expert testing of strategies - page 4

New comment
 
Integer:
Artur:

1 - the strategy is likely to be curve-fit and has not shown itself to be sustainable;

Do you think this problem is solved? Not knowing about the strategy itself, nor knowing the timing of the trades and the symbol.
And how are all sorts of neuronics different from the black box?
 
Notused, by the fact that in some of these "all kinds of neurons" (intelligent) the user knows the inputs (70% success rate), the architecture (10%), the outputs (10%), the error optimization criterion (10%). What the NS does internally is not important to the user. In addition, sufficient information must be fed to the black box input.
 
I think they are different. It is known what they were trained to do. And it was done according to a well-known algorithm.
 
kg*m/(Celsius) * Hz * (Integral of time from last year to today in m/s2) => price will go there - this is the neuronics. By the way, Mathemat, how are the degrees of freedom doing? I mean that the black box (neuron) is also fed with redundant information, and they think that the more, the better is the result. I don't know (haven't heard) about any successful neuronics on the market. At most, they help to determine the patterns, but as you know, the patterns are tricky (I was doing research, but my Expert Advisor got lost). So on the eurusd there is not a single stable set of candlesticks, which would bring profit from 1999 to early 2007. The same thing with the pound, I do not change it further. I checked the patterns black and white. the length from two to 20 candlesticks - everything I managed to get from the terminal from memory. The prognosis was to find the colour of the next candle. But maybe the approach is wrong). And the physical meaning of the variables inside the neuron - see the beginning of this post. So it turns out that with the help of sugar, an axe and a piece of wood we build a Porsche of the latest model. I think we can agree that the neuronka is a black box. And about that "What the NS does internally doesn't matter to the user" - so why does it matter in this discussion? What matters is the result. Right?
 
notused писал (а): Кстати, Mathemat, как там степени свободы поживают? Это я к тому, что к чёрному ящику (нейронке) ещё и избыточную информацию подают, причём, думают, что чем больше, тем результат краше.
Only an absolute sucker can feed a neuron input a few time series (not processed in any way), and bar close prices at the output, and still think that the NS will figure it out, because it is smart.

I do not know (I haven't heard) about any successful neuronics in the market.

Not heard - does not mean that it is impossible to create such a neuron.

At most, it helps to identify patterns [...] It didn't work. Although maybe the approach is wrong.

In general patterns do not work well on forex.

The approach is rather wrong. I've experimented with NS myself - so what, now I think that successful NS is impossible? And I haven't even taken up classifying NS yet, but I've already managed to draw some valuable conclusions...

And about "What NS does inside doesn't matter to a user" - why in this discussion does it matter? The main thing is the result. Right?

Right. And even more important is proving the result is true. If we consider backtesting of pips and nothing else to be the result, then it's bullshit. If you consider as the result the conclusions based only on analysis of a vector of trade results without time tags, then it's almost always rubbish too.

It's just that both you and I are separated by an abyss from the normal, professional application of NS.
 
Vinin:
Artur:
Good day to you all !
I couldn't resist giving a few comments.


And now about the most interesting - the purpose of the branch - many may not agree with me, but I think public discussions are absolutely ineffective, because those who only know complex words and juggle terms will not contribute to the solution of the problem anyway - they write articles that seem to be the height of professionalism to newbies, but in reality there is little behind these articles. The same few who can still really create something working or at least close to working, they will not bring their ideas to the general forum, because they know the price and they do not need it. If the idea is interesting, why publish it instead of developing it yourself? - That is my conviction.
That is why I believe that the most effective method is individual communication when two or at most three people are looking for an answer. And when everyone understands that in a small circle nobody benefits from "taking the real thing out of the house", there is a certain trust and it is possible to move forward.
Thus, this topic is an attempt to find people who have good experience and are able to think independently and constructively in order to establish personal communication with them. And one of the criteria to determine a person's ability is this tester.

So hopefully resolved all the questions that have arisen.

Then why this thread?
As I wrote - communication, acquaintance, interested in people who can think without dogma, simply and constructively.
 
Integer:
Artur:

1 - the strategy has most likely been fitted to the curve and has not proved to be stable;

Do you think this problem is solved? Without knowing about the strategy itself, nor knowing the timing of the trades and the symbol.


The problem is solved asymptotically - i.e. we cannot say for sure that everything will work well, but we can say for sure that it will hardly work.

You need to know the total period of time, but you don't need to know the symbol, the algorithms and so on. You need to know how TV sets work to determine if they are on? Same logic here.

Besides, what does it matter if the symbol is different volatility on all pairs, so what? Different volatility is a roughly speaking different scale - just adjust the strategy parameters and it will work on any pair. And if your strategy works on the euro and does not work with any parameters on the eurofrank, you can tell right away that it does not work.

 
notused:
kg*m/(Celsius degrees) * Hz * (Time Integral from last year to today in m/s2) => the price will go there - that's the neuronics. By the way, Mathemat, how are the degrees of freedom doing? I mean that the black box (neuron) is also fed with redundant information, and they think that the more, the better is the result. I don't know (haven't heard) about any successful neuronics on the market. At most, they help to determine the patterns, but as you know, the patterns are tricky (I was doing research, but my Expert Advisor got lost). So on the eurusd there is not a single stable set of candlesticks, which would bring profit from 1999 to early 2007. The same thing with the pound, I do not change it further. I checked the patterns black and white. the length from two to 20 candlesticks - everything I managed to get from the terminal from memory. The prognosis was to find the colour of the next candle. But maybe the approach is wrong). And the physical meaning of the variables inside the neuron - see the beginning of this post. So it turns out that using sugar, an axe and a piece of wood we build a Porsche of the latest model. I think we can agree that the neuronka is a black box. And about that "What the NS does internally doesn't matter to the user" - so why does it matter in this discussion? What matters is the result. Right?

I like the way you think, one has no dogmas, emotions or illusions.
 
Artur:
You need to know how the TV works to know if it's working or not? Same logic here.


I am reminded of some film about burglars. The guard thought he was watching on a monitor in the corridor, but was actually watching a video tape.

 
Mathemat - I quite agree that just because I haven't heard doesn't mean it can't happen. I'm talking about something else - how many free variables can you suggest? At first glance there are only two - time (or a function of it) and price (or a function of it). Although not the fact that they are independent - on weekends trading almost never happens. But I would like to have at least 2 :) Suggest another one (moon phases don't matter - they depend on time). If you offer me one, then I'll admit that Artur's black box is different from
neuronkey's black box.
Of course, you could use other variables as a base, but it's too late and I'm not thinking clearly.
1234567
New comment