Machine learning in trading: theory, models, practice and algo-trading - page 3006

New comment
 
mytarmailS #:
Is that an assertion?

When arbitrary permutations of increments are made, the structure of the price is lost and an SB is obtained. This is used, for example, in Monte Carlo, when the price is compared to a large number of SB variants obtained from it by permutations.

 
Aleksey Nikolayev #:

When arbitrary permutations of the increments are made, the price structure is lost and the result is a SB. This is used, for example, in Monte Carlo, when the price is compared with a large number of SB variants obtained from it by permutations.

Why are we talking about increments and not about absolute prices?

 
mytarmailS #:

And why are we talking about increments and not absolute prices?

Because initially we were talking about price movements, and a movement is first of all an increment, because if there is no increment, there is no movement.

But even if we talk about levels or tops, their arbitrary rearrangement is not a good idea, because this way you can get a fall from growth, etc.

 
Aleksey Nikolayev #:

I haven't thought about it, but I think it's unlikely, as the order of movements is important in prices.


The standard approach is to get the sample for training and the residual of the sample for checking .

 
Aleksey Nikolayev #:

Because initially we were talking about price movements, and a movement is first of all an increment, because if there is no increment, there is no movement.

But even if we talk about levels or tops, their arbitrary rearrangement is not a good idea, because this way you can get a fall from growth, etc.

you buy at the real price, your stop at the real price, your takeout at the real price...

if you come in at 1.55. and your stop is "2.05" and the market has to go to beat your stop, then if you mix up the indices at the row, nothing will change, the price of your stop will remain "2.05", it will just be in a different place ...

And if you make increments and mix them and then try to "collect" back, you will not save the information about your stop, it will be just a mess, white noise ....

Remember your article about gaps... the market will close the gap, but it can be today, or it can be in a week... but it will close.....

So it makes sense to ignore time, sequence .... only price (gap/stake/stop) which either will or won't.



This topic is about associative rules, which are essentially graphs,



There is a neuron that knows how to work with graphs, that's why I'm interested in this topic...

 
Maxim Dmitrievsky #:
I don't understand where such assumptions come from, and this is again working with traits rather than improving the way we learn. That is secondary.

Predictors also need to be dealt with, so I don't see the point of solving only one of the problems while abandoning the other. The method may reveal a fuzzy cyclicity in the predictor's probabilistic predisposition. I think so - we should check it.

There is an assumption that the models often pick up outliers. I'm thinking about how to see where on average the data in the model come from in the predictor range - to take into account the density distribution of the predictor's indicators.

Have you heard of such a metric, or maybe I have invented a bicycle again?

 
Aleksey Vyazmikin #:

Predictors also need to be worked with, so I don't see the point in solving only one of the problems, abandoning the second one. The method may reveal a fuzzy cyclicity in the predictor's probabilistic predisposition. I think so - we should check it.

There is an assumption that the models often pick up outliers. I am thinking about how to see where, on average, the data in the model come from within the predictor's range - to take into account the density distribution of the predictor's indicators.

Have you heard of such a metric, or maybe I'm cycling again?

I don't do scrupulous preprocessing ever, I have no enthusiasm for it :) if the signs are external, you need to choose them carefully. If they are derivatives of the original series, I don't see the point, just add a few variants.

I've already written above why all this doesn't work for fin BP. There the alpha is clogged with other uninformative examples from the rest of BP. You get memorisation rather than generalisation. And, on the contrary, strong regularisation also destroys TC, because not only uninformative but also good examples are extinguished, indiscriminately.

Cleaned up the redundancy. Long text is usually either desecrated or not understood by local weirdos :)

 
mytarmailS #:

you buy at the real price, your stop at the real price, your take at the real price.

if you come in at 1.55. and your stop is "2.05" and the market has to go to take out your stop, then if you mix up the indices at the row, nothing much will change, your stop price will still be "2.05" just in a different place ....

And if you make increments and mix them and then try to "collect" them back, you will not save the information about your stop, it will be just a mess, white noise....

Remember your article about gaps ... the market will close the gap, but it may be today, or it may be in a week ... but it will close ....

So it makes sense not to consider time, sequence .... only price (gap/stake/stop) which will either be or will not be.

It is not the price itself that is important for trading, but its changes (increments).

The increments are not necessarily taken at equal fixed time intervals. It is only important that the moment of the end of the increment is determined without looking into the future (Markov moment of time). Thus, gap closing is also an increment, inside of which there is another increment with the same beginning, opposite in direction and maximum in length, which is investigated in the article.

If there are several increments, their order is important. Roughly speaking, it is their order that determines what will work earlier - stop loss or take profit. That is why it is difficult to see the analogy with the set in the trading basket, which is filled in any order.

Shuffling of increments is necessary precisely to obtain the SB in order to compare with it and find price differences from it - only differences from the SB are traded. Mixing of products in the basket is not considered at all.

 
mytarmailS #:

This topic is about associative rules, which are essentially graphs,



there's a neuron that can work with graphs, so I'm interested in this whole topic...

Perhaps the analogy with a set of products can be useful in analysing the influence of news. IMHO, this model does not fit in with the analysis of a set of price patterns.

 
Aleksey Nikolayev #:

this model does not fit with the analysis of a set of price patterns.

Replace "milk," "cheese," and "beer" for "price," "price," "breakout," "pattern."

It'll all come together.

1...299930003001300230033004300530063007300830093010301130123013...3626
New comment