Machine learning in trading: theory, models, practice and algo-trading - page 1731

New comment
 
mytarmailS:

)))) I already have a caterpillar in my head)))

You can also cluster, also tried , you can also take two or three clusters, but there is already a "curse of dimensionality" few examples

Conditionally. 1-e 2 parameters hour and minute, then the quality score and cluster number 1, -1 (you can make more)

1 and 2 parameters can be any (indicator value, etc.), there can be more

You have states now, when the indicators are in some values, you know what strategy to apply. And this will persist for years, with luck

That's the shit I wrote based on your glues.

array([[ 0.        ,  0.        ,  0.        ,  0.        ],
       [ 1.        ,  0.        ,  0.92986122,  1.        ],
       [ 1.        ,  1.        ,  0.95419784,  1.        ],
       [ 1.        ,  2.        ,  0.9731913 ,  1.        ],
       [ 1.        ,  3.        ,  0.97264694,  1.        ],
       [ 1.        ,  4.        ,  0.98032991,  1.        ],
       [ 1.        ,  5.        ,  0.98203827,  1.        ],
       [ 1.        ,  6.        ,  0.98333045,  1.        ],
       [ 1.        ,  7.        ,  0.98232576,  1.        ],
       [ 1.        ,  8.        ,  0.98823056,  1.        ],
       [ 1.        ,  9.        ,  0.98317076,  1.        ],
       [ 1.        , 10.        ,  0.98527582,  1.        ],
       [ 1.        , 11.        ,  0.987291  ,  1.        ],
       [ 1.        , 12.        ,  0.98943567,  1.        ],
       [ 1.        , 13.        ,  0.99061927,  1.        ],
       [ 1.        , 14.        ,  0.9872359 ,  1.        ],
       [ 1.        , 15.        ,  0.94690234,  1.        ],
       [ 4.        , 28.        ,  0.91591829,  1.        ],
       [ 6.        , 53.        ,  0.94231733, -1.        ],
       [ 6.        , 54.        ,  0.96637643, -1.        ],
       [ 6.        , 55.        ,  0.96666239, -1.        ],
       [ 6.        , 56.        ,  0.95199225, -1.        ],
       [ 6.        , 57.        ,  0.95831038, -1.        ],
       [ 6.        , 58.        ,  0.94653855, -1.        ],
       [ 6.        , 59.        ,  0.9116516 , -1.        ],
       [ 9.        , 46.        ,  0.923074  , -1.        ],
       [10.        , 14.        ,  0.90585037, -1.        ],
       [10.        , 16.        ,  0.95169773, -1.        ],
       [10.        , 17.        ,  0.9080574 , -1.        ],
       [14.        , 18.        ,  0.90075409,  1.        ],
       [14.        , 19.        ,  0.90765833,  1.        ],
       [14.        , 37.        ,  0.96135539, -1.        ],
       [15.        , 42.        ,  0.90483235, -1.        ],
       [16.        , 56.        ,  0.91497814,  1.        ],
       [16.        , 57.        ,  0.90985375,  1.        ],
       [16.        , 58.        ,  0.93030191,  1.        ],
       [16.        , 59.        ,  0.92922849,  1.        ],
       [17.        ,  5.        ,  0.90549604,  1.        ],
       [19.        ,  7.        ,  0.90098716, -1.        ],
       [19.        , 27.        ,  0.90270895,  1.        ],
       [19.        , 28.        ,  0.93947857,  1.        ],
       [19.        , 29.        ,  0.9510106 ,  1.        ],
       [19.        , 30.        ,  0.91723467,  1.        ],
       [19.        , 31.        ,  0.90860176,  1.        ],
       [22.        , 16.        ,  0.93470451,  1.        ],
       [22.        , 17.        ,  0.95897063,  1.        ],
       [22.        , 18.        ,  0.95142488,  1.        ],
       [22.        , 19.        ,  0.92506725,  1.        ],
       [23.        ,  2.        ,  0.91332658, -1.        ],
       [23.        ,  4.        ,  0.93817147, -1.        ],
       [23.        ,  5.        ,  0.93106502, -1.        ],
       [23.        ,  6.        ,  0.94499223, -1.        ],
       [23.        ,  7.        ,  0.94141423, -1.        ],
       [23.        ,  8.        ,  0.9345415 , -1.        ],
       [23.        ,  9.        ,  0.92676692, -1.        ],
       [23.        , 10.        ,  0.94366438, -1.        ],
       [23.        , 11.        ,  0.95591665, -1.        ],
       [23.        , 12.        ,  0.96615139, -1.        ],
       [23.        , 13.        ,  0.91352493, -1.        ],
       [23.        , 14.        ,  0.93996186, -1.        ],
       [23.        , 31.        ,  0.93285471, -1.        ],
       [23.        , 32.        ,  0.97932234, -1.        ],
       [23.        , 33.        ,  0.97729135, -1.        ],
       [23.        , 34.        ,  0.9778504 , -1.        ],
       [23.        , 35.        ,  0.98064787, -1.        ],
       [23.        , 36.        ,  0.97368438, -1.        ],
       [23.        , 37.        ,  0.98026128, -1.        ],
       [23.        , 38.        ,  0.97319506, -1.        ],
       [23.        , 39.        ,  0.97524913, -1.        ],
       [23.        , 40.        ,  0.98469859, -1.        ],
       [23.        , 41.        ,  0.9807674 , -1.        ],
       [23.        , 42.        ,  0.97459257, -1.        ],
       [23.        , 43.        ,  0.96603776, -1.        ],
       [23.        , 44.        ,  0.94718308, -1.        ],
       [23.        , 45.        ,  0.93970739, -1.        ]])

