Need help! Can't solve the problem, I'm hitting hardware limitations - page 13

New comment
 
Candid:
Apparently one of the results of a passage with one date is a new date.

If the new date = the next sequence, fine. If it goes sequentially. Shit, taptology. I mean consecutive.

You can read the pack beforehand, they will be shallow.

And not even the next one, who cares. Index and call as needed.

 
Urain:

to Komposter: Andrei, if you're stuck on the dimension problem, it means you've made a mistake in formulating the problem.

There are three options here:

1 think about it yourself

2 open the problem in a public forum

3 solve the problem in private (for everyone who you think can solve it and trust to keep it secret).

Let me explain what I mean: if you save news, you can write thongs of the whole news, or you can do coding of typical phrases (compression), "account balance" turns into 1, "account equity" into 2, etc. Another variant of typical problem is desire to fill in data already sorted, for large dimensions this is death, it's easier to add to the end and do conditional sorting by indexes.

I think it's clear what I want to say by saying that there is an error in the problem statement.

I would ponder on this task
 
Since the problem is rather academic (like a job application question for a programmer) and many people have shown interest in it, why not formulate it more strictly in terms of input data description format, and everyone could generate 20 Gig of test data and present their practical solution?
 
I take it we are talking about a homemade tester/optimiser?
 
joo:
I take it we are talking about a homemade tester/optimiser?

No, it's something else.

i guess some broker/provider got the deal database. :)

 

I'll repeat the task in simplified terms

- We take a group of orders within M minutes (X+Y trades)
- take X first trades.
- calculate some criterion Kx (e.g. profit=100 or something).
- check the remaining Y deals of the group. If their calculated criterion Ky has a deviation no more than D from Kx, the order group fits us.

We do not know what to do with this group of orders and we probably will never find out, it's just insider information :)

Maybe this is a dealer who wants to lose clients, or maybe it's the other way around... a psychology study.

 
sergeev:

I will repeat the problem in simplified terms

- Take a group of orders for M minutes (X+Y deals)
- take X first deals.
- calculate some criterion Kx (for example, profit=100 or something).
- check the remaining Y deals of the group. If their calculated criterion Ky has a deviation of not more than D from Kx, then the order group fits us.

And we do not know what to do with this group of orders, and probably we will never find out, due to the insider info :)

Maybe this is a brokerage company that wants to drain clients, or vice versa...

Typical of a DB. But you have to aggregate the data... You can write in a separate table unique attributes of a sequence (c-point dates), the average profit value K and variance D and then look for the top 10 sequences that are close to the criteria you need. With indexes on these fields, the search will not take that long (even in a million records). Then, when you get the right 10 sequences, you can search through the source data, but it won't be a million searches anymore, because we have a date limit.

It is still a mystery - what is to be searched for? If we are talking about making a decision in terms of opening/closing an order, any processing of such a volume will take quite a long time.

The only way I see the effectiveness of such processing is to aggregate the data and use a probabilistic approach.

We can calculate correlation coefficients between what we have and the entire history of trades and "save" them in the robot, without using DB in the future.

There is another point. If we are talking about trades, maybe there is a reason to separate the trades for each symbol? And write robots of the same type designed for EURUSD, USDJPY, etc.

 
By the way, it's an interesting idea... Profitable deals can be generated as much as you want, by analyzing daily (High/Low) on minutes, and remembering the time of opening and closing prices (in a file). Then write an Expert Advisor that reads the file, and when the date and time matches, it opens/closes trades, and run it in the tester. I did that for a client who wanted to promote his robot's sales. And analyse already made trades in the same way )).
 
sergeev:

I think this is the only way to identify a strategy (or a set of parameters for a robot) that was used to trade in a given sequence.

It seems to me that in this way we can only identify the strategy that was used to trade (or a set of robot parameters) of a given sequence, and switch to it in a certain market situation.

 
marketeer:
Since the problem is rather academic (it looks like a question for hiring a programmer) and many people showed interest in it, why not formulate it more strictly in terms of description format of input data, and everyone could generate 20 Gigs of test data and present their practical solution?
I agree. The task is not trivial, interest is growing.
1...6789101112131415161718192021
New comment