Machine learning in trading: theory, models, practice and algo-trading - page 3052

New comment
 
I tried to write a simple q-table (without this super formula for calculating the action), simply - if the action has more points - then it is performed. If the action made a loss - the reward is equal to the loss.

I took 5 candlesticks and started to build states from them. The Close prices alone made 5 states (above/below/equal to the previous one).

Then I increased to 14 candlesticks and I have already... I think I have several billion states.

Okay, I'll try to see what's what. My method - the method of poking, thinking - too long it turns out.

The first run is the start button. Pfft. Nothing happened, I'm out.
Second run - Start button. The Grail. Year of the Grail. 5 years, 20 years of the Grail.
Started twice on each major, each major - 20 years of the Grail. I go back to the first major, Grail, nothing overwritten, everything saved.

I'm thinking, that's crazy.

Yeah. I'm printing the number of times each action was performed. And there's an average of five acts per act over 20 years.

That's a pretty good fit.

This is the fastest retraining I've ever done. I had to wait in the optimiser, but here it's just two clicks and it's done.
 
Aleksey Vyazmikin #:

Why do we need to take minutes at all? Do we have a signal generated on every bar? I don't know what kind of rules he was making up there - on returnees, I guess.

Either recognise that there can be no rules, and everything is chaos, and don't do it.

But if it is chaos, then no economic or more global events, such as the fall of a huge meteorite, can affect the quotes.

Perhaps, more elaborate modelling of price behaviour is required, rather than complete randomness. Let's say we take a dozen strategies and trade them as if creating quotes. From launch to launch, distribute different percentages of initial deposit to the strategies.

10-100 million - to make sure that there are no regularities in the random. At 1000-10000 something may happen, only because not enough variants have been added.
 

I have a problem that can be solved with MOE - looking for people willing to work on it for interest and to pump up their skills :)

The problem is solved using different sections of MO, including recognising graphical images, finding patterns in text - this is classification, plus tasks to determine approximate indicators - regression models.

The gist - there is a database of adverts, we need to determine the time a property was sold at a certain price (the price may change - the change is recorded). Probably other related metrics as well.

Interested in participating in such a project?

Added: I will provide my computing power.
 
Aleksey Vyazmikin #:

I am looking for people willing to work on it for the sake of interest and to pump up their skills :)

For the sake of interest it is more interesting to work on your own tasks. And not on someone else's, for the sake of interest.
 
mytarmailS #:
It's more interesting to work on your own tasks for the sake of interest. Not someone else's, just for the fun of it.

I propose a task that can be solved and from that get pleasure from work.

 
Aleksey Vyazmikin #:

I am offering a task that can be solved and enjoy the work.

You offer to do a real job for free to someone who doesn't need it, for someone who DOES need it....

It's the same as me calling you to do my repairs, and in return you get the pleasure of the task...

It's the same thing.

 

Aleksey Vyazmikin #:

I wouldn't equate intellectual and physical labour.

I don' t see any distinction at all, both are labour and both should be paid.

 
My offer to work together on the project is still valid. There is a target audience, which means monetisation is possible. However, I do not guarantee it. But I'm sure it will be interesting work with new experience.
 
Aleksey Vyazmikin regression models.

The gist - there is a database of adverts, we need to determine the time a property was sold at a certain price (the price may change - the change is recorded). Probably other related metrics as well.

Interested in participating in such a project?

Added: I will provide my computing power.

Just start a separate topic. There describe everything in detail.

Look, someone will be interested.

 
Aleksey Vyazmikin #:
I'm sure it will be an interesting job with new experiences.
:))
1...304530463047304830493050305130523053305430553056305730583059...3577
New comment