Machine learning in trading: theory, models, practice and algo-trading - page 554
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
Do not forget that any data redundancy leads to a delay in the model's entry into combat. That directly affects the quality of the received signals after....
I personally chose the following methodology. I reversed the model obtained using Buy signals and tested it on the same part of the market, but for Sell signals. Thus, I don't lose precious time and adequately estimate the model possibilities. IMHO
By the way, yes. The control plot can be before the trace plot as well. I do the same, but I have a classification, so the order is not important. In the case of a forecast or regression, the order of the data is IMPORTANT.
But in my case with division into Buy and Sell is the most variant, because the reference section falls on the same market period as the training section. It's just that the data is diametrically opposed, but all the laws and news that are in effect at the moment are the same for both the training and control sections. In this case the trade starts immediately after the Traine section, which potentially increases the time of the TS.
Personally, I chose the following methodology. I flip the model obtained on Buy signals and test it on the same part of the market, but for Sell signals. Thus, I do not lose precious time and adequately estimate the capabilities of the model. IMHO
Why not teach a single model with Buy and Sell values at the output, for example from 1.0 to -1.0, while everything around 0.0 is a flat?
I want to try to train and test the model using classical method with train|valid|test. If the model works, then the test will be a real trade, i.e. I feed train|valid to NS and run a real trade on obtained settings. Or maybe one train is enough? So that the training will be close to the moment of real trading. As a last resort we can validate on data before the train plot.
They argue about the number and sequence of training, test, validation sections, but don't talk about cross validation, probably all use it by default or no one uses it...
When using cross validation I read that it is used when there is little data and it is necessary to at least povalidate on something. In forex, there is a lot of data - millions of minute bars for several years. I think no one uses it.
In cross-validation chiall that it is used when the data is scarce and it is necessary to at least something to povalidate on. In forex, there is a lot of data with millions of minute bars for several years. I don't think anybody uses it.
Why not teach immediately the same model with the values of Buy and Sell on the output, for example from 1.0 to -1.0, and everything above 0.0 - flat?
That's not... Not our method. By teaching buy and sell at once the control section disappears....
Nice idea and useful MT5 library, but it is rather troublesome to synchronize it with a Python script file, IMHO.
I think it is better to synchronize MQL variables with python ones directly via a local dictionary and execute Python code fragments directly from string constants in EA's code.
I've tried the test compiling bcc64 from command line and it works fine in Python 3.6:
It would be nice to add this functionality to your library, I was going to write my own library, but unfortunately for now I am busy with P-net library for python.
By the way, I wrote about this new neural network here in a branch, by preliminary results of tests on examples with Fisher's Iris it is trained by three orders of magnitude faster than DNN in TensorFlow, with equal test results.
I'll think about it. From my library, you can run any Python script and execute any function from it. The library is not complicated. I think it will be enough for me and many others.
I'll think about it. From my library, you can run any Python script and execute any function from it. The library is not complicated. I think it will be enough for me and for many others.
I guess so. The fact that the library will change is very likely. Now I have the task of mastering Python at an advanced level and mastering data science. All the time for this. And when the working model will be ready. I will already be thinking how to attach it to MT.
In a neighboring thread Renat wrote that they are going to add Python, R and C#. I don't understand if it will be possible to simply work in MetaEditor or integrate with MQL. I don't really need to work in MetaEditor, the code editors are enough. I think integration is more likely. That would be really cool. And my work on Python integration is a temporary solution.