Machine learning in trading: theory, models, practice and algo-trading - page 3741

New comment
 
fxsaber #:

Ran a 2025 on Rann on real ticks.

And this is LMAX.

Don't know the reasons why.


Still, I would recommend for thick-skinned MO to exclude the rollover hour from trading. On Ranne Tester on it often caught up to 500 pips in a few minutes. It doesn't happen like that.

 
So the model is an intraday trader




Then there is only one of two variants: either this moshnik is a dilettante and accidentally looked up the history in the code.

Or the flat in Eurobucks continues in the long term and then such trading should work on Eurofranc, Euro pound and a couple more symbols

And on any account



Write an article, or go to the local club of trillionaires.
[Deleted]  
fxsaber #:

Ran a 2025 on Rann on real ticks.

And looked in detail through TesterReport. It is hard there at the rollover hour, but I have never traded EURUSD at this hour, so I don't know how the execution is. Of course, only limit orders should be tested there.

We will adjust it further, the main problem with the speed of learning seems to be solved, now at least you can quickly get the TS (on the five-minute x it was around a minute for all passes).
 
Ivan Butko #:
So the model is intraday.
That's quite a pullback against the position. I would have pissed my trousers and closed the position a long time ago. It's obviously a robot trading.
[Deleted]  
fxsaber #:

Ran a 2025 on Rann on real ticks.

And looked in detail through TesterReport. It is hard there at the rollover hour, but I have never traded EURUSD at this hour, so I don't know how the execution is. Of course, only limit orders should be tested there.

It turned out that it finds "patterns" only in the rollover. Example from another pair. I don't like such demanding to the quality of TS execution.

In the tester on the forward also pulled out (from the beginning of 2025)


 




Gosh, I used to make such models 6-7 years ago on third-party software and trading at Alpari


The models were only useful at night, creating another night scalper.

The example above shows 20 trained models, all of which are night scalpers.




And this is a regular MLP with prediction of either the next candlestick or % rise/drop.

So, after 7 years and hundreds of articles, there is nothing new anywhere orita. Deleted his profile again in hysterics.

Don't worry, subscribers, he will restore it soon. This is his 5th profile deletion, I think.

 
Ivan Butko #:
Ilita. Deleted his profile again in hysterics.

Don't worry, subscribers, he'll restore it soon. This is the 5th time he's deleted his profile, I think.

Now why would he do that?

Previous times, at least he had a reason to frankenstein.

 
Maxim Kuznetsov #:

Now why the fuck would you do that?

the last few times, at least you had a reason to frankenstein.

It's got nothing to do with me anymore)))

He self-dumped before I woke up today.
 
Rorschach #:

ONNX is a storage format for neural nets and graph computations.

If you approach creatively, you can get visual programming on the basis of ONNX viewer.

Rorschach #:

ONNX could be an alternative for OpenCL. But that's just an idea for now.

Rorschach #:
The ONNX-formatted model is a Protocol Buffers file, which is a message file format developed by Google
Since GPUs have been brought in, the idea has become more interesting
1...37343735373637373738373937403741
New comment