Discussing the article: "MQL5 Wizard Techniques you should know (Part 79): Using Gator Oscillator and Accumulation/Distribution Oscillator with Supervised Learning"
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Check out the new article: MQL5 Wizard Techniques you should know (Part 79): Using Gator Oscillator and Accumulation/Distribution Oscillator with Supervised Learning.
In the last piece, we concluded our look at the pairing of the gator oscillator and the accumulation/distribution oscillator when used in their typical setting of the raw signals they generate. These two indicators are complimentary as trend and volume indicators, respectively. We now follow up that piece, by examining the effect that supervised learning can have on enhancing some of the feature patterns we had reviewed. Our supervised learning approach is a CNN that engages with kernel regression and dot product similarity to size its kernels and channels. As always, we do this in a custom signal class file that works with the MQL5 wizard to assemble an Expert Advisor.
Our choice for a supervised learning model is a CNN that uses kernel regression and dot product similarity. Kernel regression estimates the outputs of each layer in a CNN by weighting the observed inputs in accordance to their similarity to a specific query point. This similarity is quantified using the dot product similarity as the kernel function. In our model, whose code is shared below in the next section, the _dot_product_kernel_regression method works out the similarity between all time positions in a feature map using the function torch.bmm(x, x.T) to which we add softmax normalization that is quite similar to self attention.
The dot product’s role is to be a data-adaptive measure of similarity, where a high product means two positions have the same pattern of activation across channels, which can lead to a greater influence in the weighted sum. An analogy, or different way of looking at this, could be like giving each position “votes” proportional to its agreement with the query position. A CNN layer can be thought of as a spreadsheet where the rows are time steps and the columns are features. In this setting, a single row is taken to represent a ‘position’. The ‘query position’ is therefore the current row that is being updated. In the process of performing these updates, comparisons are made between it and every other position to quantify how similar they are. These updates take place in Backpropagation.
Author: Stephen Njuki