具有反应性扩展的阈值检测

Threshold detection with reactive extensions

我在反应式扩展上花了一些时间后重写了这个问题 - 我可能知道现在非常危险所以请耐心等待并提前致谢。

我有以下检测级别的代码:

static class LevelExtensions
{
    public static IEnumerable<double> GetCrossovers(this double[] self, double x1, double x2)
    {
        return from level in self
               where level >= x1 && level <= x2 || level <= x1 && level >= x2
               select level;
    }

    public static IObservable<ThresholdCrossedEvent> ThresholdDetection(this IObservable<double> source, double[] thresholds)
    {
        return source
                .Buffer(2, 1)
                .Where(x => x.Count == 2)
                .Select(x => new { LevelsCrossed = thresholds.GetCrossovers(x[0], x[1]), Previous = x[0], Current = x[1] })
                .Where(x => x.LevelsCrossed.ToList().Count > 0)
                .SelectMany(x => from level in x.LevelsCrossed select new ThresholdCrossedEvent(level, x.Previous, x.Current));
    }
}

public class ThresholdCrossedEvent
{
    public ThresholdCrossedEvent(double level, double previous, double current)
    {

        Threshold = level;
        Previous = previous;
        Current = current;
    }

    public double Threshold { get; set; }
    public double Previous { get; set; }
    public double Current { get; set; }
    public Direction SlopeDirection => Current >= Previous ? Direction.Up : Direction.Down;
    public bool IsTouching => Current == Threshold || Previous == Threshold;
}


var feed = new double[] { 1, 2, 3, 5, 5.5, 6, 9, 12, 10, 9, 7.5, 6.5, 7 }.ToObservable();
var levels = new double[] { 5, 7, 8 };

feed
  .ThresholdDetection(levels)
  .Subscribe(x => Console.WriteLine(JsonConvert.SerializeObject(x)));

并且它按预期生成事件,到目前为止还可以:

{"Threshold":5.0,"Previous":3.0,"Current":5.0,"IsUpward":true,"IsTouching":true}
{"Threshold":5.0,"Previous":5.0,"Current":5.5,"IsUpward":true,"IsTouching":true}
{"Threshold":7.0,"Previous":6.0,"Current":9.0,"IsUpward":true,"IsTouching":false}
{"Threshold":8.0,"Previous":6.0,"Current":9.0,"IsUpward":true,"IsTouching":false}
{"Threshold":8.0,"Previous":9.0,"Current":7.5,"IsUpward":false,"IsTouching":false}
{"Threshold":7.0,"Previous":7.5,"Current":6.5,"IsUpward":false,"IsTouching":false}
{"Threshold":7.0,"Previous":6.5,"Current":7.0,"IsUpward":true,"IsTouching":true}

性能有异味吗?

在没有过早优化的情况下 - 是否还有其他任何尖叫性能问题(因为我将每秒处理 40,000 多条消息)?

我注意到我的 GetLevelsBetweenPair 水平交叉检测将依赖于 Linq.Where 每秒更新数千次,我的 levels 序列应该从低到高排序 - Linq 为我优化或我应该考虑的任何事情?

我将如何处理要在运行时更新的 levels 列表?

我在我的 feed 列表上调用 OnNext 从消息总线发布值,然后通过用户 interface/command 我 add/remove [=15] 上的一个元素=] 列表(它需要在修改后自动重新排序)。

鉴于不断检查这将是一个竞争激烈的数组(但根本不会经常更新),此级别列表更新过程应考虑哪些事项?

哪种结构允许以最有效的方式进行门控访问?

