What to feed to the input of the neural network? Your ideas... - page 76
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
Evgeny Shevtsov#:
Where do I go to "chat-gpt", and even more so to "dipsic".
To be honest, I've never communicated with them (hmm... maybe I should start...).
Definitely give it a try, it's worth it. https://chat.deepseek.com/
Especially the reasoning part of the thread. Dip explains well as a tutor.
...
The concept is not "if else" but "similar" or "dissimilar".
...
You see, if you don't have a threshold, then you don't have a model.
You either accept the threshold, or there is no point in MLP and others existing.
There is no such thing as "I guess a missile fired from a neighbouring country right now will hit us. Or it won't. AI says 0.694875....
The final model is decision making.
And, if you have a threshold of 0.6, then your AI is a box with rules "if the input is 0.3 / 0.7 / 0.1 / 0.8, then the output is0.694875....". And no other way.
And if you don't have a threshold, you have a model of "I don't know what this set of numbers is, I have no idea what it's for".
But even if you don't know what the box is, it has a rule inside: ""if the input is 0.3 / 0.7 / 0.1 / 0.8, the output is0.694875..."(and there's no other way round it)
I have a similarly constructed noise indicator
This is the most meaningless noise indicator that has nothing to do with noise in time series. You have a dipsic network, ask him what noise is ))
That's an interesting option. I haven't tried it. It turns out that it requires a regression model. You can try metric and R square then, or a modification - to estimate the scatter of deviation from a straight line....
I consider the problem differently - I put a question for classification - whether the price will reach the SL or TP point, and a variant - whether the price will reach the SL before the market goes to a different state (the beginning of the trend) or not. It is important to put SL on something similar to the reference points, not just in pips.
The minus here is that the price can move in the right direction in the beginning for a third of time, and then slip in flat to the starting point. If we consider that the price moves from level to level, this approach looks strange, the models should be understood at once:
1. whether a significant level will be reached in the time interval
2. How long it will take to reach the level
3. What will happen after reaching it (depends on the answer to the second question - enough time for a pullback or a breakdown after a flat).
While when a significant level is reached, it is important to evaluate the movement up to this level.
In general - a different paradigm, which one is more correct - can be determined through experimentation.
What time interval do you use?
I would also note the disadvantage of using continuous sampling - getting similar examples on neighbouring bars, which can quantitatively outweigh in one or another direction, giving an incorrect probability shift, if you trade not on every bar later on.
Regarding R/S estimation, where the time interval T is directly involved as well :
I believe trading should be done in a completely similar way to the way the estimates themselves are generated.
That is, once the network (knowingly trained) has opened a trade, it should be forcibly closed after the time interval T, where the latter can be called the "prediction horizon".
Thus, the TP level is the place of forced closing of the deal, and the SL level should be placed purely for insurance and at such a distance so as not to harm the concept itself.
It is better to calculate the time interval T by integer candlesticks, because it gets rid of the problem of calculation from Friday to Monday, where there is a two-day excess of time, if T was calculated by the concept of time, not candlesticks.
The number of candles T must be substantially smaller (by a factor of several) than the image size X used to train the network.
Of course, thinking in terms of the profit factor (the ratio of the TP distance to the SL distance, as well as the ratio of the up move to the down move, or the ratio of the down move to the up move) as an estimate of a trade or price movement is more familiar and understandable.
However, the estimate calculated in this way can and will be significantly greater than one and up to infinity.
While for network training it is required to provide estimates in the range from -1 to +1, which is satisfied by R/S estimation.
But I repeat, R/S estimation is far from being self-sufficient, because it estimates the trajectory relative only to itself, but does not estimate the R-movement relative to some longer move than S.
the noise you and dipsic were talking about has nothing to do with prices.
What does it relate to?
1) To physical signals
2) To context:
I'm saying obvious things. But even those aren't in the "rubbish in, rubbish out" rubbish articles here. Even the context is not broken down, where, what and when to call noise and where, when and how, and most importantly - why so - it is necessary to get rid of it.
An approach no better than my creative poking.
A section of the chart where other "neurons" earn money.
The whole chart is a solid zone of pure signal. And it will become "noisy" only if someone breathes on it with cigarette smoke, breaks the screen, etc.
Rubbish is data that is poorly/incorrectly prepared for training
About improperly trained data.
The result of "correctness" in the context of earning what?
Part of another TC (unaccounted for)
A section of the graph where other "neurons" are earning.
The whole graph is a continuous zone of pure signal. And it will become "noisy" only if someone breathes on it, if the screen breaks, etc.