Machine learning in trading: theory, models, practice and algo-trading - page 1182

[Maxim Dmitrievsky]

Ivan Negreshniy:

there CatBoost at iterations=100 trees instead of 10, and GBM is a beauty:)

LGBM doesn't work at all, as usual at Microsoft

and XGB by default

-------- XGB 100 Trees
 1.00*2.00=2.00 
 2.00*2.00=4.00 
 3.00*2.00=6.00 
 4.00*2.00=8.00 
 5.00*2.00=10.00 
 6.00*2.00=12.00 
 7.00*2.00=13.99 
 8.00*2.00=15.98 
 9.00*2.00=17.91 

what other boostings are there? :) later I will give them a more difficult task

[Yuriy Asaulenko]

Igor Makanu:

https://www.mql5.com/ru/forum/86386/page1180#comment_9543249

Yes, I've seen, and already looked at plugins, there are several of them. What's interesting is that chrome used to do this on its own, without installing anything else.

[Ivan Negreshniy]

Maxim Dmitrievsky:

LGBM doesn't work at all, as usual at Microsoft

and XGB by default.

what other boostings are there? :) then I will give them a more difficult task for the whole bunch at once

Maybe you can write an article on the topic of model comparison, I think it's very topical, and the name could be MO piscomer?)

[Maxim Dmitrievsky]

Ivan Negreshniy:
Maybe on the subject of model comparison you could write an article, I think it's very relevant, and the name could be MO piscomer?)

There is one article already hanging in moderation, an improved RL... then I want to make a normal feature selection, because so far this topic somehow passed me by. PCA there all sorts of things

then you can do something new purely in python, for example, add some boosting to already existing work, compare them...

[Ivan Negreshniy]

Maxim Dmitrievsky:

There is one article already hanging in moderation, an improved RL... then I want to make a normal feature selection, because so far this topic somehow passed me by. PCA there all sorts of things

and then you can do something new purely in python, for example, I can add some boosting to already existing work, for example, compare them

Improved RL with random learning or with batch learning as before?

P.S.
I think this is a very relevant article with comparative tests of EAs on a variety of modern MO models, I really think it is very relevant, can I suggest to make one on the basis of my engine?

[Maxim Dmitrievsky]

Ivan Negreshniy:

Improved RL with random learning or with batch learning as before?

P.S.
and as for the article with comparative tests of EAs on many modern MO models, it really seems very relevant to me, maybe propose to do on the basis of my engine, how do they agree on the topic now, in private to the admin or in the forum?

with batch as before, realtime also have projects, but not yet fully developed. I've sent it to Igor for the regression logit (like it quickly retrains), it works with Weierstrass functions, but it does not make money on quotients (yet)

Just write an article and that's it, send it for moderation... your topic will be accepted, I think :) you can write in the forum topic beforehand, ask Rashid

[Maxim Dmitrievsky]

Igor Makanu:

Igor Killed Chrome! I've been struggling for an hour, it doesn't work, I reinstalled it three times.... I was so worried when I looked for NeroSolutions... found it!!! - adventure...

Defender from vin10 so bad, three times rebooted, found something, left some of the trojans... Anyway, I haven't reinstalled Windows in three years, I'll have something to do (((((

This is Windows, baby, you must upgrade to MacOS

Sometimes it hangs tight and the screen of death is blue, but then everything is normal

[Maxim Dmitrievsky]

Igor Makanu:

Well, as for the subject, it's not about the type of MO or neural network (I don't even want to think about Pythons platform, etc.),@Maxim Dmitrievsky showed a lot of examples, but with all my respect all this is a one-sided implementation, here is a logit regression, here is a handmade regression... So what do we have in the output? - imho we have a "snapshot" of historical data and nothing more, that's why once again I want to find a software to visualize NS and interactively "twist", the above mentioned NeroSolution is able to do this, I used it long ago

i.e. the point is, anyway, you need to do as in the books - here is the NS, here are the inputs, and here is the output, which is fed to one of the inputs.... ideally, it should be a few ensembles of NS, then you get something adaptable

;)

PS: Chrome restored, fuh, CCleaner helped

you just do not understand yet the alchemy of grail production :) there should converge several random things, that's all

what needs to be visualized is already realized, at least partially

[Igor Makanu]

Maxim Dmitrievsky:

You just don't understand the alchemy of grail production yet :)

Yes, I do, because in this topic for a long time ;)

I don't need a grail, but a workable PBX would come in handy.

Maxim Dmitrievsky:

it's wino, baby... we need to switch to macOS

Maxim Dmitrievsky: For more than a year without problems, it hangs sometimes violently and the screen of death is blue, but then everything is normal

it is a blessing in disguise, it's high time to reinstall Win and do as I planned for long - install a virtual machine, there's a breeding ground for spam and other nasties, and if it crashes, it's one click to copy new hard drive to the virtual machine ;)

[Yuriy Asaulenko]

Igor Makanu:

I do understand, because I've been on the subject for a long time ;)

I do not need the grail, but a workable PBX would be useful

every cloud has a silver lining, it's high time to reinstall the Windows, I will reinstall and do as I had planned for a long time - install a virtual machine, it can be a breeding ground for spam and other filth, and if flies, the new hard drive in a virtual machine to copy a matter of 1-st click ;)

Comodo Internet Security and nothing flies for years.