I just put out a better score, but you can combine

on top of that the caterpillar will work (although you don't need it anymore)


 
Rorschach:

If at that moment the price went like that, the indicator could work very well, the period is almost stationary.


And no the main message was that you build a trend line (as an example) and which is almost identical for different parameters on the current bars it describes the current series well, but in the future it shows different results depending on the selected parameters, thereby shattering the myth that the market needs to remain cyclical to make money on the CSSA. The problem is that the optimization has to be done with lightning speed, not by hand. Firstly, it can be done more accurately, and secondly, with a small lifetime, when you pick it up, you don't need it, the train has gone. That's what I thought it would be great to be as round as possible to find this circle. You just forget that even though the history of the calculation does not seem great 0-250 bar itself calculation was time consuming and not always possible to find a predictive frequency, well, there is no it now for this kotir. Wait for it.
 
Maxim Dmitrievsky:

Conditionally. 1-e two parameters hour and minute, then the quality score and cluster number 1, -1 (you can make more)

And what was the clustering based on? Like what hour and minute?

And what is a quality score?

Or I didn't understand anything again????
 
Maxim Dmitrievsky:

the caterpillar can be used BEFORE the pretreatment, then you can cover yourself in Nutella and pour champagne on yourself

Oh you bastard Maxim, it was my idea :-)
 
mytarmailS:

And what was the clustering based on? Like what hour and minute?

And what is the quality score?

or did I get it wrong again????

by R^2 on the gluing of the sum of returnees - a metric. If it's falling, it's -1, if it's rising, it's 1.

Among all hours and minutes, yes. Then the same clusters can be glued together

hours and minutes as an example, we should try with the values of indicators

Z.I. If there are a lot of parameters, you can connect genetics, not a complete enumeration. But a lot of parameters is evil.

 

In fact, this is a ready-made trading strategy, which I have long wanted to try during automatic optimization, when we quickly and guaranteed get a circle and look at the behavior in the shortest possible future to make the final conclusions. So it is like this. It is possible to optimize even every hour, and it will still be in line with the main trend. Well, in general, if your interest is not fake in this question, I will be glad to help you to conduct tests and make final conclusions!!!!!!


In fact this is my childhood dream from the time of Neroshel remained unrealized.

 
Mihail Marchukajtes:

In fact, this is a ready-made trading strategy, which I have long wanted to try during automatic optimization, when we quickly and guaranteed get a circle and look at the behavior in the shortest possible future to make the final conclusions. So it is like this. It is possible to optimize even every hour, and it will still be in line with the main trend. Well, in general, if your interest is not fake in this question, I will be glad to help you to conduct tests and make final conclusions!!!!!!


In fact, this is my childhood dream from the time of Nerochelle, which has remained unrealized.

The only frequency analysis, which deserved my attention as the most adequate and not contradictory with one single proof in favor of his work. No big deal. If it is valid, the work will be a fairy tale. This is not a network, here is guaranteed to micro a small period, just a couple of bends and especially without the hassle. In any case, I think this is the maximum potential of frequency analysis.

Here and now it is guaranteed for literally not a long time.

Frequency analysis is not capable of more, because there are no fish in the frequencies at medium-long distances. Here, Max, you can not argue.

 
Mihail Marchukajtes:
Oh, you bastard Maxim, it was my idea :-)
What was it? To put Nutella on? 🤔
 
Maxim Dmitrievsky:
Which one? Putting Nutella on it? 🤔
No, the one dipped in brown chocolate :-)
 
Maxim Dmitrievsky:
Which one? Putting Nutella on it? 🤔

))))

1...172417251726172717281729173017311732173317341735173617371738...3548
New comment
Reason: