What to feed to the input of the neural network? Your ideas... - page 66
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
For humans, it is necessary to repeat many times to establish connections between neurons. But there are also people with excellent memory.
And establishing connections between neurons can be called recording. If it is recorded, even if not from the first time as in a computer, then it is remembered.
Computer and other electronic devices with memory (RAM, HDD, SSD, tapes, punched cards) memorise instantly.
For humans, it is necessary to repeat many times to establish connections between neurons. But there are also people with excellent memory.
And establishing connections between neurons can be called recording. If it's recorded, even if not the first time, as in a computer, then it's memorised.
Recording - saving on a medium, as much information as there is (compressed or not), so much will be recorded. It doesn't matter if it's the first time or the tenth time, it doesn't matter. I.e., you take information and put it on a medium, on the medium the information will be stored in an unchanged form and can be read in the same form - this is the meaning of recording.
Memorisation has a fundamentally different nature and meaning. If we take a human being (or any other living being), it is not a flash drive, it is impossible to write information to certain arbitrary parts of the brain and store it in an unchanged form. The brain can only memorise, i.e. form new connections between neurons. The number of neurons does not change during memorisation, only the connections between them change. When memorising, the brain cannot, has no way of knowing in advance how new connections between neurons should be created in order to remember information, it is an iterative process. Some people do it in a smaller number of iterations, others in a much larger number of iterations, but never clearly in one operation as it happens when writing on a medium. This is the difference between recording and memorisation, while recording is guaranteed (with a small error, which is usually solved by technical methods like error correction on hard media and special buffers) to save information on the medium, memorisation makes sense of the process of forming links between logical structures (in the brain these are neurons).
Neural networks have the same principle of memorisation as the living brain. It's not just writing values into variable weights, it's forming the right connections between artificial neurons, it's never possible to know in advance what kind of connections are needed for memorisation, so the process of memorisation is as iterative as in the living brain.
Bottom line is what was said earlier. Recording is storing information on a medium (regardless of the nature of the medium), memorisation is making connections between logical nodes.
Recording does not imply the presence of memorisation, whereas memorisation necessarily involves operations of writing to variables. We can say that memorisation is a higher level of abstraction, involving write operations, requiring iterative verification of the quality of memorisation.
A demonstration of the mechanism of memorisation in humans is the use of the anchor association technique, where the memorisation process is faster, more reliable and in fewer iterations. This is how the brain works; all information and knowledge can only be stored in the form of numerous connections between neurons. Neural networks work on the same principle, although in a very simplified form.
Hopefully, the meaning and difference between recording and memorising is now clear.
Recording - saving to media, as much information as there is (compressed or not) will be recorded. It doesn't matter if it's the first time or the tenth time, it doesn't matter. I.e., you take the information and put it on a medium, the medium will store the information in an unchanged form and can be read out in the same form - that's the point of recording.
Memorisation has a fundamentally different nature and meaning. If we take a human being (or any other living being), it is not a flash drive, it is impossible to write information to certain arbitrary parts of the brain and store it in an unchanged form. The brain can only memorise, i.e. form new connections between neurons. The number of neurons does not change during memorisation, only the connections between them change. When memorising, the brain cannot, has no way of knowing in advance how new connections between neurons should be created in order to remember information, it is an iterative process. Some people do it in a smaller number of iterations, others in a much larger number of iterations, but never clearly in one operation as it happens when writing on a medium. This is the difference between recording and memorisation, while recording is guaranteed (with a small error, which is usually solved by technical methods such as error correction on hard media and special buffers) to save information on a medium, memorisation has the meaning of the process of forming connections between logical structures (in the brain these are neurons).
Neural networks have the same principle of memorisation as the living brain. It is not just writing values into variable weights, it is forming the right connections between artificial neurons, it is never possible to know in advance what kind of connections are needed for memorisation, so the memorisation process is as iterative as in the living brain.
Bottom line is what was said earlier. Recording - saving information to a medium (regardless of the nature of the medium), memorisation - creating connections between logical nodes.
