Machine learning in trading: theory, models, practice and algo-trading - page 1854
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
how come there is no OM archive on the exchange? There are some paid services there, maybe they sell
I found a strange feature in the behavior of different methods of error detection. Regression model. Learning with teacher y = 0 or 1. Titanic data.
oob RMSEr AvgEr AvgRelEr Train RMSEr AvgEr AvgRelEr 0.409762 0.333394 0.431368 0.406325 0.330199 0.430042 0.384805 0.293965 0.377680 0.380747 0.289936 0.377605 0.383891 0.286546 0.356373 0.370912 0.275152 0.358349 0.370671 0.261466 0.321370 0.353781 0.250322 0.326013 0.380332 0.259588 0.299060 0.337633 0.227993 0.296931 0.388428 0.248587 0.245976 0.309293 0.191325 0.249176 0.401764 0.237416 0.195562 0.273932 0.150077 0.195456
By Train all is as it should be, the stronger we teach, the less errors in all methods.
On OOV the RMSE error reaches 0.370671 and then starts to grow saying that overtraining has begun. But AvgEr and AvgRelEr keep dropping further. It seemed to me that all errors should behave synchronously, i.e., they should start to grow along with the RMSE.
Calculation formulas:
The calculations are very similar ( squaring (Sqr) and modulus of value (fabs) removes the sign ), but the behavior of the curves is different...
Repeated training with the classification model, it also has a minimum at the point where RMSE reaches 0.370671.
Has anyone encountered such behavior of errors? How can it be explained?
I found a strange feature in the behavior of different methods of error detection. Regression model. Learning with teacher y = 0 or 1. Titanic data.
By Train all is as it should be, the stronger we teach, the less errors in all methods.
On OOV the RMSE error reaches 0.370671 and then starts to grow saying that overtraining has begun. But AvgEr and AvgRelEr keep dropping further. It seemed to me that all errors should behave synchronously, i.e., they should start to grow along with the RMSE.
Calculation formulas:
The calculations are very similar ( squaring (Sqr) and modulus of value (fabs) removes the sign ), but the behavior of the curves is different...
Repeated training with the classification model, it also has a minimum at the point where RMSE reaches 0.370671.
Has anyone encountered such behavior of errors? How can this be explained?
And this is actually a topic when during optimization the error first decreases and therefore starts to grow. Purely in theory, if we assume that retraining begins at the beginning of error growth, then in fact it is necessary to stop training at the very moment when the error begins to grow or just before it. It is a question of the very method of determination. What methods of estimation will implement it is another question, but the fact of the beginning of error growth as a sign of the beginning of retraining is genius. I applaud standing up... Without a bit of sarcasm :-) Now I will think what will allow to make it happen...
This is a long-known method of determining overlearning, I think about 50 years, i.e. from the very beginning of neural networks and trees. See articles by V. Perervenko, for example.
For me the different behavior of errors is not clear in this situation. By RMSE we can determine the beginning of retraining, but not by others.
This is a long-known method of determining overtraining, I think, for about 50 years, i.e. since the very beginning of neural networks and trees. See articles by V. Perervenko, for example.
For me the different behavior of errors is not clear in this situation. By RMSE you can determine the beginning of retraining, by others not.
I found a strange feature in the behavior of different methods of error detection. Regression model. Learning with teacher y = 0 or 1. Titanic data.
By Train all is as it should be, the stronger we teach, the less errors in all methods.
On OOV the RMSE error reaches 0.370671 and then starts to grow saying that overtraining has begun. But AvgEr and AvgRelEr keep dropping further. It seemed to me that all errors should behave synchronously, i.e., they should start to grow along with the RMSE.
Calculation formulas:
The calculations are very similar ( squaring (Sqr) and modulus of value (fabs) removes the sign ), but the behavior of the curves is different...
Repeated training with the classification model, it also has a minimum at the point where RMSE reaches 0.370671.
Has anyone encountered such behavior of errors? How can this be explained?
It's like the SMA and EMA, AvrEr is more laggy - I take it - a numbers game.
That's the thing - there is no archive. It has long suggested that the developers include the OM in a copy that you can take from the history. But apparently there are some restrictions. As an option, I can ask the dude who made me an EA. He writes it regularly, but not for all symbols. I stopped writing it solely because of the poor performance of the indicator :-(.
Misha, I made you a video, I will show it soon.
Misha, I recorded a video for you, I'll upload it soon