Machine learning in trading: theory, models, practice and algo-trading - page 1916

[Evgeny Dyuka]

Wizard2018:

+1000% Very very sensible thoughts.

Disagree. There is no secret knowledge about the market. There is a very noisy dataset, and a neural network must and can find patterns in it. The regularities are there, but they are few, that's why the result is not very impressive so far.
And it is definitely necessary to feed indicators, they simplify the network work because they already show some regularities ready-made. For example, the net, of course, may find the patterns we call divergence, but we can just give it that. To put it simply, you can distinguish a cat from a dog by its eyes, but you can also show its tail and ears, in this case the result will be faster and of higher quality. This is the role played by indicators.

[mytarmailS]

Mihail Marchukajtes:

Here's my answer to you. He's unconditionally infuriating in one way, purely visually. But never mind, because the main thing is the essence of the narrative, not how it looks and how creepy the sound can be... Just take note. yes I am and I think so. And yes I don't have 100% results, no more than 55% in the general sense of that understanding :-))))

What you want can be reduced to the usual classification where the signs are metrics and the target is the adequacy of the model on new data...

Generate a model, simulate the work of the model on the new data and create a dataset like in the figure.

Then check the importance and strength of the attributes (metrics) and you will know what is important and what is not.

[NeuralNetwork]

Hi all. Here is the monitoring of my account. It shows everything to shut everyone up forever. If you are losers in life, just shut up and let others do the work.Neural networks work, it is your brain that does not work, that you claim otherwise !

[NeuralNetwork]

That's it. I deleted the account and the account, too. Whatever you do not conduct surveillance, you have nothing to do. Only on the forum to talk and panic from failures.

[NeuralNetwork]

The dreams of the weak are an escape from reality; the dreams of the strong shape reality.

Believe in the dream. It has the pleasant feature of coming true.

The poor man is not the one without a penny in his pocket, but the one without a dream.

"Slowly he dies,

"He who does not travel,

"who does not read,

who does not hear music,

Who cannot find harmony in himself.

"Slowly he dies

who destroys his faith in himself,

Who does not allow himself to be helped.

Slowly dies the one

who destroys love itself,

And lives out his days with constant complaints

Of bad luck or incessant rain.

Slowly dies he who gives up his plans,

He who gives up his plans,

Before he has even begun them!


'' Quotes about Dreams (538 Quotes) | Quotes of famous personalities '' - read this interesting ))

[mytarmailS]

I have an idea to create a branch to discuss the target functions, not even a discussion, but rather to create a database of different types of target and statistics on them, what works and what does not work at all.

What do you think, who needs it?

[Aliaksandr Hryshyn]

mytarmailS:

I have an idea here, to create a branch to discuss the target functions, not even a discussion, but rather to create a database of different types of target and statistics on them, what works and what does not work at all.

Do you think anyone needs it?

Good idea

[Rorschach]

mytarmailS:

I have an idea to create a branch to discuss the target functions, not even a discussion, but rather to create a database of different types of target and statistics on them, what works and what does not work at all.

What do you think, who needs it?

My thoughts on the target. If we assume that there is no spread and commission, the maximum profit will be when guessing all of the highs and lows. That is, the task comes down to predicting one step ahead. Since it is sufficient to know the direction of a trade, we can do with binary classification. We take the increments, find the sign of the increments and use it as a target.

Now let's assume that there is some constant spread (for simplicity). Now we are interested in movements larger than that spread. And here I'm stuck, how to build such a target, what to feed in bids, asks or average

[mytarmailS]

Rorschach:

My thoughts on the target. If we assume that there is no spread and commission, then the maximum profit will be when guessing all the highs and lows. That is, the task comes down to predicting one step ahead. Since it is sufficient to know the trade direction, we can do with the binary classification. We take the increments, find the sign of the increments and use it as a target.

Now let's assume that there is some constant spread (for simplicity). Now we are interested in movements larger than that spread. And here I'm stuck, how to build such a target, what to input bids, askks or average.

It's the same as zigzag or increments to predict, in fact it's all one class of targets, a useless class... I meant in a much broader sense, to come up with a whole zoo of targets, different directions

for example :

predict the price of the day's maximum by the intraday data (regression)

at the breakdown of what level there will be excess volatility (regression + classification)

Using int raday data predict the time of the day reversal (classification)

if the first candle was black will the third be white? (classification)

mark out the support and resistance levels and forecast the level from which there will be a pounce/breakdown (classification)

Predict the optimal period for the indicator at any definite moment of time (regression)

.........

.....

...

ITP.....

You should not look at your targets too narrowly, you have to think, propose variants and test them.

Eventually you might find something interesting.

=========================

As a result, you can create a list of interesting targets which proved their worth in tests. And it will be a knowledge base for all the enthusiasts of MI.

You understand, the targets can be searched in the same way as the signs, and you probably not only can, the practice shows.

[Aleksey Vyazmikin]

Mihail Marchukajtes:

Here's my answer to you. He's unconditionally infuriating in one way, purely visually. But never mind, because the main thing is the essence of the narrative, not how it looks and how creepy the sound can be... Just take note. yes I am and I think so. And yes I don't have 100% results, no more than 55% in the general sense of that understanding :-))))

Watched the video - at one time even tried to build a model for model selection on different estimated coefficients - the effect was, but as usual in MO - a small percentage of accuracy. In general the idea is correct, you need to come up with more different predictors describing the model.

I look at error balance and analyze it using the same methods as the standard report in the strategy tester. I was just testing a script on old models, which plots the balance of errors, and found a nice variant on the sample outside of training, which was from mid 2018 to early 2019, ran it on the data for the last year - profit. And looked at the models that were selected at that time - the result is twice as bad, including the balance of errors.