What to feed to the input of the neural network? Your ideas... - page 63
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
If we consider the case of patterned data. Multiplication tables, for example. The more examples you give, the more accurate the answers will be on the new data.
The new data should not be completely different, but between the training examples. I.e. interpolation will go more or less well. 1 tree will give the nearest training example. If by other data you mean data outside the boundaries of the training data, this is already extrapolation. The tree will give the outermost example, because it is the closest.
If we consider market data, then with a large value of noise, any peak from the true pattern will be mixed with noise peaks and we need to somehow choose the true peak and not the noise peak.
Your statements are correct here.
Noise again.
Everybody talks about noise.
But how can we define noise if we don't know the rules and laws?
What if every tick is a component of rules and laws and the problem is inability of architectures to decipher the "code" of a graph?
It seems like a postulate here (the idea of noise in a price chart)
About noise and patterns
It is not known what the presence and amount of noise in market data is, nor whether and to what extent patterns exist.
About learning, memorisation and retention
Preservation- If data is simply written to variables without evaluating its quality, it can be called preservation. An example is the normal writing of data to variables (database, table, matrix, etc.).
Memorisation: If quality assessment is done during the process of saving, it is already memorisation. For example, estimating the percentage of correct answers out of the number of questions to be memorised. In this case, the more examples, the wider the scope of possible applications. For example, for a 2 x 8 task the answer would be 16, but for 18 x 67 the answer could be anything, as this question was not included in the examples.
Learning: Learning from stored and memorised data is possible with quality assessment. Training is the process of forming rules for processing memorised data. For example, practising the application of the rule of multiplication by column. Here you need to remember only the minimum necessary information (multiplication table), and using the column rule you can multiply any combination of numbers, including 18 x 67 and even 1.657875 x 3.876754.
An example of learning would be GPT-like models that not only memorise data, but also apply rules to process different data by performing calculations such as columnar multiplication.
Now, having dealt with the concepts of learning, memorisation and retention, we can ask the question: where in machine learning (not involving GPT-like models) is learning applied to analyse market data?
P.S. This is basic stuff, later we will move on to the importance of valuation and other interesting aspects.
P.P.S. Overlearning , underlearning and similar states cannot be evaluated either quantitatively or qualitatively, so it makes little sense to talk about them in a meaningful and meaningful way.
Memorisation: If quality assessment is carried out during the process of preservation, it is already memorisation.
Databases, trees and clusters that memorise information 100% do not need evaluation. But you can check what 3*3 equals by trying all possible variants. It's up to you and your time... I'll do more important things.
Only undertrained models need evaluation.
1. databases, trees and clusters that remember information 100% do not need to be evaluated.
2. only undertrained models need evaluation.
1. If there is no evaluation, it means the process of retention, not memorisation. Above showed the difference.
2. How to determine which and when a model is "underlearned" and by how much?
Probably off topic, but my experience says that only oscillators (probabilities, winrates, etc.) can be predicted qualitatively. Those who dream of predicting price movements, my personal opinion is that it is not possible yet (or maybe even at all)!
what does winrate mean?
1. If there is no evaluation, then it is a process of retention, not memorisation. I have shown the difference above.
2. How to determine which and when a model is "underlearned" and by how much?
1) it is only for you that memorisation consists in going through possible variants and evaluating each one. You don't need to impose this on others and pass it off as truth.
2) It doesn't need to be defined, before training it is set in the parameters - to train 100% or to undertrain.
1) it is only for you that memorisation consists in going through possible options. Don't impose it on others and pass it off as truth.
2) It doesn't need to be defined, before training it is set in the parameters - to train 100% or undertrain.
1. What does memorisation and "going through the possible options" have to do with it? You don't need to make up something that I didn't claim and therefore couldn't impose on anyone.))
2. In the way you apply MO, and many people in general, learning is only conditionally present. Because there is no formation of rules for processing new unknown data. There is memorisation. I have shown above.
I realise that some things can be painful to accept because they shatter established beliefs. But if accepted, it becomes clear why some methods do not bring the expected results, and therefore allow you to change the direction of research.
1. What does memorisation and "going through the options" have to do with it? You don't need to make up something I didn't claim and therefore couldn't impose on anyone.))
2. In the way you apply MO, and many people in general, learning is only conditionally present. Because there is no formation of rules for processing new unknown data. There is memorisation. I have shown above.
I realise that some things may be painful to accept, because they ruin established beliefs. But if accepted, it becomes clear why some methods do not bring the expected results, and therefore allow you to change the direction of research.
What is the difference between an ordinary database and an intelligent system?
What is the difference between an ordinary database and intelligent systems?
As an example, the same gpt does not just remember the multiplication table, it is trained to use the column rule. That means it can multiply any number, not just within the multiplication table.
This is the difference between an ordinary database and intelligent systems. Such systems can apply known rules to process new information like a human does. An even higher level of intelligence is the ability to develop new rules using old ones as a basis.
Conventional "training" of a neural network does not create rules, and therefore does not work well on unknown data. It is simply an approximation of the data and we expect this approximation to work just as well on new data as on training data.