Recording does not imply the presence of memorisation, whereas memorisation necessarily involves operations of writing to variables. We can say that memorisation is a higher level of abstraction that includes recording operations, requiring iterative checking of the quality of memorisation.
A demonstration of the mechanism of memorisation in humans is the use of the anchor association technique, where the memorisation process is faster, more reliable and in fewer iterations. This is how the brain works; all information and knowledge can only be stored in the form of numerous connections between neurons. Neural networks work on the same principle, albeit in a very simplified form.
Hopefully, the meaning and difference between recording and memorising is now clear.
You have not said anything new, because I briefly mentioned the difference between human memory and machine memory. You can prefix human memorisation and machine memorisation. They have their own peculiarities.
Databases, forests and and clustering models write = remember at once. Without iterations and repetitions.
1. You haven't said anything new, as I briefly mentioned the difference between human memory and machine memory. You can prefix human memorisation and machine memorisation. They have their own peculiarities.
2. databases, forests and clustering models write = remember at once. Without iterations and repetitions.
1. Absolutely and unconditionally I said nothing new, as these are basic, elementary things. You denied the necessity of the iterative process of memorisation with evaluation control, so I had to repeat everything known long ago in detail.
In particular, the frequent misconceptions about these basic things are surprising. Dividing the concept of memorisation into machine and "human" memorisation makes no sense, as the process is fundamentally identical.
2. If you have something written down somewhere at once, without iterations, then there is no memorisation there. I repeat - memorisation cannot be performed in one iteration, because the result of memorisation is not known in advance due to the "coherent" nature of memorisation.
1. Absolutely and unconditionally I have not said anything new, because these are basic, elementary things. You denied the necessity of a iterative process of memorisation with evaluation control, so I had to repeat everything known long ago in detail.
In particular, the frequent misconceptions about these basic things are surprising. Dividing the concept of memorisation into machine and "human" memorisation makes no sense, as the process is fundamentally identical.
2. If you have something written down somewhere at once, without iterations, then there is no memorisation there. I repeat - memorisation cannot be performed in one iteration, because the result of memorisation is not known in advance due to the "coherent" nature of memorisation.
Trees spread data over leaves in one pass.
Repeat/iterate as much as you want. It will not change existing and successfully working algorithms.
Trees spread data over leaves in one pass.
1) Always in 1 pass.
You can set the settings to always be the same.
But usually you apply randomisation: feeding random rows and/or columns to the training.
For my experiments I always disable anything that introduces randomisation: for reproducibility between runs
2) Greedy algorithms. That's why in 1 pass. There are plenty of articles about tree algorithms.
Here you can repeat the tree in the code yourself and understand how everything works there.
https://habr.com/ru/companies/vk/articles/438560/ You are an experienced programmer - you will understand it easily.
And the 2nd part https://habr.com/ru/companies/vk/articles/438562/
1) Always in 1 pass.
You can set the settings to always be the same.
But usually randomisation is applied: feeding random rows and/or columns to the training.
2) Greedy algorithms. That's why it's 1 pass. There are plenty of articles about tree algorithms.
Here you can replicate a tree in the code yourself and understand how everything works there.
https://habr.com/ru/companies/vk/articles/438560/.
The 1st is answered, the 2nd is in the article. If you understand the algorithm, you will understand everything. The code is clearer than a thousand words.
The 1st is answered, the 2nd is in the article. If you understand the algorithm, you will understand everything. The code is clearer than a thousand words.
Throw my questions to the chat, if I'm too lazy to type the answer myself)))
S.F. By the way, you can read the article on your link to answer my questions. And on mql5.com there is a huge amount of educational material in articles on the topic, and in native MQL5.
ZZY. Pay attention when reading the tutorials that trees do not branch in an arbitrary way, but pursue quite specific goals. What goals? At what stage does the iterative process of tree training take place and why is it iterative and what purpose does this iterative process pursue?
.