Machine learning in trading: theory, models, practice and algo-trading - page 3712

 
Maxim Dmitrievsky #:
Dipsic one of the links hinted at the possibility of selling books electronically on ozone, for example on MO in trading. You can use this option to achieve great fame and power :)

Imho, not a bad idea if you combine this with creating a community of MO-trading) Then add a platform for MO-trading and thematic meme-coin) Not a bad chance to become rich.

Maxim Dmitrievsky #:
You can become more famous than Prado and Pugacheva.

And even more famous than Pugacheva wearing a Prado)

Maxim Dmitrievsky #:
Then create an investment fund in Dubai and a consulting agency to attract big money.
That's even better))
[Deleted]  
Aleksey Nikolayev #:

Imho, not a bad idea if you combine this with creating a MO trading community) Then add a platform for MO trading and a themed meme coin) Not a bad chance to get rich.

And even more famous than Pugacheva wearing a Prado.)

That's even better)
I need a specialist in creating meme-coins, I don't know about it :)
[Deleted]  
Maxim Dmitrievsky #:
One gets used to everything. Ideally, to the non-stationarity of the market too :)
I have not fidgeted with regression models before, but somehow it turns out to be effortless and quite good.

h ttps:// www.mql5.com/ru/articles/18603

Some brand new type of TS has suddenly appeared. It passes forward on EURUSD at the snap of fingers with some crazy number of trades, which can be filtered quite effectively.

Classification models did not pass such a forward well, because the trend direction of the pair has changed since the beginning of 2024.

I haven't tuned it properly yet.

Before filtering by thresholds

After


[Deleted]  

What a beauty this is. it's a little off at the end, but overall.


 
Maxim Dmitrievsky #:

What a beauty this is. it's a little off at the end, but overall.


the last graph is all test?

[Deleted]  
Evgeni Gavrilovi #:

the last graph is all test?

along with the training.

It's clearer that the forward is similar to the training. That's a good criterion for me.


 
Maxim Dmitrievsky #:

in conjunction with training

test c 2024? or all mixed up with shuffle 50%? what stock index is it?

[Deleted]  
Evgeni Gavrilovi #:

test c 2024? or all mixed up with shuffle 50%? what stock index is it?

Nothing shuffled, eurobucks. Test with 2024, yes. A turning point, good for testing.

But I modified the algorithm from the article already.
Thresholds solve in regression models. Whereas in classification models they don't. The main difference.

I will rewrite my bots from binary to regression.
 
Maxim Dmitrievsky #:

Nothing mixed, eurobucks. Test from 2024, yes. Turning point, good for tests.

There is a great way to test if overlearning is present. Specify learning rate = 1 and test different random seeds, in increments of 10, ranging from 0 to 100.

If all 10 runs show more or less the same graph, then congratulations, there is no overlearning.

[Deleted]  
Evgeni Gavrilovi #:

There is a great way to check if overlearning is present. Specify learning rate = 1 and check different random seeds, in increments of 10, ranging from 0 to 100.

If all 10 runs show more or less the same graph, then congratulations, there is no overtraining.

That's a bit off the mark. There is overtraining, it can't help but exist, but cutting off signals by thresholds gives a more or less even forward.