How to form the input values for the NS correctly. - page 18
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
history is no longer responding to the surge :-)
I was talking about history - certainly the immediate one - but history
A huge request, let's not zigzag! Anything but zigzags, fractals and the like.
A huge request, let's not zigzag ! Anything other than zigzags, fractals and the like.
I'm not talking about a zigzag! What difference does it make what you use to find the pivot points!
---
GY,,,,,,,,the question is what to teach the net ?
----
if it just tends to give you +1 to hold the baja
and -1 to hold the bai and -1 to hold the sit.
then we need a point where it can switch it around
it is reasonable to switch at or after these points but at least in the vicinity of them
history is no longer responding to the surge :-)
I was talking about history - of course the nearest - but history
i meant that there are indicators that allow you to find these reversals
the question is what to teach? on the nearest history !
1 pivot points - i.e. change of trend
2 or something else?
I would rephrase the question differently:
What does the network learn better in "financial markets" in particular - discrete data (teacher outputs), i.e. "classification" or "continuous data", i.e. . everything else.
Hence, by the way, the 'architecture' - for ZigZag - one network and defined inputs, and for, let it be, 'Momentum' - other architecture and inputs.
'
Personally, I thought 'continuous data' was better.
I would reformulate the question differently:
What the network learns better in "financial markets" in particular - discrete data (teacher outputs), i.e. "classification" or "continuous data", i.e. . everything else.
Hence, incidentally, the 'architecture' - for ZigZag - one network and defined inputs, and for, let it be, 'momentum' - other architecture and inputs.
'
Personally, I thought 'continuous data' was better.
I would also rephrase:
what to teach!
---
here's the question: before you decide what to input, you need to know what you want to output.
and most likely the pros - all who work with networks design the network based on what is needed in the output!
and then decide what goes in!
---
if we want to switch networks at U-turns! then we have to teach the network to look for those U-turns!
but what to teach a network in the FINANCIAL MARKETS if not pivot points?
---
reversals are perfectly visible in some indicators.
it doesn't matter what they are ... the only important thing is that they must clearly mark these points
--
and the fact that one must learn NOT from 1999 data is probably understood by all
---
Before
deciding what to feed in, you need to know exactly what you want to get out
imho.
Yur, have a look at the mail ))))
...
---
If we want to switch the network on the pivots! then we have to teach the network to look for these pivots
but what is there to teach the network in the FINANCIAL MARKETS if not pivot points?
---
...
Not everything can be taught => a compromise must be found between "profitability" of a teacher and teachability to a network. And to come up with a "smooth teacher", i.e. a NS output that will help "identify these reversals", doesn't seem to me difficult. For example, "leading momentum" or "algorithm from page 5".
I see people are already stomping around, it's time to get high :)
IMHO, the first task (mine, anyway) is formulated as creation of two or three sets of "inputs" to be divided. By "input" is meant the moment when NS or another program of situation analysis is started. Each input should be applied to one and only one of the sets, for example to one of three: "buy", "sell" or "fence". There are other variations as well.
I am currently studying a scheme using two zigzags, a senior and a junior one. The input is the moment the segment of the junior zigzag is switched. For the case of the above three sets, the allocation of an input to one of them can be done by the distance (price) to the nearest next vertex of the senior zigzag. For example if this distance is greater than some minimum profit - it is buy or sell, depending on the direction of the current segment of the Senior ZigZ. If the distance is shorter, it is a fence. The result may look something like this
The main thing that I do not like in 2ZZ-scheme (this is my name for construction described above) is using of parameter (minimum profit). In fact, I'm doing it hoping to pick up some idea about it :) .
The main advantage - we get sets of realtime inputs, which contain quite a reasonable number of elements. And these inputs are related to the market situation. In contrast to the inputs at equal intervals of time (such as the opening of a new bar).
Concerning objections to AP, I will give a simple thought: One should not attach too much importance to particular constructions. For example by taking an integral numerically we can build rectangles and trapezoids. The essence of the process is not at all what we build, and the result in the limit will be the same. If we remove the zigzag lines from the figure above, we are left with just the inputs.
My personal opinion is that the zigzag as inputs to the NS is useless, and as an information compression too. It shows peaks, but in no way reflects the dynamics in between. Especially since it reacts to almost any spike, so, again, IMHO, screw it.
Of course, the zig-zag has zero informative value for the network.
Yur, check the post - there's no Troue there ))