Machine learning in trading: theory, models, practice and algo-trading - page 971
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Sanych, what was the result in the 18th?
I don't understand.
The model itself is also trained to 35.5% error?
It seems to me that there is no sense to use all ticks, if NS can be trained only on bars. The Open price test will be faster.
The model has nothing to do with it-it's a teacher.
I decided to post some intermediate material.
A trading system based on the classification.
Teacher = Close increments.
The system is a test one: I count close price increments on history. The increments obtained on history I start to send as a signal to a simple Expert Advisor, which ONLY places orders; I form orders obtained on the increments: BUY - SELL, i.e. the system is reversible.
I.e. the trading system ALWAYS looks ahead, into the future. In terms of classification this approach gives 100% prediction, I checked it, because the tester shows quite a different result.
Here is the result from the tester
The balance line is striking - not very smooth.
But it has absolutely amazing values of profitable and loss-making trades = 64.51% by 35.49%
And here is a picture of a loss-making long position followed by a short position:
So much for the obvious teacher that is so widely advertised here!
PS. I note that only the spread can't explain the loss of the shown long.
Well, if the teacher counted by Clowes, then the orders should be opened by him and not by the option and without taking into account the environment for longs...
Started participating in numerai again -https://numer.ai/dr_tr
Can you do 0.691616 and 83.33% in python?
There is a test dataset, no special problems to draw any numbers, and the software is a matter of taste, earlier it was possible below 0.5 loglos on the test to do, now they really have something to filter them, each time differently, at the moment below 0.67 is unlikely to get, and numerai life has long determines not the performance of their predictions, but participation in the "proof of burn" of their coins, in the form of "stacking" where on extremely absurd terms foolish people graze their bonuses, or God forbid bought on the exchange coins (( So, the shop closed about a year, given the rate NMR, which all the time almost linearly falling.
They had quite reasonable and interesting initiative, but everything stopped dead, they didn't want to connect return on NMR token with fund or make reasonable betting conditions on their forecast at least, and the classification itself isn't so valuable in real algotrading, enough to spend couple of years in MO to make almost ideal classification on fractions % different from work of super-special teams, which is not so important on market, market is not cauldron.
Well, if you counted the teachers on the clause, then the orders should be opened on it, and not on the option, and without taking into account the environment, for longs.
Corrected. The teacher on the opener. Better, but still depressing.
If it's about the spread, I don't have a chart handy, plays a significant role on M1 and looks like a smooth loss.
Just the last months I constantly use rattle - it is extremely convenient to check thoughts and no problems. It is more convenient to write a script in R for initial preparation of predictors, store it in .RData, and then load this .RData file into rattle.
Multithreading is from here. You can load all cores and neighboring computers as well.
PS.
A word of advice about learning English. It is ridiculously easy to learn, based on self-discipline and a basic knowledge of grammar.
In a couple of months there will be no problems with English, and the problems with the meaning of words in Russian or in English will come to the fore.
About Rattle - the program should work correctly out of the box. While I see the need to dance around it. I am not able to write scripts, so I cannot assess the convenience of working with .Rdate format. Maybe you have a script, which can be posted here? Well, that it would be possible to convert CSV file to .Rdate with specifying of targets.
About multithreading - thanks for the link, I'll take into account that there are options, although I can't evaluate them. Again, I need to change the code to make it all start working, as I understand it.
Thank you for the advice on learning another language, but I also have difficulty with the Russian language, I think it's my features, which is an obstacle to the study of any language. In general I don't like languages because of stupid grammar rules, their unreasonable and unstable nature.
Here's my question: what programming language are those citizens who are still sitting on pots?
SanSanych Fomenko:
R is the language of STATISTICS. A person who is not familiar with statistics does NOT need this language, and as such, R is the language of professional statisticians. If we consider the place that statistics occupies in trading, in real, professional trading, today NOT knowing R puts under a question mark professionalism of a trader.
It is this circumstance, circumstance of ignoring statistics, unwillingness/unskillfulness to use statistics in trading that I explain my astonishment at first about rejection of R.
This message only shows your ignorance of the history of programming languages. Fortran was once a standard in the banking sector (rumor has it that many servers of the London Stock Exchange still run on Fortran, because there is no one to port the old code to a modern language). The R language was created in 1993 as an open source replacement for Matlab. Now Python has evolved into a full replacement for R. R is a statistics language, but it is no longer the standard in the field.
Perl was popular at one time too, but now only regular expressions are left of it...
I don't get it.
You have a '17 state.
For the '18 is also normal?
Look at Julia, they say it's fast and on JVM. I am interested in possibilities of integration with MT5, I haven't found any information yet.
If you want to estimate the speed of integration with R, you'll need to use R as a graphics language.
If you want to try to integrate R with MT5, you'll have to study it on the website.
Look at Julia, they say it's fast and on JVM. I'm interested in possibilities of integration with MT5, I haven't found any information yet.
I've found information about integration with mt5 and it's difficult to foresee any difference.
You'll have a lot of troubles with this JVM. And firstly in Julia, and then to rewrite for MT - not a royal thing.
Python, after all, is not so complicated.