Machine learning in trading: theory, models, practice and algo-trading - page 1530

 
Maxim Dmitrievsky:
Happiness doesn't need any reason

Right, you need antidepressants to be happy, but curiosity can only be satisfied by research. I wonder what my data works or my optimizer. I think it's 50/50. Well matched data with a pretty passable optimizer. You can do a comparative analysis between the two AIs and unambiguously calculate who and what works.


Test for the optimizer:

1. On my data, your NS got a better result on the OOS. In favor of your optimizer.

2. On my data, your NS got the same result on OOS. Uncertain.

3. On my data your NS got a worse result on the feedback loop. In favor of Reshetov's optimizer.


4. On your data Reshetov got a better result on OOS. In favor of Reshetov.

5. Reshetov got the same result on OOS on your data. Undetermined.

6. Reshetov got worse result on OOS on your data. In favor of your optimizer.

Data test:

I by your baseline strategy save my data, you by my baseline save your data this makes it possible not to reveal the logic of the inputs and again optim.

1. If I get a better result on your data than on mine. In favor of your inputs.

2. If I get the same result. A draw.

If I get a worse result, in favor of my inputs.

Also do the same on your side. So you can at least try on the other researcher how good your approach in principle.

It's not a sour trick, there's a lot to negotiate, etc., but if you make this comparison, you can get a comparative estimate of the power of your TS relative to mine. Again, I'm not saying I have it so right all the time. Nope. SkyNet 2.0 is a finicky thing, you have to keep an eye on it. I understand, that it will be a great challenge to be compared in radically different approaches, but I would like to, it's boring :-(

 
Mihail Marchukajtes:

Right, you need antidepressants to be happy, but curiosity can only be satisfied by research. I wonder what my data works or my optimizer. I think it's 50/50. Well matched data with a pretty passable optimizer. You can do a comparative analysis between the two AIs and unambiguously calculate who and what works.


Test for the optimizer:

1. On my data, your NS got a better result on the OOS. In favor of your optimizer.

2. On my data, your NS got the same result on OOS. Uncertain.

3. On my data your NS got a worse result on the feedback loop. In favor of optimizer Reshetov.

What is your NS? Just take any NS and see :) messing around with other people's datasets is not a good thing.

closest to the kernelized SVM jpredictor, or rather, it's the same thing. Here try it.

 
Are you interested in integration withhttps://github.com/mlflow/mlflow/ and MT5?
mlflow/mlflow
mlflow/mlflow
  • mlflow
  • github.com
MLflow is a platform to streamline machine learning development, including tracking experiments, packaging code into reproducible runs, and sharing and deploying models. MLflow offers a set of lightweight APIs in that can used with any existing machine learning application or library (TensorFlow, PyTorch, XGBoost, etc), wherever you currently...
 

To read it I found

https://medium.com/towards-artificial-intelligence/a-simple-neural-attentive-meta-learner-snail-1e6b1d487623

https://arxiv.org/pdf/1707.03141.pdf

Attention networks and transformers are trendy nowadays, they say it's better than RNN

A Simple Neural Attentive Meta-Learner — SNAIL
A Simple Neural Attentive Meta-Learner — SNAIL
  • Sherwin Chen
  • medium.com
Traditional reinforcement learning algorithms train an agent to solve a single task, expecting it to generalize well to unseen samples from a similar data distribution. Meta-learning trains a meta-learner on the distribution of similar tasks, in the hopes of generalization to a novel but related tasks by learning a high-level strategy that...
 

Oh, ladies and gentlemen, oh... The granite of science is heavy, it's unbearable. Barely managed to read"New Neural" - the project of Open Source engine of a neural network for the MetaTrader 5 platform, I've read it for several months, it was 100 pages long, in the evenings after work, I can't follow this branch of >1500 pages for the rest of my life, I've had no luck, I tried to read selectively, I haven't seen anything useful, though I suppose that probably someone was expressing himself in florid or metaphorical terms and I had to read between the lines.

What to do ladies and gentlemen? WHAT DO WE DO?

I'm not so good at mql yet, but I really want to do something with some urgency. No super profits yet, but at least 5-10% per month as a start. I think I may start with linear regression and forecasting of next increment according to several previous ones and start it in trading. What steps do I need to take to do that?

More details: let's take a number of increments(x(t) - x(t-1)) of length for example 10, and 11-th(x(t+1)) will be a variable we are looking for, then we consider a linear regression, get a forecast, if the forecast is greater than zero we open long, if less we open short. This is the most direct application of MO to trading.

How do you like this plan?

"Новый нейронный" - проект Open Source движка нейронной сети для платформы MetaTrader 5.
"Новый нейронный" - проект Open Source движка нейронной сети для платформы MetaTrader 5.
  • 2011.10.18
  • www.mql5.com
Общее обсуждение: "Новый нейронный" - проект Open Source движка нейронной сети для платформы MetaTrader 5.
 
kapelmann:

More details: we take a series of increments(x(t) - x(t-1)) of length for example 10, and the 11th(x(t+1)) will be the desired variable, we consider the linear regression, get the forecast, if the forecast is greater than zero we open long, if less we open short. If it is less than zero, then we open long and short. In fact, this is the most direct application of MO in trading.

And you're talking about econometric archaic beliefs

AI
AI
  • www.mql5.com
AMkA Скользящая средняя по Кауффману с возможностью читать через iCustom, изменять цену, к которой применяется и способ расстановки точек: на базе пипсового фильтра или с расчётом стандартного отклонения. Сглаживание ряда Close с помощью преобразований Фурье Индикатор выделяет наиболее значимые гармоники движения, отбрасывая...
 
kapelmann:

But I very much want to make something urgent, for super profits is not aimed yet, but at least a stable 5-10% per month, as a start.

I don't even know if I will upset you or not, but I will still write that just consider the percentage of profit per month without the stipulated maximum drawdown, it's a road to nowhere.

I don't know if you're upset, but I'll write an article and tell you that looking at % profit a month without an agreed-upon maximum drawdown is a way to nowhere.)

 
Igor Makanu:

I don't even know if I'm going to upset you or not, but I'll still write that just consider the profit % per month without the agreed maximum drawdown, it's a road to nowhere.

People chop up 100% a month with a great drawdown ;)

The drawdown is at least twice less than the profit, although it is desirable that it was not there at all.

Kesha Rutov:

Direct use of MO in algotrading is ReshetovMLP and Random Forests predicting trends, while you're talking about econometric archaic

These are advanced methods, MLP and RandomForest, but "the path to thousands of LI, begins with the first step" :)

Advanced methods are confusing, convoluted, you can't figure it out without a bottle or salts, while you need to understand the principle, the gist.

 
kapelmann:

The drawdown is at least two times less than the profit, although it is desirable that it was not at all.

These are advanced methods, MLP and RandomForest, but "the way to thousands of LI, begins with the first step" :)

Advanced methods are confusing, convoluted, it is impossible to understand there without a bottle or salts, and you need to understand the principle, the gist.

Let's not promote alcoholism and drug addiction. Both go to the morass, not to the understanding of ME.
 
Maxim Dmitrievsky:
Random Forests Predicts Trends - Scam Link from the Rooster. Do not go there if you do not want to be banned for a few hours for bruteforcing mql site. I ask moderators to check it.

I understand that I do not want to learn how to make money good people want to take the hard-earned money, with the help of scam links! But there's no way, I'm not so easy to fool!