Machine learning in trading: theory, models, practice and algo-trading - page 768

[Aleksey Vyazmikin]

Mihail Marchukajtes:

In general, I trade on forex, the data is taken from the CME exchange through KD. That's why it's relevant...

I am also interested in FORTS. I checked the TS on Si, it is non-cash dollar to ruble on moex. Moscow Exchange. The result is identical in quality. The problem is that there is MT5, I have done 99% of MT5, but I have found that the way I have organized work MT5 for some reason is stupid. It is too difficult for MT5. It is too complicated calculation for MT5. If there is an opportunity to simplify it, but I do not know how, and it's a pity, Si is a quite volatile instrument and money is deposited within the country, so if there is interest in promoting TS for FORTS, I would be grateful to......

I will tell you what the problem is.....

.

If you create a branch, maybe someone will help you...

[Mihail Marchukajtes]

As promised, I'm posting the results of training according to the method of Dr.Trader....

> cat("Accuracy on train data:", Accuracy(y_pred = predictionTernTrain, trainTable[,TARGET_COLUMN]), "\n")
Accuracy on train data: 0.6666667 

> cat("Accuracy on test data:", Accuracy(y_pred = predictionTernTest, testTable[,TARGET_COLUMN]), "\n")
Accuracy on test data: 0.6666667 

And here on recommendation of the Trickster did not begin to do, can who will make and will lay out. I attach the training file ....

library(corrplot)

#  Сгенерим данные

data <- seq(0, 2*pi, 0.1)
x1 <- 5*cos(data)
x2 <- 2*sin(data)
target <- ifelse(x1 + x2 > 0, 1, 0)

#  Соберем в датафрейм

z <- data.frame(data, target, x1, x2)

head(z) #  Смотрим что получилось
str(z)  # + структуру

#  Посмотрим диаграммы размахов ("ящики с усами")

boxplot(z[,-1])

#  И корреляционную матрицу

corm <- cor(z[,-1])
corrplot(corm, method = "color", addCoef.col = "darkgreen",
addgrid.col = "gray33", tl.col = "black")

[Ivan Negreshniy]

Mihail Marchukajtes:

As promised, I'm posting the results of training according to the method of Dr.Trader....

And here on recommendation of the Trickster did not begin to do, can who will make and will lay out. I enclose the file of training ....

[Maxim Dmitrievsky]

Mihail Marchukajtes:

But I did not do it as recommended by Trickster, maybe someone will do it. I attach the training file ....

and you think he has something good to say?

he's been playing with R for God knows how many years, now he's laughing his ass off without earning anything :)

it's strange that your test and trein are equal, there's a mistake somewhere

and i guess you still have the same small set, right? then there's no point in doing it at all

[Ivan Negreshniy]

Maxim Dmitrievsky:

Do you think he's giving good advice?

He's been playing with R for who knows how many years, and now he's laughing his ass off without earning anything :)

it's strange that your test and trein are equal, there's a mistake somewhere

and looks like you still have the same small set, right? then there's no point in doing it at all

Yes, a set of 30 lines is not serious... If at least tens of thousands, I could run the training and even generate the source code, just now testing

[Ivan Negreshniy]

And here is the correlation matrix from the zigzag chart

[Maxim Dmitrievsky]

Another good article on RL

https://ac.els-cdn.com/S2212567112001220/1-s2.0-S2212567112001220-main.pdf?_tid=cde6f522-59e2-41c6-b12f-e1719dca6bad&acdnat=1521966710_8eeed067d4b093ed5ba6acacac6a0efd

[Dr. Trader]

Mihail Marchukajtes:

As promised, I'm posting the results of the Dr.Trader method training ....

What will the following code print out
cat("Best R^2 score:",max(gaResult@fitness),"\n")
if it is called after selecting parameters and creating a model? It has to be greater than 0. Very good, if it is greater than 0.5. Ideally if ==1.
This is the result of crossvalidation on training data only.

[Dr. Trader]

Maxim Dmitrievsky:

It's strange that your test and trein by akurasi are equal, there is a mistake somewhere

The guru always have equal test and trein by akurasi, all is normal

[Mihail Marchukajtes]

Ta no, the test and the train they are the same, because the test data is not yet available, they will appear in a week. That's what I did just to see what's up...

Now I'll try to add the code and see in R