Machine learning in trading: theory, models, practice and algo-trading - page 349

 
geratdc:

The main thing is stability. The main thing is stability. In less than a year - 800% and if it is true some kind of a self-training advisor on arrays similar to a neural network - I shake your hand.

I don't get it.

The point is that I do not count profit by deposit, but by the volume of the transaction. I bought, say, 1 lot - that's $100,000 - that's $100,000 profit.

We have 800% profit, divided by the leverage -100. We get 8% of profit of 100 thousand almost a year. 8% is the refinancing rate. We're playing with 100 grand. That's not much, imho.

 
Yuriy Asaulenko:

I don't get it.

The point is that I do not count profit by deposit, but by the volume of the transaction. I bought, say, 1 lot - that's $100,000 - that's $100,000 profit.

We have 800% profit, divided by the leverage -100. We get 8% of profit of 100 thousand almost a year. 8% is the refinancing rate. We're playing with 100 grand. It's not enough, imho.


1k initial deposit
 
Maxim Dmitrievsky:

I have an initial deposit of 1k

When calculating on the volume of the transaction, it does not matter what the deposit is, or even the specific volume of the transaction itself. The profit on transactions for the year is 8%.

As an example. I have 0.5% of a transaction per business day. The deal, say, 10,000 rubles. 200 days per year *0.5% = 100%/year profit from volume of transaction.

Now recalculate the leverage on the deposit, its load and the number of real contracts - we obtain the expected profit of a particular deposit. On Forts, the leverage is -4-5, the volume is N lots.

Certainly 800% looks good, but we get 8% per annum from a lot. Imagine that there is no leverage - then there is no point in playing these games. It's strange.

Probably the system does not work for me in Forex, because not to count from a deal - all some kopecks profits are obtained)).

 
Maxim Dmitrievsky:

By the way, you were looking for the most convenient grid - try this one https://www.mql5.com/ru/code/9002

I haven't learned it myself, please, let me know if it is useful or not, if I haven't manage to do it by myself.)


You should definitely use the one in standard mql5 code, in alglib library. There are code examples here -https://www.mql5.com/ru/forum/190948

The key word that appears in that thread is "BFGS", which means a special kind of grid training. This is when you don't need to spend days on learning speed and other related hyperparameters. You just feed in a learning matrix, and you're good to go. But this way of learning requires much more RAM, in deep networks for example it is not used because of this.

 
Maxim Dmitrievsky:


Nope, that's his original too, I didn't haggle

For 10 entries it would be already problematic to count even through the cloud.) But I'll try to make 3 such expert systems, applied to entry 4 ) If the opening prices are not tested for a very long period, it's okay.

Here we need to implement mathematics. You can train ReshetNNN by gradient descent, similar to the usual neuronics.

 
Dr. Trader:


You should definitely use the one that is now in standard mql5 code, in alglib library. There are code examples here -https://www.mql5.com/ru/forum/190948

The key word in that thread is "BFGS", which means a special kind of network training. It's when you don't have to spend days tweaking the learning rate and other related hyperparameters. You just feed in a learning matrix, and you're good to go. But this way of learning requires much more RAM, deep networks for example don't use it because of this.


You have to teach this one with a teacher... it's a big problem with outputs... )
 
Yuriy Asaulenko:

When calculating on the volume of the transaction, it does not matter what the deposit is, or even the specific volume of the transaction itself. The profit on transactions for the year is 8%.

As an example. I have 0.5% of a transaction per business day. The deal, say, 10,000 rubles. 200 days per year *0.5% = 100%/year profit from volume of transaction.

Now recalculate the leverage on the deposit, its load and the number of real contracts - we obtain the expected profit of a particular deposit. On Forts, the leverage is -4-5, the volume is N lots.

Certainly 800% looks good, but we get 8% p.a. from a lot. Imagine that there is no leverage - then there is no point in playing these games. It's strange.

Probably I don't have such a system in Forex, because not to count from a deal - all I get is some kopecks of profit)).


I don't think so.
 
Maxim Dmitrievsky:

In Forts without leverage it's OK, because in Forex without leverage it's not enough.
Forts without leverage does not happen.) In stocks, without leverage you can work.
 
Yuriy Asaulenko:
I don't trade there without leverage.)

I have an account at Otkritie, I've never even questioned my leverage there... I know how much a contract is and that's it... And I don't trade there, I just have some money... I don't know what the leverage is there, and I don't know how to trade there.
 
Maxim Dmitrievsky:

Maxim Dmitrievsky: I have an account at Otkritie, I've never been interested in my leverage there, I know how much a contract is and that's it... and I don't trade there, I just have some money... I don't think so... I don't know how to do it... If I wanted to trade with them, then I would probably choose the right one.

Pieces. There is no need to glue.

If you trade for 1-2 months of history on the minutes, it is enough for training.