Neural evolution algorithms implementation (CNE, NEAT, HyperNEAT) #686
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -132,22 +132,26 @@ class NEAT { | |
| // Whether mutate or not. | ||
| double p = mlpack::math::Random(); | ||
| if (p > mutateAddLinkProb) return; | ||
| printf("addlink 1\n"); | ||
| // Select from neuron | ||
| ssize_t fromNeuronIdx = mlpack::math::RandInt(0, genome.aNeuronGenes.size()); | ||
| ssize_t fromNeuronId = genome.aNeuronGenes[fromNeuronIdx].Id(); | ||
| printf("addlink 2\n"); | ||
| // Select to neuron which cannot be input. | ||
| printf("genome num input is: %d\n", genome.NumInput()); | ||
| ssize_t toNeuronIdx = mlpack::math::RandInt(genome.NumInput(), genome.aNeuronGenes.size()); | ||
| printf("toNeuronIdx is: %d\n", toNeuronIdx); | ||
| printf("num of neuron is : %d\n", genome.aNeuronGenes.size()); | ||
| ssize_t toNeuronId = genome.aNeuronGenes[toNeuronIdx].Id(); | ||
| if (fromNeuronId == toNeuronId) return; | ||
| printf("addlink 3\n"); | ||
| // Check link already exist or not. | ||
| ssize_t linkIdx = IsLinkExist(genome, fromNeuronId, toNeuronId); | ||
| if (linkIdx != -1) { | ||
| genome.aLinkGenes[linkIdx].Enabled(true); | ||
| return; | ||
| } | ||
| printf("addlink 4\n"); | ||
| // Check innovation already exist or not. | ||
| ssize_t innovIdx = CheckLinkInnovation(fromNeuronId, toNeuronId); | ||
| if (innovIdx != -1) { | ||
|
|
@@ -159,7 +163,7 @@ class NEAT { | |
| genome.AddLink(linkGene); | ||
| return; | ||
| } | ||
| printf("addlink 5\n"); | ||
| // If new link and new innovation, create it, push new innovation. | ||
| LinkInnovation linkInnov = AddLinkInnovation(fromNeuronId, toNeuronId); | ||
| LinkGene linkGene(fromNeuronId, | ||
|
|
@@ -293,74 +297,80 @@ class NEAT { | |
| // If not, we will need to change CrossoverLinkAndNeuron, and Disjoint, and WeightDiff. | ||
| void CrossoverLinkAndNeuron(Genome& momGenome, Genome& dadGenome, Genome& childGenome) { | ||
| // Add input and output neuron genes to child genome. | ||
| printf("crossover 0\n"); | ||
| for (ssize_t i=0; i<(momGenome.NumInput() + momGenome.NumOutput()); ++i) { | ||
| childGenome.aNeuronGenes.push_back(momGenome.aNeuronGenes[i]); | ||
| printf("crossover 0.5\n"); | ||
| } | ||
| childGenome.NumInput(momGenome.NumInput()); | ||
| childGenome.NumOutput(momGenome.NumOutput()); | ||
| childGenome.GenomeDepth(); | ||
|
|
||
| printf("crossover 1\n"); | ||
| // Iterate to add link genes and neuron genes to child genome. | ||
| for (ssize_t i=0; i<momGenome.NumLink(); ++i) { | ||
| ssize_t innovId = momGenome.aLinkGenes[i].InnovationId(); | ||
| ssize_t idx = dadGenome.GetLinkIndex(innovId); | ||
|
|
||
| bool linkContainedInDad = (idx != -1); | ||
| double randNum = mlpack::math::Random(); | ||
| printf("crossover 2\n"); | ||
| if (!linkContainedInDad) { // exceed or disjoint | ||
| childGenome.AddLink(momGenome.aLinkGenes[i]); | ||
| printf("crossover 3\n"); | ||
| // Add from neuron | ||
| ssize_t idxInChild = childGenome.GetNeuronIndex(momGenome.aLinkGenes[i].FromNeuronId()); | ||
| ssize_t idxInParent = momGenome.GetNeuronIndex(momGenome.aLinkGenes[i].FromNeuronId()); | ||
| if (idxInChild == -1) { | ||
| childGenome.AddHiddenNeuron(momGenome.aNeuronGenes[idxInParent]); | ||
| } | ||
| printf("crossover 4\n"); | ||
| // Add to neuron | ||
| idxInChild = childGenome.GetNeuronIndex(momGenome.aLinkGenes[i].ToNeuronId()); | ||
| idxInParent = momGenome.GetNeuronIndex(momGenome.aLinkGenes[i].ToNeuronId()); | ||
| if (idxInChild == -1) { | ||
| childGenome.AddHiddenNeuron(momGenome.aNeuronGenes[idxInParent]); | ||
| } | ||
| printf("crossover 5\n"); | ||
| continue; | ||
| } | ||
|
|
||
