Machine learning in trading: theory, models, practice and algo-trading - page 1036

New comment
 

AlgLib modification of training for sample partitioning using the R parameter.

The interpretation of the ready random forest itself in AlgLib is not different from the same Spark.

override protected def predictRaw(features: Vector): Vector = {
    // TODO: When we add a generic Bagging class, handle transform there: SPARK-7128
    // Classifies using majority votes.
    // Ignore the tree weights since all are 1.0 for now.
    val votes = Array.fill[Double](numClasses)(0.0)
    _trees.view.foreach { tree =>
      val classCounts: Array[Double] = tree.rootNode.predictImpl(features).impurityStats.stats
      val total = classCounts.sum
      if (total != 0) {
        var i = 0
        while (i < numClasses) {
          votes(i) += classCounts(i) / total
          i += 1
        }
      }
    }
    Vectors.dense(votes)
  }
The same division of the sum of tree predictions by the number of these trees.
 
Roffild:

The idea itself: throw some data from different indicators into a random forest and get one super-indicator.

Ways to find the grail require testing of unconventional ideas.

A questionable idea, or rather the resulting indicator will give questionable results

There is such a notion in information theory as reliability: "Reliability is a property of information without hidden errors", and if you have indicators as input data (you call them predictors), then these data do not always carry reliable information... and what result should the mathematical apparatus find?

Well, if you use indicators for training, which have positive expectation in the strategy tester, then why complicate the whole process - the strategy tester will show the result at once....

such a vicious circle

In essence it turns out that you are trying to take all the instruments from the dashboard of the car, mix up and then take their readings to calculate...? well let it be the angular velocity of the car - you know there is no such instrument, but here you want to find the angular velocity and period!!!

imho, either use mathematically sound formulas( from physics, math, geometry) and OHLC or all this manipulation is still that self-deception

 

The trained model is built on correlation points. It is possible to pile up all the indicators in one array. It all depends on the training points, which will be selected during data dumping. If the training points correlate well with one particular indicator, the model will predict this very indicator.

The strategy tester cannot output the priority values, as it is done when training the model.

 
Roffild:

The strategy tester cannot output priority values as it does when teaching the model.

The priority value can be set in the signal module:

  • Each market model is set a significance, measured from 1 to 100. The higher the value, the stronger the model.

I mean theWeight parameter https://www.mql5.com/ru/docs/standardlibrary/expertclasses/expertbaseclasses/cexpertsignal/cexpertsignalweight

I doubt that if you feed the machine learning with a mix of randomly chosen indicators the machine will find a mathematical model that describes the market condition

Your method is more like optimizing an Expert Advisor with a lot of indicators

 

Igor Makanu:

I don't think that if you feed a mix of randomly chosen indicators to machine learning, then the machine will find a mathematical model that describes the state of the market

Your method is similar to the optimization of an Expert Advisor with a lot of indicators

This method is not like the optimization, because in addition to the enumeration of indicator values they are compared to each other.

We can propose a theory for a long time. But why, when you can practically check it all?

 
Roffild:

This method is not like optimization at all, because in addition to the enumeration of indicator values, they are also compared to each other.

You can put forward a theory for a long time. But why, when you can practically check it all?

This is the right answer to all "I think", "I suppose", etc.

 

Hello!

Does anybody know a smart eksporter of quotes from mt4 to txt or csv file

I need to export just one instrument - eurodol, time frame, 5 minutes.

I need to export just one tool - Eurodol, 5 min time frame, 40-50k OHLCV candlesticks and update the file every 5 min with a new candlestick.

You can download the whole file every time,

or upload it once and then update it every 5 min by a new candlestick.

The second way is certainly faster.


I've found two exporters on the Internet

the first one Output History just won't start because it is old

(NZ_ExcelLive works and works fine, but when I set the output of 40k candlesticks the terminal hangs for ever))


Who can help? Let me remind you that I am a total zero in McMule

 
Igor Makanu:

A questionable idea, or rather the resulting indicator will give questionable results

There is such a notion in information theory as reliability: "Reliability is a property of information without hidden errors.", and if you have indicators as input data (you call them predictors), then these data do not always carry reliable information... and what result should the mathematical apparatus find?

Well, if you use indicators for training, which have positive expectation in the strategy tester, then why complicate the whole process - the strategy tester will show the result at once....

such a vicious circle

In essence it turns out that you are trying to take all the instruments from the dashboard of the car, mix up and then take their readings to calculate...? well let it be the angular velocity of the car - you know there is no such instrument, but here you want to find the angular velocity and period!!!

imho, either use mathematically sound formulas (from physics, mathematics, geometry) and OHLC or all this manipulation is still that self-deception

Information theory is certainly a good thing!!!! But AIs have gone further. They try to find reliable information in a stream of garbage. Or rather, to take from each indicator only what is reliable in combination with the same reliable information of other indicators. Unfortunately, it is impossible to find a reliable indicator for the market, in accordance with the theory of information. The indicator which will carry only reliable information. Naturally, IХO.

 
Igor Makanu:

Your method is something like optimizing an Expert Advisor with a lot of indicators

Not something, but many times more complicated and inefficient analogue

And he himself can't explain what he screwed up and why :)

 
mytarmailS:

Hello!

Does anyone know a smart extractor of quotes from mt4 to txt or csv file

I need to export just one instrument - eurodol, time frame, 5 minutes.

I need to export just one tool - Eurodol, 5 min time frame, 40-50k OHLCV candlesticks and update the file every 5 min with a new candlestick.

You can download the whole file every time,

or upload it once and then update it every 5 min by a new candlestick.

The second way is certainly faster.


I've found two exporters on the Internet

the first one Output History just won't start because it is old

(NZ_ExcelLive works and works fine, but when I set the output of 40k candlesticks the terminal hangs for ever))


Who can help? Let me remind you that I don't know a thing about McMule

I have an in-house one.


1...102910301031103210331034103510361037103810391040104110421043...3629
New comment