Machine learning in trading: theory, models, practice and algo-trading - page 3607

New comment
 
mytarmailS #:
How do I trade on the minute? Turns out it's impossible, and I didn't know it.

You hold a position for one minute?

I'm writing about predicting on H1 and holding a position on H1.

 
СанСаныч Фоменко #:

I write about predicting on H1 and holding the H1 position.

If spread is important at such a forecast horizon, then 0 price for such forecasts.

People trade on minutes and even ticks and they don't care about spreads.

 
Maxim Dmitrievsky #:

The idea of the approach has already been described in the links. A dataset is grouped into similar clusters (patterns, if you like) by attributes. And labels for a given (subset_size) number of clusters are corrected, i.e. all labels become 1 or 0, depending on what is more in the cluster. This removes ambiguity for the final model, it stops overtraining for noise and making unnecessary splits.

Is the final model necessary? Maybe just become 1 clusters and trade? I assume the charts would be pretty good too.

 
Aleksey Nikolayev vertical lines) is interesting.

On gold, for example, such peaks are much weaker.

Perhaps it is time to add round price levels to round time levels)

Probably tens-hundreds of thousands of robots trading on H1, M30, M15 open/close positions and for a few seconds sort liquidity up and down from the current price. This increases the spread.
And different 14:15, 15:30... - time of important news releases. The same thing - they sort out liquidity with a large number of almost simultaneous trades in both directions.

 
Forester #:

Do we need a final model? Maybe we should just make 1 clusters and trade them. I assume the charts will be pretty good too.

I was wondering why the model
 
Maxim Dmitrievsky #:
I haven't done such measurements because you have to spend a lot of time on them to have reliable statistics. Some algorithms stand on reals, pluses.

Good curves are obtained after almost every training, depending on the settings of the algorithm. Training is very fast, it was a priority.

Periodically I come up with something new. If the new one turns out to be better than the old one, it means that I am moving in the right direction.

I like the fact that these TS are wooden (in the sense of oak) and trade often on hours. They are hard to break by spread or news. There is almost no need to follow them.

In general, in my case it does not want to work. I tried markup on each H1 bar - and different number of clusters - the necessary clusters are found not more than 1%, and the process is not constant and the shift of probability is not stable. Apparently, this is a matter of luck.

 
СанСаныч Фоменко #:

Here are the stats on the 35,000-bar watchmakers

Half have an increment of less than 4 pips and 75% have less than 9 pips in four digits.

If we take into account that the stop should be less than the profit, i.e. we need marks with increment over at least 15 pips, then it is impossible to build anything on hourly indicators.


Something weakly draws out on H2

And something can be discussed in terms of spread on H3

Only profitable trades are marked taking into account markups.

Starting from 1 bar, the passage forward and stop at the nearest bar, where there is profit. To markup the next deal, return to the initial position of the beginning of the markup of the previous deal +1 bar.

 
Forester #:

Do we need a final model? Maybe we should just make 1 clusters and trade them. I assume the charts will be pretty good too.

Trade, who's stopping you? I use a more flexible system that includes sets of clusters for the final model.
 
Aleksey Vyazmikin #:

In general, in my case it does not want to work. I tried markup on each H1 bar - and different number of clusters - the required clusters are found not more than 1%, and then the process is not constant and the probability shift is not stable. Apparently, it is a matter of luck.

So it is suitable only for my training system, which I have been working out for a couple of years. Or you're doing something wrong.
 
Maxim Dmitrievsky #:
Trade, who's to say you can't?
I'm not MO busy at all right now. So I won't test it soon. It would probably be enough for you to just test the idea with a small code adjustment. Of course, if it is interesting to compare with and without the final model.
1...360036013602360336043605360636073608360936103611361236133614...3625
New comment