proportional to len(tree.samples)/tree.fitness) on exceeding maximal count, as well as [planned] recombination and spreading'''
#ALARME!! try clustering here. may be good to cluster pairs with p ~ distance_between_samples/sum_of_importances but distance must be measured in terms of STATEMENTS of STATEMENT POOL proportionally to their IMPORTANCE... maybe
#ALARME!! a place for spreading of existent samples, IF IT IS NECCESARY given that they spread via tree recombination
#ALARME!! it may happen that a rubbish sample is kept cause it happened to be in one good chromo's list, test it and mb fix it somehow. p.e. expand 'good' samples in proportion to their fitness/len(self.samplesPool[sample]), but not to the 'best' nodes as won't fix the initial problem (mb uniformly?)
if self.writeLog:
self.logWriter.writeLine( '\nManaging samples. Initial samples count = ',self.currentSamplesCount,'/',self.maxSamplesCount)
self.logWriter.timeMarker('manageSamples',False)
deletionSequence = helpers.proportionalrandom(self.treesPool,
lambda x: len(x.samples)/x.fitness,
max(0,self.currentSamplesCount- self.maxSamplesCount ))
if self.writeLog:
self.logWriter.writeLine('Deletion sequence length = ', len(deletionSequence))
_s=0
while len(deletionSequence) !=0:
count = 0
elem = deletionSequence[-1]
while deletionSequence[-1] == elem:
deletionSequence.pop()
count+=1
if len(deletionSequence) ==0:
break;
outcome, delSequence = elem.removeSamples(count)
if outcome:
self.currentSamplesCount -= count #if an entire tree is removed, the same is done inside removeTree method.
for sample in delSequence:
self.samplesPool[sample].remove(elem)
if len(self.samplesPool[sample])==0:
del self.samplesPool[sample]
else:
self.removeTree(elem)
_s+=1
if self.writeLog:
self.logWriter.writeLine( _s,' trees removed due to zero samples')
self.logWriter.timeMarker('manageSamples')
self.logWriter.writeLine( 'OnAfterStep samples count = ',self.currentSamplesCount,'/',self.maxSamplesCount)
self.logWriter.writeLine( 'End of samples management.\n')
def initialise(self):
'''create initial pool of trees'''
rateTrees = .3
baseFitnessForIdealTree = 1
if self.writeLog:
self.logWriter.writeLine('starting initialisation')
self.logWriter.timeMarker('init',False)
samplesPerTree = int(round(self.maxSamplesCount/(rateTrees*self.maxNodesCount)+0.49))
nodesPerTree = int(1/rateTrees)
countTrees = int(self.maxNodesCount*rateTrees)
self.maxAttrsCount= countTrees
#ALARMA!! may need some details for samples/statements pools once they're implemented
for i in range(countTrees):
newCh = chromo(self,
set(helpers.proportionalrandom(self.samplesPool.keys(),lambda x:1,min(samplesPerTree,len(self.samplesPool)))) ,
self.boolStatements,
self.numStatements,
1)
for i in range(nodesPerTree):
if not newCh.expandBestNode():
break
if newCh.numNodes ==0:
continue
newCh.fitness = helpers.getSuccessProbability(newCh,self.samplesPool.keys())*baseFitnessForIdealTree
self.currentNodesCount+= newCh.numNodes
self.currentSamplesCount+=len(newCh.samples)
self.currentAttrsCount += len(newCh.boolStatements)+len(newCh.numStatements)
self.treesPool.add(newCh)
for sample in newCh.samples:
self.samplesPool[sample].add(newCh)
for sample in self.samplesPool.keys():
if len(self.samplesPool[sample])==0:
del self.samplesPool[sample]
if self.writeLog:
self.logWriter.timeMarker('init')
self.logWriter.writeLine('init phase end.', len(self.treesPool),'trees created with',
nodesPerTree,'nodes and',samplesPerTree,'samples each')
def manageTrees(self):
''' update the pool'''
recombinationRate = .025
