Discussion of article "Neural networks made easy (Part 16): Practical use of clustering"

[MetaQuotes]

New article Neural networks made easy (Part 16): Practical use of clustering has been published:

In the previous article, we have created a class for data clustering. In this article, I want to share variants of the possible application of obtained results in solving practical trading tasks.

To evaluate the performance of the Expert Advisor, we tested it using the 500-cluster clustering model which we trained in the previous article and used in the previous test. The training graph is shown below.

As you can see, the training graph is quite smooth. To train the model, I used the Adam parameter optimization method. The first 20 epochs demonstrate a gradual decrease in the loss function, which is associated with the accumulation of momenta. And then there is a noticeable sharp decrease in the loss function value to a certain minimum. Previously obtained training graphs of supervised models, had noticeable broken lines of the loss function.

Author: Dmitriy Gizlyk

[Rodrigo Pagani]

2022.05.30 21:57:27.477 kmeans (WDO$,H1) 800 Model error inf

2022.05.30 22:00:23.937 kmeans (WDO$,H1) 850 Model error inf

2022.05.30 22:04:22.069 kmeans (WDO$,H1) 900 Model error inf

2022.05.30 22:08:04.179 kmeans (WDO$,H1) 950 Model error inf

2022.05.30 22:10:56.190 kmeans (WDO$,H1) 1000 Model error inf

2022.05.30 22:10:56.211 kmeans (WDO$,H1) ExpertRemove() function called

Como resolver este erro?

[Dmitriy Gizlyk]

Rodrigo Pagani #:

2022.05.30 21:57:27.477 kmeans (WDO$,H1) 800 Model error inf

How to resolve this error?

This is not a program execution error. This line displays the model error (average distance to the centres of the clusters). But we see inf - value beyond the accuracy of calculations. Try to scale the original values. For example, divide by 10,000

[Rodrigo Pagani]

Dmitry Gizlyk # :

Este não é um erro de execução do programa. Esta linha exibe o erro do modelo (distância média aos centros dos clusters). Mas vemos inf alem da precisao dos valores - valor. Tente dimensionar os valores originais. Por exemplo, divida por 10.000

I still couldn't find a solution.

[Rodrigo Pagani]

Rodrigo Pagani #:

I still couldn't find a solution.

data[shift] = NormaliseDouble((open - Rates[bar].low)/100,2);

data[shift + 1] = NormaliseDouble((Rates[bar].high - open)/100,2);

data[shift + 2] = NormaliseDouble((Rates[bar].close - open)/100,2);

data[shift + 3] = NormaliseDouble(RSI.GetData(MAIN_LINE, bar)/100,2);

data[shift + 4] = NormaliseDouble(CCI.GetData(MAIN_LINE, bar)/100,2);

data[shift + 5] = NormaliseDouble(ATR.GetData(MAIN_LINE, bar)/100,2);

data[shift + 6] = NormalizeDouble(MACD.GetData(MAIN_LINE, bar)/100,2);

data[shift + 7] = NormalizeDouble(MACD.GetData(SIGNAL_LINE, bar)/100,2);