All about MQL5 Wizard : create robots without programming. - page 6

[Sergey Golubev]

MQL5 Wizard Techniques you should know (Part 31): Selecting the Loss Function

MQL5 wizard can be a test bed for a wide variety of ideas, as we have covered so far in these series. And every once in a while, one is presented with a custom signal that has more than one way of being implemented. We looked at this scenario in the 2 articles about learning rates, as well as the last article on batch normalization. Each of those aspects to machine learning presented more than one potential custom signal, as was discussed. The loss , also by virtue of having multiple formats, is in a similar situation.

[Sergey Golubev]

MQL5 Wizard Techniques you should know (Part 32): Regularization 

Regularization is another facet of machine learning algorithms that brings some sensitivity to the performance of neural networks. In the process of a network, there is often a tendency to over assign weighting to some parameters at the expense of others. This ‘biasing’ towards particular parameters (network weights) can come to hinder the network’s performance when testing is performed on out of sample data. This is why regularization was developed.

[Sergey Golubev]

MQL5 Wizard Techniques you should know (Part 33): Gaussian Process Kernels

Gaussian Process Kernels are covariance functions used in Gaussian processes to measure the relationships among data points, such as in a time series. These kernels generate matrices that capture the intra-data relationship, allowing the Gaussian Process to make projections or forecasts by assuming the data follows a normal distribution. As these series look to explore new ideas while also examining how these ideas can be exploited, Gaussian Process (GP) Kernels are serving as our subject in building a custom signal.

[Sergey Golubev]

MQL5 Wizard Techniques you should know (Part 34): Price-Embedding with an Unconventional RBM

We continue these series that explore various trade setups and ideas thanks to MetaTrade-5’s rapid development and prototyping environment with the MQL5 wizard. These articles, in principle, seek to explore how else traders can set themselves apart from the pack by exploring ideas that may not be so common and could deliver an edge to the interested trader, depending on how he chooses to exploit them. So, we are into exploring here, not necessarily exploiting and the reason why an edge matters a lot is many working trade ideas that are available tend to correlate too positively with each other.

This is super when the trends are bullish and everyone is in the green however, as many would agree, diversification is what would mitigate drawdowns when trends reverse and yet simply finding inversely correlated securities is a lot harder than it seems on paper. That’s why trade entries and exits that are specific to a trader could be a better haven than simply relying on commonly used setups. With that, this article looks at the Restricted Boltzmann Machines (RBMs) when implemented with Backpropagation, as opposed to their traditional implementations of Gibbs Sampling and Contrastive Divergence.

[Sergey Golubev]

MQL5 Wizard Techniques you should know (Part 35): Support Vector Regression

Support Vector Regression (SVR) is a form of regression derived from Support Vector Machines. At its core, SVR uses kernel methods to map input data into higher-dimensional spaces, allowing for more complex relationships to be captured, which contrasts with dimensionality reduction. For this article though we are exploring strictly its loss function role when used with a multi-layer perceptron. A related but different form of regression we looked at in an earlier article was Gaussian Process Regression. So perhaps it is key we start by drawing distinctions between the two.

[Sergey Golubev]

MQL5 Wizard Techniques you should know (Part 36): Q-Learning with Markov Chains

Custom signal classes for wizard assembled Expert Advisors can take on various roles, that are worth exploring, and we continue this quest by examining how the Q-Learning algorithm when paired with Markov Chains can help refine the learning process of a multi-layer-perceptron network. Q-Learning is one of the several (approximately 12) algorithms of reinforcement-learning, so essentially this is also a look at how this subject can be implemented as a custom signal and tested within a wizard assembled Expert Advisor.

[Sergey Golubev]

MQL5 Wizard Techniques you should know (Part 37): Gaussian Process Regression with Linear and Matern Kernels

We continue these series on the different ways key component classes of wizard assembled Expert Advisors can be implemented by considering 2 Gaussian Process Kernels. The linear-kernel and Matérn Kernel. The former is so simple you cannot find its Wikipedia-page, however the latter has a reference page over here. 

There are several additional advantages and uses to GPs like noise handling & reduction as well as adaptation to regime change and many others. As traders, though, we want to capitalize on these benefits, so let’s look at a very basic linear kernel.

[Sergey Golubev]

MQL5 Wizard Techniques you should know (Part 38): Bollinger Bands

Bollinger Bands are a popular technical indicator developed by John Bollinger in 1987 that consists of 3 lines (or data buffers). Its primary function is to find a way of quantifying market volatility by identifying over bought and over sold price points of traded securities. Bollinger Bands expand and contract due to market volatility, whereby if the volatility increases the two outer bands draw apart more, while if there is a contraction in volatility the two outer bands are drawn closer together.

[Sergey Golubev]

MQL5 Wizard Techniques you should know (Part 39): Relative Strength Index

In the last article, we started looking at Bollinger Bands in use within a custom signal class for wizard assembled Expert Advisors. We looked at up to 6 of the intended 8 possible setups that can be useful to a trader within a custom signal class in weighting long and short conditions. Because the article was running too long, the final two setups as well as a combined testing of all possible Bollinger Bands setups or patterns as we referred to them then. Let’s continue therefore by looking at pattern 6.

[Sergey Golubev]

MQL5 Wizard Techniques you should know (Part 40): Parabolic SAR

We stick to this format for this article, where we go over pattern by pattern for the parabolic SAR before concluding with a test run that combines multiple patterns as we did in the last articles. The parabolic SAR is computed almost independently with each new bar, since some of the parameters that go into its formula need to be adjusted, as we will see below. This trait, though, makes it very sensitive to price changes and trends in general, which in turn makes the case for its use within a custom signal class. For this article, we are going to explore 10 separate patterns of this indicator by testing each independently and then concluding, as in the recent articles, with a test run that combines a selection of these patterns.