Machine learning in trading: theory, models, practice and algo-trading - page 3586
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
I call a rectangle not a two-dimensional figure, but a multidimensional figure with dimension equal to the number of features (n-hyperrectangle). It is more convenient for me, so you should understand it as well.
.
It is also possible to search for a multidimensional figure, but it is more complicated. However, I find it doubtful to search on all predictors at once - even in theory I cannot imagine how it can be.
A pair of features can be connected by the same logic.
We were talking about trees that do nothing but cut the space of N features into N rectangles.
Suppose we do have some N-rectangle in the feature space in which one class strongly dominates the other. To select it with a tree we need to make at least 2N splits and get a total of 3^N (this is a huge number at large N) N-rectangles. Isn't it better to search for one N-rectangle at once?
The main idea is that the task of complete "mapping" of the feature space is too ambitious - the limited information contained in the sample is wasted. What if we limit ourselves to trying to simply cut out the "good" pieces from it?
the feature space is overly ambitious - wasting the limited information contained in the sample. What if we limit ourselves to trying to simply cut out the "good" bits from it?
.
Chewing the same gum....
In much deliberation of realisation came to euristic search, symvoyl regression and logical rules as the final product.
Faced with the problems of the curse of dimensionality (despite the large amount of data\quotations of good examples/patterns found 5-20 instances) and moral burnout :).
We were talking about trees that do nothing but cut the space of N features into N rectangles.
If you take the range of predictor values, you find the side of a conditionally multidimensional figure.
Let's assume that we do have a certain N-rectangle in the feature space.
As it seems to me, this is possible if the probability on the predictor scale changes smoothly, but I see in my predictors sharp changes in the probability shift. So, it all comes down to finding rows where the predictors showed values in a certain range and predominantly emphasised one class. And, it's easier to look for this through quantised cutoffs, once obtained, a grouping of these by the responses in the rows is carried out.
Even if you get what you intend, there is no guarantee of working on new data. It is necessary to dig exactly in the direction of searching for such guarantees.....
You're just standing still...do you want a business idea?
algorithms (or better principles of algorithms/formulas/NN) that are the most quickly and easily optimised by an optimiser are in high demand in the signals and market sections.
people need a sense of opportunity, you need money. You can find each other
The signals section has ordered a long life, and the marketplace section needs marketing to get through the pile of other marketing that has filled its top :) Right now it's clearly dominated by bots with rollbacks for 5 stars.
they're both in there, taking orders from each other :-)
just philosophical: there is no market without a speculator without a middleman, and we can see the proof of it in person.