Machine learning in trading: theory, models, practice and algo-trading - page 3506
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'm going to bed - wasting time discussing your desire to learn - well - I guess I don't have that kind of time.
Or you think you do.
Again using an obsolete term at will?) Reducing the choices for a split does not mean giving up greed (a local optimum is always chosen). And using a different criterion, "considering sustainability", does not mean giving up greed.
We are talking about standard metrics for evaluating a split, and the term relative to them is used because my approach is being compared to the generally accepted one - no need to take it out of context. It feels like you want to find something in the form, but the content is not interesting at all.
I'm talking about standard split evaluation metrics and the term relative to them is used because my approach is being compared to the accepted one - no need to take it out of context. It feels like you want to find something in the form, but the content is not interesting at all.
Giving common terms self-made meanings greatly reduces the possibility of understanding the content.
It turns out that a) thinning of possible points for splits is used (quantisation), b) on the set of points for splits thinned in point (a) a tree is built according to a custom criterion of "stability" (this is the darkest place, probably it is more correct to call it clustering), c) "stable" points for splits obtained in point (b) are used in catbusta to build the final working model.
I got the schematic.
Based on this statement " And not pulling split borders or whatever out of a trained catbuster. " - no.
Quantum splits are clusters of data.
If there is clustering in one dimension, then - yes, you can say so .
Then you do clustering of already existing clusters, you get like branches and leaves.
Where do I do this?
I have already written that this is hierarchical clustering.
I have already written about a number of disadvantages of this method...
because the algorithm is almost the same.
Can you describe the algorithm, what we take and what operations we perform? Maybe it's really as you say - algorithms can be different, from the ones I looked at - I didn't notice anything similar, but I couldn't look through all of them obviously.
you can make your further inferences.
Thank you.
Giving commonly accepted terms self-defined meanings greatly reduces the ability to understand the content.
It's a generally accepted term. But you have transferred it to the maximum of my custom FF for selection - however it is expressed there - i.e. you yourself have expanded a concept that in the context of wooden models is limited within the generally accepted concepts. It's not even this that is surprising, but the desire to identify and discuss it.....
It turns out that a) thinning of possible points for splits is used (quantisation), b) on the set of points for splits thinned in point (a) a tree is built using a custom criterion of "stability" (this is the darkest place, it is probably more correct to call it clustering), c) "stable" points for splits obtained in point (b) are used in catbusta to build the final working model.
Your understanding of the process is much improved. In (b) all candidates are selected according to a number of criteria (probability bias and number of activations - number of examples in the sample), selection from those selected for the split is done according to an additional criterion. For the catbuster, a quantum table is made of the quantum splits selected at different iterations from point b. There are variants there.
The term is a generally accepted term there. But you have transferred it to the maximum of my custom FF for selection - however it is expressed there - i.e. you yourself have extended the concept, which in the context of wooden models is limited within the generally accepted concepts. It's not even this that is surprising, but the willingness to identify and discuss it....
I won't get into arguments, I just recommend to read at least the wiki about greedy algorithms. Trees are always built by greedy algorithms.
Your understanding of the process has greatly improved. In (b) all candidates are selected according to a number of criteria (probability bias and number of activations - number of examples in the sample), selection from those selected for the split is made according to an additional criterion. For the catbuster, a quantum table is made of the quantum splits selected at different iterations from point b. There are options there.
Judging by that statement " And not pulling split borders or whatever out of a trained catbuster. " - no.
If in one dimensional clustering, then - yes, you could say that .
Where am I doing that?
Already wrote about a number of disadvantages of such a method....
Can you describe the algorithm, what we take and what operations we perform? Maybe, indeed, as you say - algorithms can be different, from those that I looked - similar did not notice, but I could not look through everything obviously.
Thank you.