Machine learning in trading: theory, models, practice and algo-trading - page 1114
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
Well, you know... I'm also deciding which model to put on by the look of the curve. What good is a model if it made 90% of profitable trades, and at key moments it was losing terribly. The type of the balance curve does matter. Of course, it will not be enough, but I will still have some idea.
How much data do you need for training????
If there is any correlation, the machine will find it.
The more data, the better.
it was the 10th year of mastering the Optimizer...
but happy people don't watch the clock))) hilarious...
Indeed, let's reason with....
All current metrics talk about the current quality of training and do not tell one bit about the future performance of the model. That is, getting a great result of the best metric does not guarantee quality results in the future. That's why I strongly believe that in addition to the training and validation period, there must be a testing period to assess the model's performance and decide if it is suitable. If you look at it through the percentage, the evaluating sample should consist of training/validation/control as 45/45/10 percent respectively IMHO, that's what I want to dovisetail in the optimizer. So that after model acquisition, when learning goes with a glance to the test section, iterative evaluation will take place and if the result is satisfactory, then the learning will stop with an opportunity to save the model, if control failed, then we start training again and so either until the desired result at the test section or after a certain number of training iterations. Now there is just one thing left to do.
We need to run the resulting model on a test, and if the test fails, then start training again, and so on 10 times. The problem is that the control section eats up the precious period of model operability, which begins to come to zero immediately after the period of training on the OOS. And I've come up with one way, which I told Doc a while ago, a couple of months ago, and he basically found it to be quite logical. Now I do not use it for certain reasons so I want to share it with you, if you don't mind....
If there is any correlation, the machine will find it.
The more data, the better.
Okay. I'll tell you right away that there's not a lot of it. I can make a giant unloading, but to do MT it will be long and I'm afraid will hang, so try to train it on what is. I'll do the preprocessing myself and post the already pre-processed data set. ok?
My targeting is fine, don't worry about it, and the optimizer is written in Java. You don't think it's possible to implement metrics as complex as you want???? please....
I think you gave me your data and from it I concluded that your predictors have nothing to do with the target.
Or am I wrong?
Can you start by proving that your predictors are relevant to the target?
If there is any correlation, the machine will find it.
The more data, the better.
You yourself divided into training and oos. In training you do everything you can do with a square. With what and how you did not have to tell, of course.
and I'll post the pre-processed data set. okay?
No, not ok, raw + preprocessed.
It seems to me that you gave me your data and from it I concluded that your predictors had nothing to do with the target.
Or am I wrong?
Can I start by proving that your predictors are relevant to the target?
All true, but since then I've made a number of key changes that have increased the amount of meaningful data. Especially since I was discounting the full set, and now after preprocessing the number of inputs is reduced and only significant ones remain. You can try again to evaluate.... if you are interested!!!
No not ok, raw + preprocessing.
Doesn't matter, let him prove first that he does NOT have noise.
It was the 10th year of mastering the Optimizer...
but happy people don't watch the clockSo I'm a programmer with a chicken paw, what do you want. I only studied eclipse for half a year in order to arrange the windows conveniently. I have no helpers ..... this forum is full of programmers, so everything is so slow :-(