Machine learning in trading: theory, models, practice and algo-trading - page 1136

 
Oleg Papkov:

Heiken Ashi indicator without settings.

No shit... The picture is not serious

Everyone knows that any indicator can be fitted to a certain part of the history that it will become a grail, the same can be done for "any parameters" just to fit the history, this is for ***, for "grail" sellers

All this is very easy to check, fit an indicator on a tray and then backtest it on OOS (once), on a big piece of history so that at least 500 trades were made, the error decreases inversely as the square of the number of trades.

 
Igor Makanu:

Hmm, I don't even know whether to write here or in the humor topic:

In the morning disabled ad blocking in the browser on android tvbox for RBC banned for banning (on TV used to surf the web), now I was surprised that I was on the sites began to offer to sign up for machine learning courses, until I remembered that the banning is disabled, to paid courses for 12 weeks, not going to, but the training program promises to give free, have anyone training program on the MOE ??? - Needed thing when you want to pull up the theory!



ZSY: Some unlucky day, finally hijacked my box for spam, two years as the first, a new need to start again, for all sorts of leftist reg ... I'll do it tomorrow.

https://www.youtube.com/watch?v=qLBkB4sMztk&list=PLJOzdkh8T5kp99tGTEFjH_b9zqEQiiBtC

there's a playlist on the right
 
Igor Makanu:

So you want to say that there is no more explanatory information on MO in Russian than Vorontsov on the Internet?

I left Vorontsov for later, I have already made myself a youtube prop, while I'm still stuck on theory, a lot of little things that are not written either in articles or on forums

All about the same, in general.

I don't know what to study, I know almost everything ))
 
Igor Makanu:

I was interested in making an AI that learns to trade itself, having only prices, without a teacher or expert information. I put what I made in the kodobase, I will improve it later.

the rest is not the AI, but just rattles

For example, the question of the best choice of outputs (automatic, of course, without intervention) is not solved

 
Maxim Dmitrievsky:

I was interested in making an AI that learns to trade itself with only prices, without a teacher or expert information

This is the AI (strong artificial intelligence) that created the matrix and pitted the Russians against the Americans.

Look at thishttps://cloud.mail.ru/public/HRZX/uS3xg38cg say that the SII, but crookedly made.

There is a simple version from the Bandera man Andrei Kuchemenko:


void AI()

{

std::map<std::string, std::string> memory; // Словарь


while (true) // цикл 

{

cout << "enter question" << endl; // попросить ввести вопрос

string question;

cin >> question; // ввести вопрос с клавиатуры в переменную question

string answer = memory[question]; // запросить вопрос в словаре

// если в словаре есть такой ответ(не пустой) то вывести ответ на консоль

if (answer != "") cout << question << " this is " << answer << endl;

else // если нет в словаре ответа попросить ввести ответ

{

cout << "enter answer" << endl;

std::cin >> answer;

memory[question] = answer; // записать пару вопрос-ответ в словарь

}

}

}

Файл из Облака Mail.Ru
Файл из Облака Mail.Ru
  • 0s.mnwg65le.nvqws3booj2q.nblz.ru
Облако Mail.Ru - это ваше персональное надежное хранилище в интернете. Все нужные файлы всегда под рукой, доступны в любой точке мира с компьютера или смартфона.
 

I was selecting leaves from trees, and I found out that on the test period (M1) I was dangling around zero, and I think shit - I killed almost a year on all this MO, and then I accidentally tested an EA with another TF (M5), and it suddenly showed a normal result, I tried different TFs and found one more where the result is good (M2), and both on the training period and on the test period. Training period of the minute from 2015 to 2017 inclusive, tests on 2018.

Is this a fluke, or a manifestation of fractality?


The training period is only M1.

 
Maxim Dmitrievsky:

I was interested in making an AI that learns to trade itself, having only prices, without a teacher or expert information. What I made I threw in the kodobase, later I will improve it some more.

the rest is not the AI, but just rattles

For example, the question of the best choice of outputs (automatic, of course, without intervention) has not been solved

As far as I understood, there the agent randomly signals to open orders in the tester, and at the end the forest is built on collected samples from the selected through polynomials features.

I think that inputs can be marked by price difference and without orders, of course, IMHO, but for search of outputs, with maximum profit, just for...

 
Ivan Negreshniy:

As far as I understood, the agent signals randomly to open orders in the tester, and at the end the forest is built on samples selected via polynomials.

I think that entries can be marked on price difference and without orders, of course, IMHO, but for search of exits, with maximum profit, just for...

1st yes, and the polynomials can be changed, you can plug in whatever you want, as well as the forest itself on something else

I don't quite get it.

 
Maxim Dmitrievsky:

1st yes, and the polynomials can be changed, you can plug in whatever you want, as well as the forest itself on something else

2-th didn't quite catch it

2. Why should I look for inputs by random order shooting? They are almost uniquely determined as tops and bottoms of existing BP, and outputs are less unequivocal, so you can look for them.
 
Ivan Negreshniy:
2. Why should we look for inputs by randomly shooting orders? They are almost unambiguously determined as tops and bottoms of existing BP, while outputs are less unambiguous, so we can look for them.

By exits we mean the marks (any marks), both to buy and to sell. I don't consider any special algorithms of exiting positions, because a sell signal is a natural signal to close a buy order(though it may be wrong, I haven't thought about it)

so this is an interesting task - how to sample marks randomly, from some distribution, or by enumerating distributions in order to obtain a set of unique strategies and choose the best then

in other words, how to change the redistribution function in an efficient way