Machine learning in trading: theory, models, practice and algo-trading - page 62
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 you want a grail... You train once and spend the rest of your life clipping coupons? Is that it? I`m surprised, what year have you been in the market? If it's no secret...... Do you know that the market is changing constantly and after some time the model just disappears, if you want the model to work for a long time, then switch to a monthly chart, in this case it is just right. And so, for 5 minutes the work in a week is quite normal result, then comes overtraining....While if two weeks, then in general class....
"I understand your concern," S.
I've been in forex for 8 years now, with no systemic profits, if that's interesting.
See:
I've ticked the box to mark the beginning of the forward. It has been going on for almost 6 years! The previous data was used for training. I am getting a profitable signal on such a long interval.
This is a satisfactory result (by 3). I will be fine-tuning it. If I get a 4, I can bet on the real. I am not afraid of seeing weak results, it only motivates me to improve, and I've been doing it for years.
I've never seen a result like that before. Show me a forward test of at least 5 years, at least 2,000 trades with a constant lot and no tricks in the tester.
I am promoting the idea that there is a single, constant signal in the market. And I am already catching it.
PS: if you want to make a periodic re-learning model, make a sliding forward, in other words forward gluing.
Everything is much simpler. I calculate the difference between the clone of the current signal and the clone of the previous signal, if this difference is positive, taking into account the direction of the signal and the difference is more than 100 pips, then the previous signal is a one, if less - zero.
"I understand your concern," S.
I've been in forex for 8 years now, with no systemic profits, if that's interesting.
See:
I've ticked the box to mark the beginning of the forward. It has been going on for almost 6 years! The previous data was used for training. I get a profitable signal on such a long interval.
This is a satisfactory result (by 3). I will be fine-tuning it. If I get a 4, I can bet on the real. I am not afraid of seeing weak results, it only motivates me to improve, and I've been doing it for years.
I've never seen a result like that before. Show me a forward test of at least 5 years, at least 2,000 trades with a constant lot and no tricks in the tester.
I am promoting the idea that there is a single, constant signal in the market. And I'm already catching it.
The main reason for that is the difference between the two types of correlations. By the way I'm in the market since 2004 and actively engaged in neural network trading. Even on neuroschel we made a gang on a closed forum....
So my point is, after the tick you go forward, and then there is a burning drawdown, and it is at this point you will doubt that the system will pull the deposit up again, so it's all bullshit. There are certain rules, the balance of the system should be at a 45 degree angle and grow steadily without sharp jumps and dips, then it is considered adequate. And the fact that Beter managed to raise the deposit in 2007 with the help of the NS, it is only because the NS caught the market rhythm and successfully entered it as they say lucky and nothing more, it was discussed when he reaped the rewards. Just like Reshetov's model can be trained successfully and it works for three months without a hitch, but it is all at the level of luck. After the championship, he could not repeat such a result ... So it's like this....
Hmm well good luck catching the true signal, I can not provide such tests, because I have to retrain the system every week. By the way, I'm in the market since 2004 and I'm actively involved in neural network trading. Even on neuroschel we made a gang on a closed forum....
So my point is, after the tick you go forward, and then there is a burning drawdown, and it is at this point you will doubt that the system will pull the deposit up again, so it's all bullshit. There are certain rules, the balance of the system should be at a 45 degree angle and grow steadily without sharp jumps and dips, then it is considered adequate. And the fact that Beter managed to raise the deposit in 2007 with the help of the NS, it is only because the NS caught the market rhythm and successfully entered it as they say lucky and nothing more, it was discussed when he reaped the rewards. Just like Reshetov's model can be trained successfully and it works for three months without a hitch, but it is all at the level of luck. After the championship, he could not repeat such a result ... So it's like this....
Quite original and really much more trivial than I thought it would be. Thank you!
Could you chew on your idea of making up a target variable a little bit? I don't understand it yet, but I found it interesting.
Well, everything has already been said there. For signals with a profit more than 100 pips we put 1 for all the others 0. If you read my post carefully, everything is clearly described, isn't it?
Correct me if I am wrong.
MMMM is not quite clear, but I will try to explain. Suppose the system gives buy and sell signals. For all signals that lead to a given profit or more I set 1 for the rest 0. I subtract the cloze of the current signal from the cloze of the previous signal, in the case of a buy signal and if this difference is greater than the specified one I put 1 for the previous signal, otherwise zero. The target signal is not used anywhere else, only to build a model, and then ... In the future, when the signal appears from the indicator the model says 1 or 0. Here is how it is implemented in my code, if that makes sense......
Where Maximum is the profit (I have it so named) In this example this code is from the sell signal....
LastI is the buffer index of the previous signal to write the value there. As if in the past, because now we do not know whether the signal will bring us profit or not, we can now say about the profit of the previous signal, but Reshetov's model is just the one that removes this shift and answers this question in real life right now. Will the current signal be profitable or not.... Whether the current signal belongs to the class of profitable signals or not. So no re-rating, if that's what you mean.....
MMMMM all kind of muddled I did not quite understand, but I will try to explain. Suppose the system gives buy and sell signals. For all signals that led to a given profit or more I put 1 for the rest 0. I subtract the cloze of the current signal from the cloze of the previous signal, in the case of a buy signal and if this difference is greater than the specified one I put 1 for the previous signal, otherwise zero. The target signal is not used anywhere else, only to build a model, and then ... In the future, when the signal appears from the indicator the model says 1 or 0. Here is how it is implemented in my code, if that makes sense......
Where Maximum is the profit (I have it so named) In this example this code is from the sell signal....
LastI is the buffer index of the previous signal to write the value there. As if in the past, because now we do not know whether the signal will bring us profit or not, we can now say about the profit of the previous signal, but Reshetov's model is just the one that removes this shift and answers this question in real life right now. Will the current signal be profitable or not.... Whether the current signal belongs to the class of profitable signals or not. Therefore, no re-rating, if that's what you mean.....
I did not say anything about any re-rating, do not twist my words.
I see, it means the time of transaction from signal to signal. Well, the rest is clear.
The same method, I've never tried it, so I'll shut up. The question is whether the signal system itself is adequate. I.e. why in those places, where there is a signal, we start watching, and not in other places.