Machine learning in trading: theory, models, practice and algo-trading - page 311
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
Again, I tell it to those who want and try to understand, those who are in the subject, if only have something to say, so give out clever or at least on the subject!!!!
And what is the MO? ) I was too shy to ask
Machine Learning
Now I understand why you couldn't understand what Gerchik was talking about. The kettle is not boiling:).
The courses are too early to look through and review it, start with the wonderful book "The Stock Market Grail or Adventures of the Trader Pinocchio", this is a good starting point for your level.
All Padawans in blacklist, spammed already :)
You should at least read this book carefully, and then smear yourself and fill up the blaclists. I don't know what to do with it.
You should at least make a couple of profitable trades first and then give out tips... Or start with a cab driver, go all the way from nobody to the great Guru... Because that soldier is bad...
PS: Do you know where Mukhanchikov works?
What is MO? ) I was too shy to ask.
Machine learning is kind of like...
A couple of interesting conclusions that I've gotten over the last few months.
1) Classification or regression?
It seems to be regression. In my code examples I often took the increment of the next bar as a target for model training and rounded it to -1 and 1 (i.e. bar color, or price rise/decline), so that I could then use classification. Recently I compared results of training and prediction of different models both with and without rounding of a target (classification); I got better results with regression. But standard means of regression model evaluation such as R^2 did not suit me, I build a balance graph of trading and calculate the recovery factor to evaluate the model.
2) Estimation of the model using new data.
I somehow got used to the Expert Advisors from Market that with them it is possible to achieve almost perfectly growing line of means during optimization and to get similarly beautiful line up on new data. But this is an ideal case. In reality if the model is not good enough it will sometimes fail in trading and no optimization can fix that, the model simply does not understand some regularities of the market.
Here's an example of one weak but interesting strategy. In that example it loses on new data, but then suddenly begins to recover, but that is not the most interesting thing. Even more interesting is if you take the optimization window all the way to the end where the model was not able to trade that last long optimization time. Something happened in the market that went against this strategy and optimization cannot fix it.
This leads us to an interesting conclusion - in new data we should not expect the model to achieve ideal growth of funds upwards but to make the balance chart form at new data coincide with the new optimization chart at this data, it will mean that the model has caught some regularities of the market, but it is too simple and cannot take into account everything. Bad models and strategies will not have such a match.
Here is this example, optimization and fronttest
Now the optimization window has been moved to the end to the right, the dates on the chart are the same, but the horizontal scale is a bit jittery due to differences in the trades
The right part of both charts is very similar, despite the fact that in the first case it was new data for the model, and in the second case the mt5 optimizer spent about a day trying to achieve better trading in this area.
There's a lot of silence among the common people. Ahh I get it.... With the release of my article appeared a lot of what you need to check and try. That's why all are silent?
A couple of interesting conclusions that I've gotten over the past few months.
1) Classification or regression?
It seems to be regression. In my code examples I often took the increment of the next bar as a target for model training and rounded it to -1 and 1 (i.e. bar color, or price rise/decline), so that I could then use classification. Recently I compared results of training and prediction of different models both with and without rounding of a target (classification); I got better results with regression. But standard means of regression model evaluation such as R^2 did not suit me, I build a balance graph of trading and calculate the recovery factor to evaluate the model.
2) Estimation of the model using new data.
I somehow got used to the Expert Advisors from Market that with them it is possible to achieve almost perfectly growing line of means during optimization and get similarly beautiful line up on new data. But this is an ideal case. In reality if the model is not good enough it will sometimes fail in trading and no optimization can fix it, the model simply does not understand some market regularities.
Here's an example of one weak but interesting strategy. In that example it loses on new data, but then suddenly begins to recover, but that is not the most interesting thing. Even more interesting is if you take the optimization window all the way to the end where the model was not able to trade that last long optimization time. Something happened in the market that went against this strategy and optimization cannot fix it.
This leads us to an interesting conclusion - in new data we should not expect the model to achieve ideal growth of funds upwards but to make the balance chart form at new data coincide with the new optimization chart at this data, it will mean that the model has caught some regularities of the market, but it is too simple and cannot take into account everything. Bad models and strategies will not have such a match.
Here is this example, optimization and fronttest
Now the optimization window has been moved to the end to the right, the dates on the chart are the same, but the horizontal scale is a bit jittery due to differences in the trades
The right part of both charts is very similar, despite the fact that in the first case it was new data for the model, and in the second case the mt5 optimizer spent about a day trying to achieve better trading in this area.
As I said before it all depends on the input data. If the input data is the reason for the output, then the performance of the network on optimization and out-of-sample will be about the same. If the inputs are not, then the result will be significantly different. I have also done some manipulation here with building my models and the result is much improved, time will tell....... I hope Wazard continues to follow my signal????
That's spot on!!!! He is a good trainer mainly for beginners. He has knowledge, but his populism is in stereotypes. Trader, girls, expensive cars. Want to be like me? etc. In our case, the trader is a guy in underpants with unwashed face in front of the monitor. In his head a lot of formulas. Trading is a hell of a job. You know, all my friends and relatives have the impression that I just sit at the computer and do nothing. But if you think about it. I usually get up at eight, check the volumes, start building models-select up to 12 hours and then if the day does not hang :-(. Built a model - put.... I sit and monitor the whole day. It has gone the wrong way, nerves, etc. If you want to make money in the market, you have to work hard. I work hard and then.... Well you all have seen it :-). But I believe everything will be fine in the end!!!!
That's for sure. I work 12-14 hours a day. Well, you still need a distraction once in a while. Over the years, I've crooked my spine.
I don't believe Pepper either. Whoever sings beautifully is usually a liar. Go to smardlab there are a lot of such "ambitious" gurus.
But, still, success is possible and Larry Williams with 10K earned more than a million and it is officially documented, in the championship results and other people, such as Ed Secota.
They should make a film about Levermore, it would be much more interesting.