Where is the line between fitting and actual patterns? - page 35
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
-this is also the first type.
I'm going to jump in, if you don't mind, Vitaly.
That's why Pardo's book was written. There are a lot of criteria there, it's not that simple...
But the most important criterion is known: FN must be somewhere inside some "stability area", i.e. the area in which small changes in parameters do not lead to a drastic change in the system profile.
I see that the main criterion has its own price - a tightly fitted graph formula will also fit the criterion. The "Stability Area" criterion allows you to choose a fit that behaves stably in some parameter area, which is intuitively acceptable. For me, the problem remains the same - did we manage to fit the parameters so tightly to the graph, or did we find the parameters of a real pattern?
-This is also the first type.
Why do we need two types of councillors? We will also have to write them. I think the result will be the same, and it will depend on how we prepare it.
IMHO we should use something based on "perseptron" from AI (Yuri Reshetov). It is transparent and understandable, and the Expert Advisor is able to classify simple patterns. There is plenty of space for "chasing", what else do we need? "What's the pack thinking?
Why? We wait for the same 5 candlesticks and close. Is it really necessary to open at a certain time?
->
paukas:
Okay.
5 candles. If the last one is larger than the previous 4, then above its high is a buy, below its low is a sell.
There is no criterion for no exit (and the second type requires an exit within a given time), the reverse combination may not happen again, and there is no time limit on the trade.
paukas:
And what is the distinguishing feature of the second one?
For the second type, TS have a limited time of trade and clearly know in advance the moment of the next signal arrival.
It's blurry, but it's not the kind of area where you can strictly divide everything into black and white. I actually understand the logic of such a division into types, but:
1. All this is very relative and pulled by the ear, the diversity of TC and principles used in them is much wider than their division into "good" and "bad" by trading time.
2. We, for our purposes, as far as I understood it, do not need it. I suggest to leave it to trader/developer to distinguish a "hopelessly "driven" TS from one that is able to look for patterns and use them. With some experience, it is easy enough to understand whether the TS has common sense or we are just playing with numbers. Let's consider that our TSs ;) are capable to discover regularities when used correctly. It is also suggested to consider that these patterns actually exist.
----
In the context of the branch I see our task as purely practical: to understand how to find with the highest probability a set of parameters, set, FC or whatever else, in which TS will use the patterns that it is built to use. This is relevant to me. There are 2 points here too.
а. How to train? (What to train on? How much to train on? etc.)
б. How to filter? (analysis of results).
----
So far there is no agreement :) Here are some people think for example that OOS steals profits, others have contrived to put OOS in the training sample and as an explanation draw such schemes, that half a litre of cognac did not help me understand, how can the control sample can be used for training, that it would not lose its meaning. Some excess of unsupported logic dissent in our ranks)
1) So the answer on the subject is obtained, the more so the author himself has shown us the knowledge of the answer to his question.
--------
2) Maybe then return to that thread? After all it was created for these purposes by you and many things have calmed down since then...
--------
3) The essence is the same. The main thing is that each period should have a clear name.
I do not insist on my model. And I can easily accept any that suits everybody.
One of the variants of universal classification, for example,
i.e. only the past and the future (at any arrangement of the "Present moment" in history), but at the same time with many realizations.
And everyone's adored OOS remains in place.
...........................................
It's just not clear what TheXpert is hiding up his left sleeve?
How about introducing a one and a half ) type then? A time limit, but without a clear entry point.
That's what I'm talking about, they don't divide them in a straight line into two piles).
->
There is no criterion for no exit (and the second type requires the exit to take place within a specified time), the inverse combination may not repeat, and there is no time limit for the trade.
For the second type TCs have a trading time limit, and the moment of the next signal arrival is known in advance.
This is what I am talking about, they are not divided in a straight line into two piles).
It's even worse. Even so, there is no mention of trend watchers at all in the classification so far. And this is a completely different class.
Or are we just parsing a narrow example? Or are we proving the merit of 2 models?
How about introducing a one and a half ) type then? A time limit, but without a clear entry point.