neural network and inputs - page 26
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
You see, you write your assumptions, while I checked it half a year ago - 80% of historical data does not repeat, if it does, then 1-2 times, which is insignificant in such an array of data - if it doesn't repeat, then why is it not chaos? Well, admit that not to repeat the combination, for example of 10 or 20 bars over 10 years of history, not once at one-hour timeframe - it takes some effort, and the market has done so not once, but constantly
What does it have to do with whether they repeat or not? You're making the claim that if they don't repeat, it's chaos. Is that right? You're just sort of cycling on the fact that you have to look for similar plots (very simplistically speaking)... But there are other approaches, after all... I fundamentally disagree with your assertion - that if it doesn't repeat then it's random wandering.
You see, the conversation touched upon an old sore subject - SB or not SB? Well personally I'm thinking - what kind of SB can there be when it comes to money? Generally speaking? Well the price in the market can't be formed by an HF generator, can it? (By the way a good HF generator, really random, is not such an easy task). It's just that the process is complicated... And if my or your approach "doesn't catch" any segment of the participants' body movements, then you shouldn't immediately be talking about SB. It just means it's the wrong approach.
An example (very simplified) is that MTS was invented, which catches the elite. But it's not designed to catch the MASH guys or the Zigzags. So it catches let's say 30%, the rest 70 (MA and ZZ are chaos for it).
And what you wrote that you checked - 80% do not repeat ... Sorry, it depends on what kind of patterns you were looking for, and in general, what you can call a pattern. You have to agree...
For example, if we consider a pattern to be a price movement of more than 30 points per hour (of course, it is just an example, but formally - isn't it a pattern?) - then such moves will be a ton and a half. The more complex is YOUR pattern (I am sure it is a binary one) - the less frequently it will occur on the history.
What does it have to do with whether they repeat or not? You're making the claim that if they don't repeat, it's chaos. Is that right? You're just sort of cycling on the fact that you have to look for similar plots (very simplistically speaking)... But there are other approaches, after all... I fundamentally disagree with your assertion - that if it doesn't repeat then it's random wandering.
You see, the conversation touched upon an old sore subject - SB or not SB? Well personally I'm thinking - what kind of SB can there be when it comes to money? Generally speaking? Well the price in the market can't be formed by an HF generator, can it? (By the way a good HF generator, really random, is not such an easy task). It's just that the process is complicated... And if my or your approach "doesn't catch" any segment of the participants' body movements, then you shouldn't immediately be talking about SB. It just means it's the wrong approach.
An example (very simplified) is that MTS was invented, which catches the elite. But it's not designed to catch MASH guys or Zigzags. So it catches let's say 30%, the rest 70 (MA and ZZ are chaos for it).
And what you wrote that you checked - 80% do not repeat ... Sorry, it depends on what kind of patterns you were looking for, and in general, what you can call a pattern. You have to agree...
For example, if we consider a pattern to be a price movement of more than 30 points per hour (of course, it is just an example, but formally - isn't it a pattern?) - then such moves will be a ton and a half. The more complex is YOUR pattern (I am sure it is a binary one) - the less frequently it will occur on the history.
And there are still some reasonable people here. It's nice to read you.
Well, hello. :)
And there are still some reasonable people here. It's nice to read you.
Hey, there. :)
good evening to you too :)
good evening to you too :)
Thank you. :)
I remember fiddling (:wore?:) with patterns about four years ago. then i put it aside, like until better times. maybe it will come soon.
One of the intermediate results was even posted (here).
Probably soon (or not very soon) I will get back to these things on a new level of technical (and ideological) potential. there are fish there but it is hard to test - scanning of patterns on history is similar to testing, with similar time of scanning. if i run all this stuff in a tester, the process gets slower... But generally speaking, the approach works, especially if you have good inputs. For my method raw history is just as bad for inputs as for neural nets. actually, the method is similar to training nets. we can even say that it is "training of single-layer neural nets to recognize one pattern using one-pass method". :)
good luck!
Thank you. :)
I remember fiddling with patterns about four years ago. Then I put them off, like until better times.
my results were quite positive. i.e. all this is quite computationally intensive. i even posted one of my intermediate results (here).
maybe soon (or not) I will come back to these things on a new level of technical (and ideological) potential. there are fish there but it is hard to test - scanning of patterns on history is similar to testing, with similar time of scanning. if you run all this stuff in a tester you get quadratic slows down the process... But anyway, the approach works, especially if you have good inputs. For my method raw history is just as bad for inputs as for neural nets. actually, the method is similar to training nets. we can even say that it is "training single layer neural nets to recognize one pattern using one-pass method". :)
good luck!
Personally I don't like patterns because I always get small statistics with them, it's hard to make statistically reliable conclusions. Somewhere I met a conclusion I don't know where it's coming from, that 30 instances are enough for evaluation (generally speaking) (where's that number coming from?). But I have here, in the course of torture, it turns out that 200-300 is not enough. Perhaps in stocks such numbers are available, but in Forex, personally for me - no. In general, I'm a supporter of somewhat different methods, where you can set more precedents for SUSTAINABLE conclusions.
