Machine learning in trading: theory, models, practice and algo-trading - page 1346
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
Learn the theorist not from Cicero's speeches, but from textbooks. Look, for example, Korolyuk recommended by you.
Why are you fixated on the theorist... Probably because this limits the range of your knowledge.
I see that you are not familiar with the concepts:
1) generalized spectra;
2) instantaneous spectra;
3) spectral structure of an unsteady process.
You do not know how nonstationary processes are analyzed. You don't even know how to approach it.
Look it up on the Internet, read it thoroughly. And don't say stupid things.
How do I make a normal sliding window for a feature?
Let's say price is a feature.When you go outside the price range, on the test sample, the model starts to stall, because it hasn't seen them before. At the same time, any cotier transformations take away important information. Suppose standardization and normalization over the entire sample does not solve the problem, since it will still be recalculated when going out of the range, moreover the training subsample is a small fraction of the whole chart. I'm not even talking about all sorts of oscillators and increments, it's not a feature, it's trash for NS.
Removal of spikes may help. There is some information about it in Perervenko's articles.
Of course the necessary information will be removed, but for example I don't care if the model will die because of 1000 pt or 2000 pt emissions. And so it is the end of such jumps.
Why are you so hung up on theorists... Probably because your knowledge is limited to that.
I see that you are not familiar with the concepts:
1) generalized spectra;
2) instantaneous spectra;
3) spectral structure of an unsteady process.
You do not know how nonstationary processes are analyzed. You don't even know how to approach it.
Look it up on the Internet, read it thoroughly. And don't say silly things.
For those random processes that radio amateurs deal with, these concepts can probably make sense sometimes. For prices, hardly.
Consider this with the example of the Wiener process, which has been considered an initial approximation for prices since Bachelier.
For those random processes that radio amateurs deal with, these concepts can probably make sense sometimes. For prices, hardly.
Consider this with the example of the Wiener process, which has been considered an initial approximation for prices since Bachelier.
1) Is this "For prices - unlikely" thing of yours sufficient and convincing for you???
2) You are stuck in the time of Bachelier.
zy
your - through the lip - dismissive "radio amateurs" does not add weight to you and does not do you credit, but only indicates your complete lack of knowledge in this field.
A dozen posts and not a single thought other than "you're all d..."
How do you do it?
divide the graph into many parts (levels), and when moving to another level, normalize prices only within it? I.e., the price range will always be normalized when going to another "level". Maybe there are some names for such methods.
Sounds like large size ranged bars. For example, take 100 p. (4-digit) rank-bars, start calculation from a circular level - we obtain a natural breakdown into 100 p. Inside this rank-bar we investigate the regularity of price movement - autonomously for each bar, but taking into account the large-scale trend.
But here again there appears the main question - what to trade - breakout or return? Almost every trader who comes to Forex first tries to trade with the swing, i.e. to use trend trade, and after the fiasco with the swing they switch to Bollinger - i.e. to return trade. The art of combining these methods, imho, shows how mature a trader is. Respectively, within these levels it is possible to trade in the trend, and it is possible to return from the edge to the center of the range. How to choose what is better in the current situation? Maybe the NS is able to do it.
I tried to train a neural network in Python. The package is scikit-learn, the NS itself is sklearn.neural_network.MLPRegressor. Neurons over 100, hidden layers -7, inputs -19, output - 1. The task is to predict a random process.
Have you checked what the booster can do with this data?
Can you post samples for CatBoost experiment?
How do I make a normal sliding window for a feature?
Let's say price is a feature. When you go outside the price range, on the test sample, the model starts to stall, because it hasn't seen them before. At the same time, any cotier transformations take away important information. Suppose standardization and normalization over the entire sample does not solve the problem, since it will still be recalculated when going out of the range, moreover the training subsample is a small fraction of the whole chart. I'm not even talking about all sorts of oscillators and increments - they are not features, but garbage for NS.
How can I divide a chart into many parts (levels) and, when moving to another level, normalize prices only within that level? I.e. the price range will always be normalized when going to another "level". Maybe there are some names for such methods.
again, all the "classic" transformations are not suitable for non-stationarity, so what I suggested above looks reasonable, at first glanceWhat don't you like about my variant with ATR of upper TF?
1) Is it your "For prices, hardly" that is sufficient and convincing for you???
2) You are stuck in the time of Bachelier.
zy
Your - through your lip - dismissive "radio amateurs" doesn't add any weight or honor to you, but only indicates your complete lack of knowledge in the field.
1) To paraphrase Tolstoy - every unsteady process is unsteady in its own way. Prices and signals are very different.
2) Modern financial mathematics began with Bachelier.
3) This is the result of observing the persistent attempts to unreasonably apply signal theory to prices.
1) To paraphrase Tolstoy - every unsteady process is unsteady in its own way. Prices and signals are very different.
2) Modern financial mathematics began with Bachelier.
3) This is the result of observing persistent attempts to apply signal theory unreasonably to prices.
:))) Alexius! You, I see - you know. However, I hasten to report - the market cannot be taken by theorists alone. For one reason - it doesn't take into account such a thing as "time". Yes, yes, ordinary time - seconds, hours, ... Without understanding the role of time and its consideration, all the power of TViMS is also powerless.