Recognising images ( rhetorical theme ) - page 12
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
denis_orlov:
думаю, именно это ты и просишь, т.к. сам конкретику представить не в состоянии.
I'm not asking for anything. And I didn't ask for any.
I told you - I do not need to give you the algorithm. I can be given the result of its work, and I will test the result. And I will tell you whether you can get something out of such an algorithm. Is it a request or a petition? It was a suggestion, which is primarily aimed at benefiting the author of the recognition algorithm. If someone can make a good recognizer, but cannot implement it in profit (it's hard!), he should directly seek cooperation with another developer.
You can upload a file with recognition signals here, in this thread, and someone else besides me will try to use these signals to test the algorithm for entering and exiting a trade.
But I see that you are trading with profit? So this offer is not for you :)
I'm not saying they are useless at all, but in my case they are not applicable.
The candlestick pattern recognition approach itself does not involve recognition per se, but a search for simple patterns. This loses >99% of the information encoded in the pattern.
Well, if you look at it that way, basically every indicator that moves away from the formal price value loses its informativeness and makes an error. For example, what is lost is the candlestick coding method used by Richie - K= (HL, LO/HL, LC/HL). Now the question is, what method of pattern selection on BP should be used?
Why do I ask the question, I've already faced this question many times when processing - how to formalize a pattern. I see two methods, first is coding - when we form codes similar to Richie or Lihovidov. The second method is to introduce boundary criteria. Again, in case of boundary criteria we lose one important factor - time. The number of frames during which the figure is formed should also be considered.
The problem is not recognising patterns (whatever the word 'pattern' implies) . The problem is preprocessing the data for analysis. Pre-processing is a surprisingly low priority in this forum. But all information must be presented in a form suitable for subsequent analysis.
Noise, gaps and spikes are typical obstacles for adequate analysis (no matter what methods are used). Let me draw a parallel with a mirror. Noise is analogous to the roughness of a mirror surface, the reflection becomes blurred and smeared. Gaps are cracks and shifts of parts of the mirror, as if it is broken. Ejections, or abnormally large bars (there are no abnormally small bars) are analogous to a crooked mirror. And some parts of the reflection are undistorted, while others are distorted beyond recognition.
These three problems must be solved separately. And then we can talk about pattern recognition.
Try taking a picture of the defective mirror (which I wrote about) and applying some kind of pattern recognition system to the photo. You might not recognise yourself in the reflection, let alone the "glitchy iron".
PS Each of the mirror defects carry information, but not about the original light that was reflected by the mirror, but about the causes of the defects (holidays and other factors).
There are other phenomena that support my argument. The human brain has 'built-in' filters of information coming from the senses. Thus people are easily able to talk to each other in a very noisy place, even if hundreds of people are talking to each other. Vision has the same property. The brain is able to focus on a single image element amongst the noisy elements - captcha is an example.
Is that why manual trading is difficult to formalize? Is it the reason why manual traders pay much attention to a single trading instrument and sharpen their brain-filters?
It seems to me that the concept of a pattern is better applied here in a generalised way, as a recognisable section of the price chart. It does not necessarily have to be a graphical figure. The main thing is that it should be recognized steadily, without omissions and failures if possible. Accordingly, the recognition method can be almost anything. Encoding is fine. There may be many encoding methods. Graphically is okay. Criterionally, that is what I understand by the indicators - also good. By pattern search - no. It works unstably and misses the most of them. True, the template search can be adaptive, but I have not seen this. Neuronet is good, but their training is a complicated thing, everyone only trains for trade. I don't think there's been any discussion here about using neural networks solely for pattern recognition. What other methods can you think of?
There are other phenomena that support my argument. The human brain has 'built-in' filters of information from the senses. Thus, people are easily able to talk to each other in a very noisy place, even if hundreds of other people are talking nearby. Vision has the same property. The brain is capable of concentrating on a single image element among noisy elements; captcha is an example.
Isn't that why manual trading is difficult to formalize? Is that why "manual traders" pay much attention to a single trading instrument while sharpening their brain filters?
I disagree :) According to works by Nobel laureates in the field of medicine Torsten Nils Wiesel and David H. Hubel, who conducted studies of the visual cortex of the cat, which found that there are so-called simple cells, which respond particularly strongly to straight lines at different angles, and complex cells, which respond to the movement of lines in one direction, ie the brain carries out the separation of traits. On this basis the whole class of NS called convolutional networks, which is based on convolutional mechanism, is developed. So here is the most interesting that this class of NS shows some of the best performance in image recognition with distortion (it is about the curve mirror and distortion) is very well shown in the works of Dr. Jan LeCun. But you can not apply convolutional nets to BP forex :) nets is good for distorted data recognition, but bad for image reconstruction.
There are other phenomena that support my argument. The human brain has 'built-in' filters of information from the senses. Thus people are easily able to talk to each other in a very noisy place, even if hundreds of people are talking to each other. Vision has the same property. The brain is able to concentrate on a single image element amongst the noisy elements - captcha is an example.