Discussion of article "Neural Networks: From Theory to Practice" - page 5

New comment
 

I will try to separate the concepts of "fitting" and "training".

Fitting - selection of model parameters in order to match the modelled process. An analogy can be made - "rote learning".

Learning - the process of cognition of the process by the model, which includes memorisation of "rules" and "exceptions to the rules" for the purpose of further possibility to draw conclusions on the basis of the incoming unknown information. In NS training, the validation part of the training sample is used for this purpose.

Thus, we can say that the NS in the Expert Advisor from the article is adjusted rather than trained, as well as any Expert Advisors optimised in the tester. The optimiser's functionality is not sufficient for training EAs (with or without NS) (although there is forward testing, but it only means that we only select the one that passes the forward test).

 
joo: For training EAs (with and without NS) the optimiser's functionality is insufficient ...
Is there a way out of this situation? Should we write our own training algorithms? I hope that MQL5 articles already contain it.
 

Yedelkin:

................., and the term "training" has been given a new highly specialised meaning, namely: training is the usual selection (fitting) of parameters .

Any cycling (playing badminton, etc., etc.) is learnt by fitting parameters of own neural networks.

Copying (rote learning) is a much more primitive way of "learning". // Including learning dictionary definitions.

By the way, no one learns a language (native language) from dictionaries, they learn by "fitting".

 

MetaDriver: Отнюдь не новый.  Любой езде на велосипеде / игре в бадминтон  обучаются путём подгонки параметров собственных нейросеток. 

...By the way, nobody learns a language (native language) from dictionaries, they learn it by "fitting".

Wonderful example of explaining ordinary phenomena with the help of highly specialised terms :)

I'm reminded of this:

- Son, what are you doing?

- Studying MQL5.

- Don't be silly, you learn the language by fitting parameters of your own neural networks.

MetaDriver: Copying (rote learning) is a much more primitive way of "learning". // Including learning dictionary definitions. By the way, no one learns a language (native language) from dictionaries.

It only remains to add that dictionaries are not for rote learning and "learning dictionary definitions", but for reflecting the meanings of words that are considered generally accepted.

 
Yedelkin:
Is there a way out of this situation? Should we write our own learning algorithms? I hope that the articles on MQL5 already contain it.

There is already something on optimisation algorithms, yes. On learning algorithms, no.

And there are no articles on methods of selecting the most optimal variants from the total number of runs (including in the standard tester).

 
joo:

There is already something on optimisation algorithms, yes. On learning algorithms, no.

And there are no articles on methods of selecting the most optimal variants from the total number of runs (including in the standard tester).

There you go... I.e. we won't be able to create a self-learning programme in the next year. Too bad. I will remain a dummy :)
 
Yedelkin:

A wonderful example of explaining ordinary phenomena using highly specialised terms :)

Inspired:

- Son, what are you doing?

- Studying MQL5.

- Don't be silly, you learn the language by adjusting the parameters of your own neural nets.

Well, the sense of humour is awake, so "the patient is on the mend." :)


The only thing left to add is that dictionaries are not used for rote learning and learning "dictionary definitions", but for reflecting the meanings of words that are considered generally accepted.

Oh, for what they are not used only...! Here on our favourite forum they are mainly used to piss on each other's heads with them.........

;)

 
papaklass:
Generally speaking, a mathematical model of any process or phenomenon is a description in mathematical language of the laws to which this process or phenomenon obeys. But the control of this process with the help of parameters can be called fitting. Maths is an exact science, so definitions must be exact.
Holy shit. And you probably believe it yourself...
 
papaklass:
Generally speaking, a mathematical model of any process or phenomenon is a description in mathematical language of the laws to which this process or phenomenon obeys. But the control of this process with the help of parameters can be called fitting. Mathematics is an exact science, so definitions must be exact.

"Laws" are already models. They exist only in the head.

Real processes do not obey any laws ))

 

Dear speakers.

Of course, I am not against discussing the subtleties of neural networks in this thread, but originally, the article was planned for beginners. It omits some details, because these very details are able to confuse a novice neural networkist. Of course, the article does not specify various methods of training (fitting) neural networks, but it is not necessary at the initial stage. If you realise that neural networks are not that difficult, it does not give you an excuse to turn away and say "this is very difficult and not for me". If you know more - that's great, then the article is probably not for you.

Regarding self-study - it is possible that changes were made during the moderation process that were not in the original version. At the moment the source of the article is not at hand, but as soon as possible I will check this article for errors of this nature.

After some reflection, it was decided to write the second part of the article.
At the moment the second part will cover the work with multilayer neural networks.
If you have any wishes about its content - please, briefly, write them.
Those ideas that I can convey on my fingers will be described in the article.

Thank you.

1234567891011
New comment