Machine learning in trading: theory, models, practice and algo-trading - page 2155
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
Regarding the TF invariant normalization for the model ...
we take the series, we single out the important break points
we leave only the extrema points, all the rest is removed
normalize
now we take the distances between the breakpoints in the first series, create a new series from them and normalize too
thereby we obtain the normalized series, both by scales (amplitudes) and by time (frequencies)
All that is needed is to keep the number of extrema in the pattern even, everything else is normalized.
Thus, you can feed the model with data, no matter how many minutes or weeks, it will see it as the same thing, it will be invariant to the TF.
You may train one model for all TFs at once
=============================================
For those who haven't understood what it is and what it is for
For the model it will be the same pattern, because it is one and the same pattern
I am doing almost the same thing, only I have 100% time and points. But I don't understand how it is suggested here to normalize the time - by minimum distance?
I guessed it quickly ) but no one else...
Apparently I missed something?
Apparently I missed something?
Yes, the hyphen in the pronouns
I'll say it again, for the dumb ones.
There are points in the feature space. Some are to buy, others are to sell.
Suppose these points can move in such a way that the buy-sell sequence is not observed, i.e. the information about the spread in the dataset is lost.
the spread can be equated to the Euclidean distance between points, or between two classes of points
How to add this information. FNF, acceleration, and other stuff you can shove in your s. This is for clarity of perception, so to speak.
The constant bar does not work. there is a probability in both directions. the spread in the deal is always minus on the price.
zy worsen the conditions
Karoch read, read , read, read....
I still don't understand what you want to do with that spread (my brain is obviously out of shape today, or what you didn't say...
I didn't even get the essence of what you want to do and what for...
Karoch read, read , read, read....
I still do not understand what you want to do with the spread (my brain is obviously not in shape today, or what you did not say ...
I don't even understand what you want to do and what for...
Ask Valery, he began to catch up...
I find it hard to think of any other wording for itAlmost do so, only I have time and points 100%. But I do not understand how it is proposed here to normalize the time - by the minimum distance?
normalization 0-1 normalization
If in an already marked dataset with labels, you subtract or add a spread from the features, depending on the label, what effect would this have?
Would the feature space look better separable?
It is clear that this is done only for training.
In my article, I used exactly this approach - I separated the labels by a meaningful distance, which greatly improved learning. Usually we have labels replacing regression in essence, so the greater the deviation from zero (average?) the potentially greater the difference in traits - reducing noise by not considering small tees. But this is useful in the in/out classification and the triple classification of buy/sell/expectation. It is likely that the success of the approach also depends on the underlying strategy (formed or emerging). To be further studied.
In my article, I used exactly this approach - spread the marks over a meaningful distance, which significantly improved learning. Usually marks replace regression in essence, so the greater the deviation from zero (average?) the potentially greater the difference in signs - we reduce noise by not taking into account small tees. But this is useful in the in/out classification and the triple classification of buy/sell/expectation. It is likely that the success of the approach also depends on the underlying strategy (formed or emerging). Subject to further study.
I have only hard resampling with class separation in mind so far, but I think there are simpler ways
how did you do it, which letter should i read?
Yes, the hyphen in the pronoun
Added it :)