expandedPoolCapacityRate = 1.2
if self.writeLog:
self.logWriter.writeLine( '\nManaging trees pool. Initial nodes count = ', self.currentNodesCount,'/',self.maxNodesCount, ', trees count = ',len(self.treesPool))
self.logWriter.timeMarker('manageTrees',False)
self.logWriter.timeMarker('filterNegFitness',False)
#remove all trees with their fitness <=0
_s = 0
for i in copy.copy(self.treesPool):
if i.fitness <=0:#better be a deadline. check if this one is ever reached
self.removeTree(i)
_s+=1
if self.writeLog:
self.logWriter.timeMarker('filterNegFitness')
self.logWriter.writeLine(_s,' trees removed due to negative fitness')
self.logWriter.writeLine('\nRecombination phase')
self.logWriter.timeMarker('recombination',False)
_s = 0
# add some new trees
mutationSequence = helpers.proportionalrandom(self.treesPool,lambda x:(x.fitness),int(self.maxNodesCount*recombinationRate*2))
if self.writeLog:
self.logWriter.writeLine('Recombination phase: sequence len =',len(mutationSequence))
random.shuffle(mutationSequence)
#ALARME!!actual mutation for trees may be implemented either here or separately. or not implemented due to it's simulated by all the alterations with samples and nodes.
while len(mutationSequence)>1:
self.recombine(mutationSequence.pop(),mutationSequence.pop())
_s+=1
if self.writeLog:
self.logWriter.timeMarker('recombination')
self.logWriter.writeLine('End of recombination phase')
self.logWriter.writeLine('\nExpansion phase')
self.logWriter.timeMarker('expansion',False)
#expand nodes up to maximal nodescount times expandedPoolCapacityRate (actually less due to absense of repetitions),
# prioritised by tree fitness times expected benefit from expansion
expandees = set(helpers.proportionalrandom(self.treesPool,lambda x:x.fitness*x.getBestReplacementPotential(),max(int((self.maxNodesCount*expandedPoolCapacityRate-self.currentNodesCount)), 0)))
if self.writeLog:
self.logWriter.writeLine( 'Expansion phase: sequence len =',len(expandees))
for tree in expandees:
if tree.expandBestNode():
self.currentNodesCount+=1
if self.writeLog:
self.logWriter.timeMarker('expansion')
self.logWriter.writeLine('End of expansion phase')
self.logWriter.writeLine('\nPruning phase')
self.logWriter.timeMarker('pruning',False)
#remove up to all nodes above nodescount(actually less due to possibility that trees have less nodes than tis demanded to remove)
# of least effective nodes in terms of nodes per fitness, remove empty trees
reducees = helpers.proportionalrandom(self.treesPool,lambda x:x.numNodes/(x.fitness),max(self.currentNodesCount - self.maxNodesCount,0))
if self.writeLog:
self.logWriter.writeLine('Pruning phase: sequence len =',len(reducees))
_s=0
for tree in reducees:
if tree in self.treesPool:
if tree.pruneWorstNode():
self.currentNodesCount-=1 #if pruning fails, -1 node is done in removeTree
else:
_s+=1
self.removeTree(tree)
if self.writeLog:
self.logWriter.writeLine( _s,' trees removed in pruning phase')
self.logWriter.timeMarker('pruning')
self.logWriter.writeLine('End of pruning phase')
if self.writeLog:
self.logWriter.timeMarker('manageTrees')
self.logWriter.writeLine( 'OnAfterStep nodes count = ', self.currentNodesCount,', trees count = ',len(self.treesPool))
self.logWriter.writeLine( 'End of tree management.\n')
def manageTrees(self):
''' update the pool'''
expandedPoolCapacityRate = 1.2
targetTreesCount = 0.3*self.maxNodesCount
minTreesCount = 10
if self.writeLog:
self.logWriter.writeLine( '\nManaging trees pool. Initial nodes count = ', self.currentNodesCount,'/',self.maxNodesCount, ', trees count = ',len(self.treesPool))
