Machine learning in trading: theory, models, practice and algo-trading - page 2620

 
He has his own website with articles not included in the books. Read the books
 
Maxim Dmitrievsky #:
What is there to count? 200 models in 5 minutes can be trained on a mac, it's like Intel 9
I realise the flaw, but I want a TC generator on the MO

I don't care if there are a thousand models, if they look at the last 10 candles it's useless, even if it's GPT-3 with all its guts.

The generator is there, the implementation is missing and the power ...

 
mytarmailS #:

No matter how many models there are, if they look at the last 10 candles, it's useless, even if it's GPT-3 with all the guts.


There's a generator, there's no implementation and there's no power...

It is possible to change the signs in the course of training, depending on the conditions, it will not be quite a squeeze window
 
Maxim Dmitrievsky #:
You can change signs in the process of learning, depending on the conditions.

it won't work, it's primitive...

No matter how many feathers you stick in a hedgehog, it won't become a bird.


There's only one way to do it: by learning the rules.

The rules aren't the ones in Forrests and Boosts, like X>0.1 & X2<0.003... they are approximating rules, they are as useless as any regression, it works only for table data, not for us


And rules called knowledge, event-driven rules, with the ability to be unrelated to indexes

event1...... event2...... event3.....

event1................event2...................event3

event1.event2......event3.


an event is a lower-order rule

a significant sequence of events is a full-fledged rule

a pool of complete rules is a TS.


This is the only algorithm you can use to reveal the TC of people on the market, and no other

Automatically generate and check full-fledged rules convenient "grammatic regression", it's an algorithm which uses a genetic algorithm to essentially write the code itself


Believe it or not, if you keep trading, you will come to this, it's just a matter of time.

 
mytarmailS #:

it won't work, it's primitive...

No matter how many feathers you stick in a hedgehog, it won't become a bird.


There's only one way to do it: by learning the rules.

The rules aren't the ones in Forrests and Boosts, like X>0.1 & X2<0.003... they are approximating rules, they are as useless as any regression, it works only for table data, not for us


And rules called knowledge, event-driven rules, with the ability to be unrelated to indexes

event1...... event2...... event3.....

event1................event2...................event3

event1.event2......event3.


an event is a lower-order rule

a significant sequence of events is a full-fledged rule

a pool of complete rules is a TS.


This is the only algorithm you can use to reveal the TC of people on the market, and no other

Automatically generate and check full-fledged rules is convenient "grammatic regression", it's an algorithm which uses a genetic algorithm to write the code itself


Believe it or not, if you keep trading, you will come to this, it's a matter of time.

If I understand correctly, it's quite a strong idea. About the "how" part, I just don't get the idea. You can bruteforce - like different sizes of windows, you mark what is called here the events. Then I have to somehow move it from the space of different windows to the space of events. I haven't tried it before, maybe it's not so hard in terms of calculations.

 
mytarmailS #:

it doesn't work, it's primitive...

No matter how many feathers you stick in a hedgehog, it won't become a bird.


There's only one way to do it: by learning the rules.

The rules aren't the ones in Forrests and Boosts, like X>0.1 & X2<0.003... they are approximating rules, they are as useless as any regression, it works only for table data, not for us


And rules called knowledge, event-driven rules, with the ability to be unrelated to indexes

event1...... event2...... event3.....

event1................event2...................event3

event1.event2......event3.


an event is a lower-order rule

a significant sequence of events is a full-fledged rule

a pool of full-fledged rules is a TS.


This is the only algorithm you can use to reveal the TC of people on the market, and no other

Automatically generate and check full-fledged rules is convenient "grammatic regression", it's an algorithm which, using a genetic algorithm, writes the code itself


Believe it or not, if you keep trading, you will come to this, it's just a matter of time.

So what are the rules to bind to, what to generate? Throw in a couple of examples, I'll pencil them in.
 
Maxim Dmitrievsky #:
So what are the rules to bind to, generate what? Just throw in a few examples, I'll pencil you in.
How I see rule generation, I'll text you later. I'm getting bombarded...

You're at least minimally familiar with grammatical regression.
 
mytarmailS #:
I'll let you know how I see the rule generation later. I'm getting a little bombarded here.

In the meantime, you should at least know a little about grammatical regression.
Oh, man, I'm sorry.
I'll read .
 
Maxim Dmitrievsky #:
There is a package for PythonPonyGE2, but I'm doing it on Pke, so I can't say what it does.
 
A sequence of events/rules is effective, but each rule has a dimension and a long sequence has curses.