How do you practically assess the contribution of a "specific" input to the NS? - page 5
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
Generally speaking, not at all.
Quite right, regression is just a special case of NS.
Let me think I'll look on the internet for an appropriate definition/description of regression, and I stumbled across this, sharing a smile:
(Regression; Regression) ≈ a return movement of the libido to an earlier mode of adaptation, often accompanied by infantile fantasies and desires.
An intelligent man came along and gave an adult answer to my childish question.
By the way my post did not contain an assessment of your mental abilities
Not only the regression and NS are not quite the same thing, but the proposed option is at least not simpler.
Look carefully at the regression equation - it takes the results of NS and does not touch anything in NS. After all, the topic is a question on results, not on the NS arrangement, or am I missing something?
Well I have done it, but I have gone from the opposite, not took any inputs and made of them input combinations, and excluded inputs and some combinations and looked at the result - that in general is one and the same. Switch on, switch off - what's the difference? Due to the specifics of the implementation, I found it more convenient to exclude.
The difference is homegrown. What has been suggested is much richer and in particular your result, only the minimax search is already ready.
Quite right, regression is just a special case of NS.
What are you discussing? Let's evaluate the result of NS, which gives a bunch of inputs.
I do not have regression as a substitute for NS, see the equation
By the way, NS is also a regression. The same dependence of the current countdown on previous countdowns. But that's not the point.
What faa suggests applies to linear regression, while neural network is a non-linear regression.
I'm not suggesting linear regression - I don't know what it would be.
Again on my understanding of linearity in regression. I distinguish between linearity of variables and linearity of parameters. Non-linearity of variables is not considered a difficulty at all. The difficulty is the non-linearity of the parameters, which tend to be stochastic.
The difference is homegrown. What I suggested is much richer.
Ok, then let's go one at a time, shall we?
Doing a regression:
Profit = s(1) * A0 + ... with(n) * A(n)
We estimate the coefficients of this regression.
Look carefully at the regression equation - it takes the results of NS and does not touch anything in NS. After all, the topic is a question on results, not on the NS arrangement, or am I missing something?
How do we "do" regression? I have the NS doing classification. What is this "profit" and where do the coefficients c(1),...,c(n) proposed to be evaluated come from? Or are they just the weights of my NS? And the whole regression equation then is all my NS rewritten in "one line" with all non-linear transformations and all hidden layers as an equation which is incomprehensibly equal to what?
Ok, so let's go one by one, shall we?
I have an NS doing classification.
That aside - not touching, do's and don'ts.
Or is it just my NS weights?
Nothing to do with NS.
And the whole regression equation then is all my NS rewritten in "one line" with all non-linear transformations and all hidden layers as an equation that is incomprehensible to what is equal?
Regression has nothing to do with NS. We are interested in the result of NS in the form of profit/loss and inputs
How do we "do" regression? What is this "profit" and where do coefficients c(1),...,c(n) estimated come from?
Let's take an NS with n inputs, run it on some sample and get the result - a profit.
Shift the sample and get profits again. Get at least 30 profits. Then we use the least squares method to calculate the given coefficients.
I have the NS doing the classifications.
I do not know, just an idea, by the way if it will work, it is applicable to any TS with many inputs.
And so it is) And let me ask, if the regression equation does not correlate in any way with the work of NS, then why was it concluded that the inputs will behave the same, or at least be equally useful, when used differently? This transition requires at least some justification.
Once again we take a MACD with XYZ periods and obtain conditionally 0.5 coefficient and estimate that it adds +100 rubles to the moneybox of any trading system? And this conclusion is supposed to be drawn on only 30 training examples? And my NS has thousands of them and there may be contradictory examples, so how should we select them? And in the end of our analysis we will get a "misfit"?
All in all I got it. I do not need any more answers. After briefly discussing the question I was interested in, I solved the task in 30 minutes on the whole, having written 3 lines of code, excluding inputs and their combinations, your suggestion to my task is barely applicable, and pulls on a good thesis-diploma, or even a Ph.
And I apologize, for:
The clever man came and gave a grown-up answer to my childish question.)
By the way my post did not contain an assessment of your mental abilities.
It wasn't really an accident, it just happened to stick. Well, don't cross it out?) I'm sorry. I'm generally a peaceful, balanced, not angry, I love everyone, I control myself, I control myself, I control myself, I control myself, I control myself, I'm not angry, I love myself, I'm balanced........)