| if (linkContainedInDad && randNum < 0.5) { | ||
| childGenome.AddLink(momGenome.aLinkGenes[i]); | ||
| printf("crossover 6\n"); | ||
| // Add from neuron | ||
| ssize_t idxInChild = childGenome.GetNeuronIndex(momGenome.aLinkGenes[i].FromNeuronId()); | ||
| ssize_t idxInParent = momGenome.GetNeuronIndex(momGenome.aLinkGenes[i].FromNeuronId()); | ||
| if (idxInChild == -1) { | ||
| childGenome.AddHiddenNeuron(momGenome.aNeuronGenes[idxInParent]); | ||
| } | ||
| printf("crossover 7\n"); | ||
| // Add to neuron | ||
| idxInChild = childGenome.GetNeuronIndex(momGenome.aLinkGenes[i].ToNeuronId()); | ||
| idxInParent = momGenome.GetNeuronIndex(momGenome.aLinkGenes[i].ToNeuronId()); | ||
| if (idxInChild == -1) { | ||
| childGenome.AddHiddenNeuron(momGenome.aNeuronGenes[idxInParent]); | ||
| } | ||
| printf("crossover 8\n"); | ||
| continue; | ||
| } | ||
| printf("crossover 9\n"); | ||
| if (linkContainedInDad && randNum >= 0.5) { | ||
| childGenome.AddLink(dadGenome.aLinkGenes[idx]); | ||
| printf("crossover 10\n"); | ||
| // Add from neuron TODO: make it a function?? check whether crossover is correct. | ||
| ssize_t idxInChild = childGenome.GetNeuronIndex(dadGenome.aLinkGenes[idx].FromNeuronId()); | ||
| ssize_t idxInParent = dadGenome.GetNeuronIndex(dadGenome.aLinkGenes[idx].FromNeuronId()); | ||
| if (idxInChild == -1) { | ||
| childGenome.AddHiddenNeuron(dadGenome.aNeuronGenes[idxInParent]); | ||
| } | ||
| printf("crossover 11\n"); | ||
| // Add to neuron | ||
| idxInChild = childGenome.GetNeuronIndex(dadGenome.aLinkGenes[idx].ToNeuronId()); | ||
| idxInParent = dadGenome.GetNeuronIndex(dadGenome.aLinkGenes[idx].ToNeuronId()); | ||
| if (idxInChild == -1) { | ||
| childGenome.AddHiddenNeuron(dadGenome.aNeuronGenes[idxInParent]); | ||
| } | ||
| printf("crossover 12\n"); | ||
| continue; | ||
| } | ||
| } | ||
|
|
@@ -518,7 +528,9 @@ class NEAT { | |
|
|
||
| averageRank = averageRank / speciesSize; // smaller is better. | ||
| speciesAverageRank.push_back(averageRank); | ||
| printf("size speciesAverage is %d \n", speciesAverageRank.size()); | ||
| } | ||
| printf("size speciesAverage is %d \n", speciesAverageRank.size()); | ||
| } | ||
|
|
||
| // Remove weak species. | ||
|
|
@@ -565,24 +577,30 @@ class NEAT { | |
| // NOTICE: how we organize different mutations is kind of flexible. | ||
| void Mutate(Genome& genome) { | ||
| // NOTICE: we can change mutate rates here. Randomly let mutate rates be bigger or smaller. | ||
| printf("mutate 1\n"); | ||
| // Mutate weights. | ||
| MutateWeight(genome, aMutateWeightProb, aPerturbWeightProb, aMutateWeightSize); | ||
| printf("mutate 2\n"); | ||
| // Mutate link. TODO: check link mutate implementation is correct or not. | ||
| double p = aMutateAddLinkProb; | ||
| while (p > 0) { // so p can be bigger than 1 and mutate can happen multiple times. | ||
| if (mlpack::math::Random() < p) { | ||
| printf("mutate 3\n"); | ||
| MutateAddLink(genome, aMutateAddLinkProb); | ||
| printf("mutate 4\n"); | ||
| } | ||
| --p; | ||
| } | ||
|
|
||
| // Mutate neuron | ||
| printf("mutate 5\n"); | ||
| p = aMutateAddNeuronProb; | ||
| printf("mutate 6\n"); | ||
| while (p > 0) { | ||
| if (mlpack::math::Random() < p) { | ||
| printf("mutate 7\n"); | ||
| MutateAddNeuron(genome, aMutateAddNeuronProb); | ||
| printf("mutate 8\n"); | ||
| } | ||
| --p; | ||
| } | ||
|
|
@@ -591,7 +609,9 @@ class NEAT { | |
| p = aMutateEnabledProb; | ||
| while (p > 0) { | ||
| if (mlpack::math::Random() < p) { | ||
| printf("mutate 9\n"); | ||
| MutateEnableDisable(genome, true, aMutateEnabledProb); | ||
| printf("mutate 10\n"); | ||
| } | ||
| --p; | ||
| } | ||
|
|
@@ -600,7 +620,9 @@ class NEAT { | |
