在单个项目上成功应用this遗传算法后,我想优化对象数组的值。
为了做到这一点,我尝试以下列方式使用forEach循环:
function getOptimalValues(req, res){
mdl.getItems({limit: 5}, function(err, items){//this function retrieves 5 items from database
items.forEach(function (elem) {
var queryObj = {
properties: elem,
targetEnergy: req.targetEnergy
};
genetic.getOptimalQuantities(queryObj, function (err, optimalValues) {
geneticResults.push(optimalValues);
if (geneticResults.length == items.length) {
res(null, geneticResults);
}
});
});
});
}
genetic.getOptimalQuantities定义如下:
function getOptimalQuantities(req, res){
var properties = req.properties;
var targetEnergy = req.targetEnergy;
var targetValues = {energy: targetEnergy, properties: properties};
var Task = require('genetic').Task;
var options = { getRandomSolution : getCandidateQuantities // previously described to produce random solution
, popSize : 100 // population size
, stopCriteria : stopCriteria // previously described to act as stopping criteria for entire process (set to 100 generations)
, fitness : getFitnessValue // previously described to measure how good your solution is
, minimize : false // whether you want to minimize fitness function. default is `false`, so you can omit it
, mutateProbability : 0.1 // mutation chance per single child generation
, mutate : mutate // previously described to implement mutation
, crossoverProbability : 0.3 // crossover chance per single child generation
, crossover : crossoverFunction // previously described to produce child solution by combining two parents
};
var t = new Task(options);
t.targetValues = targetValues;
//t.on('mutate', function () { console.log('MUTATION!') });
t.on('statistics', function (statistics) {
console.log('statistics',statistics.maxScore);
});
t.on('iteration start', function (generation) {
console.log('iteration start - ',generation)
});
t.run(function (stats) {
var dataObj = {quantities: stats.max, items: t.bestCombination}
res(null, dataObj);
});
}
在运行此过程时,我得到以下输出:
iteration start - 1
statistics 0.008126878121533886
iteration start - 1
iteration start - 1
statistics 0.007777620410591467
statistics 0.007777620410591467
iteration start - 1
iteration start - 1
iteration start - 1
statistics 0.008133385505205764
statistics 0.008133385505205764
statistics 0.008133385505205764
iteration start - 1
iteration start - 1
iteration start - 1
iteration start - 1
statistics 0.0093968469349952
statistics 0.0093968469349952
statistics 0.0093968469349952
statistics 0.0093968469349952
iteration start - 1
iteration start - 1
iteration start - 1
iteration start - 1
iteration start - 1
statistics 0.008431076204956763
statistics 0.008431076204956763
statistics 0.008431076204956763
statistics 0.008431076204956763
statistics 0.008431076204956763
iteration start - 2
iteration start - 2
iteration start - 2
iteration start - 2
iteration start - 2
...
对于传递给函数getOptimalQuantities的五个项目中的每一个,应该有100次迭代,并且每次迭代的每个项目的适应度值(作为统计数据)应该是不同的(所有元素的相等的机会接近于零)。因此,通过观察输出,我猜想遗传算法并不是针对本例中传递的5个项目中的每一个单独运行。
有谁知道如何确保功能(本例中的遗传算法)不会混淆多个输入的数据?直观地,更容易想象每个调用的顺序执行(这就是为什么要求顺序执行的原因),但通常任何建议的方法,顺序或并行都会非常有用。
感谢您的帮助。
尼科
答案 0 :(得分:0)
我相信这是你正在使用的库中的一个错误。它的Task
类继承自节点的EventEmitter
类,该类包含一个名为init
的方法,用于设置每个对象的侦听器哈希。但随后Task
会使用自己的init
方法覆盖此方法:
Task.prototype.init = function (callback) {
var self = this
self.emit('init start')
async.until(function () { return (self.population.length == self.popSize)}
, function (callback) { self.getRandomSolution(function (solution) { self.population.push(solution); callback() }) }
, function (err) {
if (err!=null) { self.emit('error', err) }
else {
self.emit('init end', self.population)
callback()
}
}
)
}
这意味着EventEmitter.init()
不会被调用,这意味着Task
的每个实例都没有自己的侦听器数据结构,而是使用相同的全局实例。当你只有一个Task
实例时,这样可以正常工作,但只要你同时处理多个实例,它们就会被添加到同一个事件监听器哈希中,并且只要它们中的任何一个发出事件,它们都会得到通知