Interpolation, approximation and the like (alglib package) - page 4
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
Well, discrete points are selected, of course. And you can do it on an irregular grid. This is what makes the interp. convenient for transforming a series.
Why select discrete points if you already have a continuous function? The problem of approximating a mathematical analytically defined (continuous) function makes no sense.
Why choose discrete points when there is already a continuous function? The task of approximating a mathematical (continuous function) makes no sense.
the links there say why. And why exactly is it written to me too
the links there say why. And why exactly did it say to me as well?
It's all clear with you... goodbye))
Interpolation requires a data series, not a mathematical function.
Strictly speaking, a data series is a mathematical function. In a school course, good textbooks tell you this straight away.
Classical math function: takes value 1 at rational points and 0 at irrational points.
I apologise for the off-topic.
Dear Maxim,
If I am not wrong, then by using splines you are trying to feed the Mt5 screen price data in discrete packets to a neural network in which each segment or packet of price data will represent a separate function by itself and then, the neural network will choose the best function automatically for a specific price segment based on least mean square error (MSE) of past trained data. Am I correct in my understanding?
I mean you are trying a similar approach of game theory of feeding pixels to a game and in your case, you are trying to feed the price in the form of splines. Is that correct?
Thanks...
Hi, yes, you absolutely understood correctly. But I'm not sure about splines, because there are other ways:"inverse distance weighting", for example. But all about interpolation.
Strictly speaking, a data series is a mathematical function. In a school course, good textbooks tell you this right away.
Classical mathematical function: takes value 1 at rational points and 0 at irrational points.
I apologise for the off-topic.
Okay. So what's the correct way to say "analytically defined" then? Or should I say "given by an analytical expression"?
Hi, yes, you absolutely understood correctly. But I'm not sure abouts splines, because there are other ways: " inverse distance weighting", for example. But all about interpolation.
Ok, but are you sure that it is a necessary feed price using spline for a neural network?
I mean why we can't feed the open, close, High, Low prices of the candles directly to a neural network?
Why do you think that we need a function to define the price structure of a price segment and then, interpolate the prices again?
I am not sure if it is feasible in MT5 or not, but I am referring to an approach used in the game "ALPHA GO ZERO". So in MT5 we can feed the open, close, High, Low price of last 50 candles (example) to a neural network. Have you already tried this approach or it is not feasible for Mt5?
Can you please explain a little more why is it important to use a function or spline to feed to the neural netowork?
Ok, but are you sure that it is required feed the price using a spline to a neural network?
I mean why can't we feed the open, close, High, Low prices of the candles directly to a neural network?
Why do you think that we need a function to define the price structure of a price segment and then, interpolate the prices again?
I am not sure if it is feasible in MT5 or not, but I am referring to a computer screen feeding approach of "ALPHA GO ZERO". So we can feed the open, close, High, Low price of last 50 candles (example) to a neural network. Have you already tried this approach or it is not feasible for Mt5?
Can you please explain a little more why is it important to use a function or spline to feed to the neural netowork?
We just need to minimize cross entropy (or mutual information) beteen inputs and outputs by transforming the input information. It means that classifier will work better in a test subset and futher (better separation of points). Such techniques are widely used in machine learning.
But we are dont know apriori which transform will be better, so we just transform it iteratively and check model errors.We just need to minimise cross entropy (or mutual information) beteen inputs and outputs by transforming the input information. It means that the classifier will work better in a test of subset and futher. Such techniques are used in machine learning.
Well, I understood your goal what you are trying to achieve by using a different set of indicators for different segments of price as decided by the neural network based on minimal error of past trained data.
Obviously, in machine learning, it is very important to use the cross entropy and minimisation in order for the algorithm to converge over time instead of diverging from the goal.
There is already an article which uses automatic selection of strategies and I am not sure if you are aware of it or not. But it doesn't use machine learning. You can have a look if it can be of any help to you.
https://www.mql5.com/ru/articles/143
Well, I understood your goal what you are trying to achieve by using a different set of indicators for different segments of price as decided by the neural network based on minimal error of past trained data.
Obviously, in machine learning, it is very important to use the cross entropy and minimisation in order for the algorithm to converge over time instead of diverging from the goal.
There is already an article which uses automatic selection of strategies and I am not sure if you are aware of it or not. But it doesn't use machine learning. You can have a look if it can be of any help to you.
https://www.mql5.com/ru/articles/143
we dont care about indicators or anything else at this moment, end can use freamwork for any strategy and get the best result that can't be calculated analytically.