neural network and inputs - page 40
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
So I'm thinking... how to interpret the result... to make it faster and certainly not by hand........
To choose: c, c++, Java, JavaScript, mql4, mql5
Yuri plz.
Can you tell me how to link MT4.5 with JAVA? I really want to try, at least to read something.
Yuri plz.
I want to use your mt4.5 in JavA and I've got no idea how to use it.
There are a lot of methods, but they are all cumbersome. Java was not designed for low-level applications, with the purpose of being platform independent. And MT lacks high-level communication channels that are not tied to plaform - Windows.
This is why I don't tie them and write code generators in Java, the results of which I can easily paste into MQL, C or Java.
I can't really understand the positivity of this thing yet. But during optimization, the total number of errors is not always counted correctly. And in this case, for some reason, at the end of the optimization it displays a message Bad Data!!! What could it be?
Bad Data means that the model is wrong in its 'predictions' more often than if we had not used the model. That is, the model is not fit for application. The reason: rubbish in the inputs. That is, the inputs are insignificant.
Let's take the results of testing on a control sample and compile a contingency table:
In this case the frequency of positive outcomes in the sample, i.e. if we take at random any sample from the sample, will be equal to (TP + FN) / (TP + FN + FP + FN)
If we use model predictions, the frequency of positive outcomes correctly predicted by the model will be equal to: TP / (TP + FP)
For the model to err less often in predicting positive outcomes for the samples than if we take arbitrary samples from the sample at random and treat their outcomes as positive by default, it is necessary and sufficient that the condition
TP / (TP + FP) > (TP + FN) / (TP + FN + FP + FN)
If the condition is not met, then the model is not suitable for predicting positive outcomes in practice, because if we take any arbitrary example from the sample and treat its outcome as positive by default, we will be wrong no less often than if we treat the positive outcomes predicted by the model.
I see and I understand that it is necessary to look for such set of input data that would significantly reduce errors and increase generalization ability..... And that's with enough data records ....
Exactly right. I.e. up to 1023 predictors (input data) can be crammed in at a time. After that, if no BadData message appears, then exclude from the sample (delete columns from the spreadsheet) those predictors which are marked as under reduction in the model description.
A sample of the remaining predictors can already be run again and will be subject to kernel transformations. And kernel transforms in turn will give increased generalizability.
Kernel transforms are automatically activated if the number of inputs (i.e. columns with explanatory variables) does not exceed 44. If the number of inputs does not exceed 10, then MSUA is included, which gives the maximum generalisability.