| p = aMutateDisabledProb; | ||
| while (p > 0) { | ||
| if (mlpack::math::Random() < p) { | ||
| printf("mutate 11\n"); | ||
| MutateEnableDisable(genome, false, aMutateDisabledProb); | ||
| printf("mutate 12\n"); | ||
| } | ||
| --p; | ||
| } | ||
|
|
@@ -609,18 +631,31 @@ class NEAT { | |
| // Breed child for a species. | ||
| // NOTICE: can have different ways to breed a child. | ||
| void BreedChild(Species& species, Genome& childGenome, double crossoverProb) { | ||
| printf("breed 1\n"); | ||
| double p = mlpack::math::Random(); | ||
| ssize_t speciesSize = species.aGenomes.size(); | ||
| printf("species size is %d \n", speciesSize); | ||
| printf("breed 2\n"); | ||
| if (speciesSize == 0) | ||
| return; | ||
|
|
||
| if (p < crossoverProb) { | ||
| ssize_t idx1 = mlpack::math::RandInt(0, speciesSize); | ||
| ssize_t idx2 = mlpack::math::RandInt(0, speciesSize); | ||
| printf("idx1 is %d\n", idx1); | ||
| printf("idx2 is %d\n", idx2); | ||
| printf("breed 3\n"); | ||
| Crossover(species.aGenomes[idx1], species.aGenomes[idx2], childGenome); | ||
|
There was a problem hiding this comment. Is it possible to do a crossover using the same genomes? |
||
| printf("breed 4\n"); | ||
| } else { | ||
| ssize_t idx = mlpack::math::RandInt(0, speciesSize); | ||
| printf("breed 5\n"); | ||
| childGenome = species.aGenomes[idx]; | ||
| printf("breed 6\n"); | ||
| } | ||
| printf("breed 7\n"); | ||
| Mutate(childGenome); | ||
| printf("breed 8\n"); | ||
| } | ||
|
|
||
|
|
||
|
|
@@ -633,31 +668,46 @@ class NEAT { | |
| void Reproduce() { | ||
| // Remove stale species. | ||
| RemoveStaleSpecies(aPopulation); | ||
| printf("hehe 0\n"); // DEBUG | ||
| // Remove weak genomes in each species. | ||
| CullSpecies(aPopulation, aCullSpeciesPercentage); | ||
|
There was a problem hiding this comment. I'm not sure, is this a technique to avoid local minima? Truncating each species by 50% sounds like an counterproductive idea. Maybe I missed a reference in the NEAT paper? There was a problem hiding this comment. I referred to this implementation which seems performs good on Mario game: There was a problem hiding this comment. Thanks for the information, maybe it's a good idea, but it sounds counterproductive to me. I think it should work without, since RemoveStaleSpecies should remove old species anyway. There was a problem hiding this comment. My current plan is keep the functions, but when we do the Reproduce(), we need further decision about whether to use it. Organizing these steps in Reproduce() are flexible. There was a problem hiding this comment. Sounds good, since there is a parameter (aCullSpeciesPercentage) it's super easy to disable the function. |
||
| printf("hehe 1\n"); // DEBUG | ||
|
|
||
| // Remove weak species. | ||
| if (aPopulation.NumSpecies() > 10) { | ||
| RemoveWeakSpecies(aPopulation); | ||
| } | ||
| printf("hehe 2\n"); // DEBUG | ||
| // Breed children in each species. | ||
| std::vector<double> speciesAverageRank; | ||
| CalcSpeciesAverageRank(aPopulation, speciesAverageRank); | ||
|
There was a problem hiding this comment. Regarding https://groups.yahoo.com/neo/groups/neat/conversations/topics/2203 we don't use the adjusted fitness. There was a problem hiding this comment. So you mean, as we are calculation the average fitness of species, we doesn't need to use the adjustedFitness, because it likes we average it for twice, right? There was a problem hiding this comment. Yes, I think that's what Stanley is trying to explain. There was a problem hiding this comment. I now removed all adjustedFitness, even in the genome class. |
||
| printf("size speciesAverageRank is %d \n", speciesAverageRank.size()); | ||