具有不同的 levels 列表并对其调用相同的订阅

我会有不同的级别“类别”,每个级别都创建相同的 ThresholdCrossedEvent,但意味着不同的东西,即 HistoricalThreshold 或 WeeklyThreshold。

作为多个订阅执行以下操作是否可以(最好),或者我应该将此阈值类型数据编码到警报中并将其从 double[] 更改为包含必要条件的 class 对象字段?

feed
  .ThresholdDetection(historicalThresholds)
  .Subscribe(x => Console.WriteLine("Historical " + JsonConvert.SerializeObject(x)));

feed
  .ThresholdDetection(weeklyThresholds)
  .Subscribe(x => Console.WriteLine("Weekly: " + JsonConvert.SerializeObject(x)));

您的实施看起来不错。几件事:

  1. 您还想将不断变化的 levels 建模为可观察对象。
  2. 我不是最好的 Linq2Objects 专家,但我认为 list/array 强制转换是不必要的并且对性能有轻微影响(如果您担心的话)。显然是测试。
  3. 我讨厌 Linq 语法,所以我在下面的示例中删除了它。我认为这不重要。

解决方案如下所示:

static class LevelExtensions
{
    public static IEnumerable<double> GetCrossovers(this IEnumerable<double> self, double x1, double x2)
    {
        return self
            .Where(level => level >= x1 && level <= x2 || level <= x1 && level >= x2);
    }

    public static IObservable<ThresholdCrossedEvent> ThresholdDetection(this IObservable<double> source, IObservable<double[]> thresholdSource)
    {
        return thresholdSource
            .Select(thresholds => source
                .Buffer(2, 1)
                .Where(x => x.Count == 2)
                .Select(x => new { LevelsCrossed = thresholds.GetCrossovers(x[0], x[1]), Previous = x[0], Current = x[1] })
                .Where(x => x.LevelsCrossed.Count() > 0)
                .SelectMany(x => x.LevelsCrossed.Select(lc => new ThresholdCrossedEvent(lc, x.Previous, x.Current)))
            )
            .Switch();
    }
}

这是对您的代码的直接改编:唯一的区别是外部 SelectSwitch 和输入。对于每个 levels 通知,新的 Select 创建一个新的 Observable。 Switch 表示总是切换到最新的,并丢弃旧的。因此,您始终使用最新的 levels 阈值。

我用 运行 测试了下面的代码:

var feed = new double[] { 1, 2, 3, 5, 5.5, 6, 9, 12, 10, 9, 7.5, 6.5, 7 };
var levels = new double[] { 5, 7, 8 };

var feedSubject = new Subject<double>();
var levelsSubject = new BehaviorSubject<double[]>(levels);

feedSubject
    .ThresholdDetection(levelsSubject)
    .Subscribe(x => Console.WriteLine(JsonConvert.SerializeObject(x)));
    
foreach (var d in feed)
    feedSubject.OnNext(d);

levelsSubject.OnNext(new double[] { 5, 8 });
Console.WriteLine("---");

foreach (var d in feed)
    feedSubject.OnNext(d);

它 运行 使用您的原始阈值通过该馈送阵列一次,然后我 运行 通过它并删除 7 个阈值。我的输出是这样的:

{"Threshold":5.0,"Previous":3.0,"Current":5.0,"SlopeDirection":0,"IsTouching":true}
{"Threshold":5.0,"Previous":5.0,"Current":5.5,"SlopeDirection":0,"IsTouching":true}
{"Threshold":7.0,"Previous":6.0,"Current":9.0,"SlopeDirection":0,"IsTouching":false}
{"Threshold":8.0,"Previous":6.0,"Current":9.0,"SlopeDirection":0,"IsTouching":false}
{"Threshold":8.0,"Previous":9.0,"Current":7.5,"SlopeDirection":1,"IsTouching":false}
{"Threshold":7.0,"Previous":7.5,"Current":6.5,"SlopeDirection":1,"IsTouching":false}
{"Threshold":7.0,"Previous":6.5,"Current":7.0,"SlopeDirection":0,"IsTouching":true}
---
{"Threshold":5.0,"Previous":3.0,"Current":5.0,"SlopeDirection":0,"IsTouching":true}
{"Threshold":5.0,"Previous":5.0,"Current":5.5,"SlopeDirection":0,"IsTouching":true}
{"Threshold":8.0,"Previous":6.0,"Current":9.0,"SlopeDirection":0,"IsTouching":false}
{"Threshold":8.0,"Previous":9.0,"Current":7.5,"SlopeDirection":1,"IsTouching":false}