self.logWriter.timeMarker('manageTrees',False)
self.logWriter.timeMarker('negative fitness removal',False)
_t = 0
_r = 0
for chromo in copy.copy(self.treesPool):
if chromo.fitness <=0:
self.removeTree(chromo)
_t +=1
while len(self.treesPool) < minTreesCount:
newCh = self.generateChromo(self.currentNodesCount/len(self.treesPool),len(self.treesPool),self.currentSamplesCount/len(self.treesPool),self.currentAttrsCount/len(self.treesPool))
if newCh:
newCh.fitness = helpers.getSuccessProbability(newCh,self.samplesPool.keys())
self.addChromo(newCh)
_r+=1
if self.writeLog:
self.logWriter.timeMarker('negative fitness removal')
self.logWriter.writeLine(_t,'trees removed for negative fitness, ',_r,' replaced')
self.logWriter.writeLine('\nRecombination phase')
self.logWriter.timeMarker('recombination',False)
treeGrowthRate = 5
_s = 0
# add some new trees
mutationSequence = helpers.proportionalrandom(self.treesPool,lambda x:(x.fitness),2*treeGrowthRate)
if self.writeLog:
self.logWriter.writeLine('Recombination phase: sequence len =',len(mutationSequence))
random.shuffle(mutationSequence)
#ALARME!!actual mutation for trees may be implemented either here or separately. or not implemented due to it's simulated by all the alterations with samples and nodes.
while len(mutationSequence)>1:
newChromo = self.recombine(mutationSequence.pop(),mutationSequence.pop());
if newChromo:
self.addChromo(newChromo)
_s+=1
if self.writeLog:
self.logWriter.timeMarker('recombination')
self.logWriter.writeLine('End of recombination phase')
self.logWriter.writeLine('\nExpansion phase')
self.logWriter.timeMarker('expansion',False)
#expand nodes up to maximal nodescount times expandedPoolCapacityRate (actually less due to absense of repetitions),
# prioritised by tree fitness times expected benefit from expansion
#this NoRep is recalculating priorities every round.
expandees = helpers.proportionalrandomRounds(self.treesPool,lambda x:x.fitness*x.getBestReplacementPotential(),max(int((self.maxNodesCount*expandedPoolCapacityRate-self.currentNodesCount)), 0))#that one was set(random) just a sec ago
if self.writeLog:
self.logWriter.writeLine( 'Expansion phase: sequence len =',len(expandees))
for tree in expandees:
if tree.expandBestNode():
self.currentNodesCount+=1
if self.writeLog:
self.logWriter.timeMarker('expansion')
self.logWriter.writeLine('End of expansion phase')
self.logWriter.writeLine('\nPruning phase')
self.logWriter.timeMarker('pruning',False)
#remove up to all nodes above nodescount(actually less due to possibility that trees have less nodes than tis demanded to remove)
# of least effective nodes in terms of nodes per fitness, remove empty trees
if self.writeLog:
self.logWriter.writeLine('Pruning phase: initial nodes count =',self.currentNodesCount,'/',self.maxNodesCount)
_t = 0
while self.currentNodesCount > self.maxNodesCount:
reducees = helpers.proportionalrandom(self.treesPool,lambda x:x.numNodes/(x.fitness),max(self.currentNodesCount - self.maxNodesCount,0))
for tree in reducees:
if tree in self.treesPool:
if tree.pruneWorstNode():
self.currentNodesCount-=1 #if pruning fails, -1 node is done in
else:
_t+=1
self.removeTree(tree)
if self.writeLog:
self.logWriter.writeLine( _t,' trees removed in pruning phase')
self.logWriter.timeMarker('pruning')
self.logWriter.writeLine('End of pruning phase')
if self.writeLog:
self.logWriter.timeMarker('manageTrees')
self.logWriter.writeLine( 'OnAfterStep nodes count = ', self.currentNodesCount,', trees count = ',len(self.treesPool))
self.logWriter.writeLine( 'End of tree management.\n')