| printf("hehe 3\n"); // DEBUG | ||
| double totalAverageRank = std::accumulate(speciesAverageRank.begin(), speciesAverageRank.end(), 0); | ||
| printf("size speciesAverageRank is %d \n", speciesAverageRank.size()); | ||
| printf("hehe 3-0\n"); | ||
| std::vector<Genome> childGenomes; | ||
| printf("hehe 3-01\n"); | ||
| for (ssize_t i=0; i<aPopulation.aSpecies.size(); ++i) { | ||
| // number of child genomes by this species. | ||
| printf("hehe 3-02\n"); | ||
| printf("size species is %d \n", aPopulation.aSpecies.size()); | ||
| printf("size speciesAverageRank is %d \n", speciesAverageRank.size()); | ||
| ssize_t numBreed = std::floor(speciesAverageRank[i] * aPopulationSize / totalAverageRank) - 1; | ||
|
There was a problem hiding this comment. That's the right position, I would go with offspring here. |
||
| printf("hehe 3-1\n"); // DEBUG | ||
| for (ssize_t j=0; j<numBreed; ++j) { | ||
| Genome genome; //!!!!!!!!!!!!! | ||
| printf("num of species is %d \n", aPopulation.aSpecies.size()); | ||
| BreedChild(aPopulation.aSpecies[i], genome, aCrossoverRate); | ||
| childGenomes.push_back(genome); | ||
| } | ||
| printf("hehe 3-2\n"); // DEBUG | ||
| } | ||
| printf("hehe 4\n"); // DEBUG | ||
|
|
||
| // Keep the best in each species. | ||
| CullSpeciesToOne(aPopulation); | ||
| printf("hehe 5\n"); // DEBUG | ||
|
|
||
| // Random choose species and breed child until reach population size. | ||
| while (childGenomes.size() + aPopulation.aSpecies.size() < aPopulationSize) { | ||
|
|
@@ -666,11 +716,13 @@ class NEAT { | |
| BreedChild(aPopulation.aSpecies[speciesIndex], genome, aCrossoverRate); | ||
| childGenomes.push_back(genome); | ||
| } | ||
| printf("hehe 6\n"); // DEBUG | ||
|
|
||
| // Speciate genomes into new species. | ||
| for (ssize_t i=0; i<childGenomes.size(); ++i) { | ||
| AddGenomeToSpecies(aPopulation, childGenomes[i]); | ||
| } | ||
| printf("hehe 7\n"); // DEBUG | ||
|
|
||
| // Reassign genome IDs. | ||
| aPopulation.ReassignGenomeId(); | ||
|
|
@@ -680,7 +732,9 @@ class NEAT { | |
| void Evaluate() { | ||
| for (ssize_t i=0; i<aPopulation.NumSpecies(); ++i) { | ||
| for (ssize_t j=0; j<aPopulation.aSpecies[i].SpeciesSize(); ++j) { | ||
| printf("start eval fitness\n"); | ||
| double fitness = aTask.EvalFitness(aPopulation.aSpecies[i].aGenomes[j]); | ||
| printf("end eval fitness\n"); | ||
| aPopulation.aSpecies[i].aGenomes[j].Fitness(fitness); | ||
| } | ||
|
|
||
|
|
@@ -706,7 +760,9 @@ class NEAT { | |
| // Repeat | ||
| while (generation < aMaxGeneration) { | ||
| // Evaluate all genomes in population. | ||
| printf("start evaluate\n"); | ||
| Evaluate(); | ||
| printf("end evaluate\n"); | ||
|
|
||
| // Output some information. | ||
| printf("Generation: %zu\tBest fitness: %f\n", generation, aPopulation.BestFitness()); | ||
|
|
||
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -51,7 +51,9 @@ class TaskXor { | |
|
|
||
|
There was a problem hiding this comment. I guess, using an arma::mat object to store the input is a better representation. This way, we can use all kinds of already implemented methods to preprocess the data. There was a problem hiding this comment. @zoq Reasonable. I will revise it then. |
||
| double fitness = 0; | ||
| for (int i=0; i<4; ++i) { | ||
| printf("start activate\n"); | ||
| genome.Activate(inputs[i]); | ||
| printf("end activate\n"); | ||
| double output = genome.Output()[0]; | ||
| //fitness += FitnissFunction::Error(output, outputs[i]); incorrect | ||
| fitness += pow((output - outputs[i]), 2); // TODO: revise. | ||
|
|
||
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think we have to make sure, that
fromNeuronIdx != toNeuronIdx.