Machine learning in trading: theory, models, practice and algo-trading - page 695

 
elibrarius:

I've been thinking about regression....
Regression in financial markets is not a smooth function, but rather a step function with a step of 1 pt. (both for the teacher and for the forecast). If, for example, we limit ourselves to a movement of +/- 100 pt, then there is an analogy with classification by 200 classes. That is, at the output we predict the most probable class - for example +22 pt.
Doesn't this mean that for good results the structure/complexity of the model (number of neurons) for regression should be 200 times larger? Well if you increase the step to 5 pt, then 40 times would be a little more economical at the expense of less accuracy.

No ideas on this topic?
 
elibrarius:
Any ideas on this topic?

You can't do regression on prices, do it on increments, then the number of variants will be less

You don't need more neurons, in fact you don't need a lot of them for regression... in linear regression you have 1 coefficient per trait :)

 
Maxim Dmitrievsky:

You can't do regression on prices, do it on increments, then the number of variants will be less

You don't need more neurons, in fact you don't need a lot of them for regression... in linear regression you have 1 coefficient per trait :)

I mean +/- 100 pt. increment, not price.

Well, the analogy with 200 classes is direct... Although they go sequentially and can still be smoothed from a step function into a smooth one

 
elibrarius:

The increment is what you mean by movement +/- 100 pt.

Well the analogy with 200 classes is direct...

Well, the classes also get more than one value at the output, it is simply divided by a sigmoid

 
elibrarius:
No ideas on this subject?

To classify into 200 classes, you need 200 neurons in the output, and correspondingly many neurons in the inner layers to make it all work somehow.

And for regression you only need 1 output neuron, what value it outputs will be Pronosis, "+22p, plus or minus error". And the hidden neurons will probably be enough for less.
The teacher will be discrete, with a number sign like a symbol. But the forecast is a regular noncontinuous double number with 16 digits, and so the estimate function (mean square error for example) will also be continuous.

 
SanSanych Fomenko:

Interesting article on a study of eight machine learning models

Unfortunate that the authors of the article took artificial prices instead of oil prices, for example. The results were obtained for ideal conditions, and it is not clear whether they will be applicable for real trading.

 
Mihail Marchukajtes:

Today is grail day, but we know what it looks like and how much work it takes to hold it in our hands!!!!

I'm not afraid of this word, but today I found the grail for myself. I did a lot of tests and the results were amazing. Special thanks to Dr. Trader for his support, which actually led to the discovery. I'm not afraid of this word........ With R I was able to effectively find a set of important predictors, and given that the target has the same number of classes, then playing it up a little (adding or removing one) set of important predictors can be expanded by one, two columns. I tried it once and it was just fine to add them. Then we start to tune in and select the model with maximal values of the training result.


Confused by the size of the polynomial is not great, but it will work in theory 50% of the training interval, that is a week, and this is enough for me!!!!!! But here's the thing.... And I now turn here to those people who are looking for reliable and stable patterns. It's easier to explain with an example.........

I keep a data table of 1000 rows and 111 columns where 110 predictors and so output. BUT I don't take the whole table, but take a small fresh section of 40 records (that's 2 weeks of TS work approximately) As a result I have a training set of size 40 by 110 plus the target. In fact, I take a slice of the market on this particular day at this particular interval. This slice is stationary. Further I make a choice of significant input variables in relation to the output and get from 3 to 5 columns, which, I understand, has the proverbial alpha allowing me to have an advantage over other market participants. And now the most important thing.... What was the point of all this discussion. As soon as I add one more line to the data table for training, the set of columns will change dramatically, that is, the alpha will run into another set of columns. Maybe not immediately, but after adding not one, but several rows. I.e. signals TS!!!! Alpha is exactly the same pattern in its purest form, which is minimal and sufficient for the target function. But this pattern is not explicit, that is, to see it with the naked eye is extremely difficult. It is at this stage and joins the AI and does its job.

And now imagine how alpha can jump on all data field, which I unload, if it is rarely contained in more than five inputs, and the total field of 110 inputs. In other words, with each new slice I get a completely different set of predictors. And how do you want to keep up with it, and even at a distance of a YEAR!!!!!!! if it is here for weeks, you can hardly catch it normally....... But you are absolutely right, the Grail exists, but everyone has his own, and in order to keep it you need to make a lot of effort.......

And again, referring to the theoreticians of demo accounts, this is how it is done.......

I've been working on the theory and did some tests with it. The tests showed good results. Models are trained UPU with the robot is loaded. Watch my signal this week and you will immediately see what my assumptions are worth.

I don't see what the problem is, you cluster the bars (to compare them reliably) and plot the recurrence statistics of some element, you can start with the most frequent one. Then build statistics of repetitions in pair with another element (better the second of the statistics, although you still have to check all of them), choose the maximum statistic, put the second point on the graph, and so on. As soon as the graph shows an inflection, this is the optimal length for this word. And so you check all the letters.

Get a set of words, of which will already be putting sentences together, here's a words can already be applied to the TC but first should assign the words coding for closeness to each other, although with this cope and the TC, there are also autoencoders after all. In short, at this stage there is freedom for imagination.

 

Can someone check the increment indicator? For some reason on the low TF there are holes in the drawing, and not only if you rewind... the indicator window becomes blank. Or maybe it's time for me to reinstall the terminal

Files:
loglog.mq5  5 kb
 

this is more correct

double pr2 = (pr!=0?log(pr):0);

 
Dr. Trader:

It is unfortunate that the authors of the article took artificial prices instead of oil prices, for example. The results were obtained for ideal conditions, and it is not clear whether they will be applicable to real trading.

Made on purpose to cover the theoretical variety of prices and explicitly call this variety.