Otherwise, it turns out - the patterns are found, then run out, "put them in the cans", then work again... allegedly...
I don't even know who I'm answering to. No one.
We all do things at home with our hands. I recently got the task of hanging a curtain rod for a heavy drape in my house, and the customer insisted that the number of hooks must be at least N.
Because of my disagreement with this design solution, I was forced to justify its untenable nature.
Each hook has a platform on wheels, the length of the platform is L. The product N*L=F determines the fixed part of the length of the ledge that will ALWAYS be occupied by the drape. If F=C (the full length of the curtain track), the drape will become unmovable, i.e. there will NEVER be any daylight entering the house.
My point is that the size of the fractal in my example is the same as the length of the carriage, and any solution to the problem where F is close to C is idiotic.
Do I need to explain that there are no metaphysical markets and the market saturation with fractals (pardon, paternals) is 20% not bad at all? And the porterage opens at 80%, which corresponds to a rather understandable market.
Don't believe me? Do a field experiment on the curtains in your house :)
........
Don't believe me? Do a field experiment on the drapes in your house :)
"Jolly Barin..." (c) the cabman taking Hippolyte Matveich // from the restaurant where he, while seducing Liza, squandered his deposit of his share of cash, with the result that the aggregate dough was not enough to buy the chairs at the auction. :)
I personally do not like patterns because I always get small statistics with them, it's difficult to make statistically correct conclusions. Somewhere I met a conclusion I don't know where it's coming from, that 30 instances are enough for evaluation (generally speaking) (where's that number coming from?). But I have here, in the course of torture, it turns out that 200-300 is not enough. Perhaps in stocks such numbers are available, but in Forex, personally for me - no. In general, I'm a supporter of somewhat different methods, where you can set more precedents for SUSTAINABLE conclusions.
Otherwise, it turns out - the patterns are found, then run out, "put them in the cans", then work again... supposedly...
Lack of data for statistics collection can be compensated by the number of analyzed (1) instruments and (2) indicators.
Regarding (1) it seems to be clear. if a pattern "works" on one instrument and "does not work" on another one, you may wonder "was the pattern boy?", maybe the pattern "worked" just by accident ("tried to fool us" (c) Taleb)
Regarding (2) - I've already mentioned above that I analyzed not only (and not so much) patterns on raw quotes, but patterns on indicators. By the way, "non-standard" indicators are a special song. Here's a simple reasoning: let's assume that all liquidity providers in the forex market are "conspiring against traders" and move quotes at any given time against the total trader's position (a very likely scenario). Let's forget about "how scary to live" and all that. Let's get scared for a while, wring our hands in indignation, and then ask ourselves a practical question: how do we exploit this "law of the banking law jungle " when we are "on this side of the movie? we don't have level3 information (the banks do). so how do we live if we're bored of being afraid? After all, without any of such insurance guarantees, we can reason something like this: if aggregate position (remember - unknown to us) is formed by traders using mainly "standard" analysis tools, then it is "statistically enough" for us to know what exactly these very standard tools recommend to a "standard trader" - the very one who will form an "aggregate position" in the market. And then - more bells and logic, plus a set of non-standard analytical tools... and something may come together. Maybe not a grail, but it will make your bread and butter...
something like this.
// I hope this reasoning helps to understand one of the (quite possible) reasons why standard patterns "do not work".
// Only profits return to those who work even better than "standard patterns" - to those who know (or make good predictions) about a trader's aggregate market position.
We don't have levelled3 information (banks do). so how do we live if we're bored of being afraid? After all, without such insurance guarantees, we can reason reason in the following way: if aggregate position (remember - unknown to us) is formed by traders using mainly "standard" analysis tools, then it is "statistically enough" for us to know that exactly these very standard tools recommend a "standard trader" - the one who will form an "aggregate position" in the market. And then - more bells and logic, plus a set of non-standard analytical tools... and something may already be brewed. well, maybe not a grail, but for bread and butter...
Like this.
Believe it or not, this is exactly the kind of approach I am trying to exploit, right on. Only I do not start from the "standard set of the standard trader" (perhaps in vain), but from the standard... uh, how shall I put it... the actions... of the standard trader.
Standard set is also a good topic by the way, indeed. MAs for example. A lot of people are sitting on them. And all the rest sit on the same MAs, only in profile :)
Believe it or not, this is exactly the kind of approach I am trying to exploit, right on. Only I do not start from the "standard set of the standard trader" (perhaps in vain), but from the standard... uh, how shall I put it... the actions... of the standard trader.
Standard set is also a good topic by the way, indeed. MAs for example. A lot of people are sitting on them. And all the rest sit on the same MAs, only in profile :)
Believe it or not, this is exactly the kind of approach I am trying to exploit, right on. Only I do not start from the "standard set of the standard trader" (perhaps in vain), but from the standard... uh, how shall I put it... the actions... of the standard trader.