Machine learning in trading: theory, models, practice and algo-trading - page 1339
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
Then you are in art, not science, because scientific knowledge is hierarchical, from the simple to the complex and there is continuity.
And good programmers and mathematicians have problems with imagination, according to you.
Fantasy is something naive. Abstract thinking is more correct. The height of abstractionism is mathematical formulas and abstract concepts, not mush in the mind of some particular fantasist
so I wouldn't position myself so that they're like mathematicians and I'm a fantasist and I'm full of ideas.
+++
No offense to Alexey - strike me dead if I understand a word of what he writes. Neither the goals nor the methods of achieving them are clear or substantiated. The spirit of the Teacher, who spent 15 years on neural networks, and now works at a car wash, so hovers over him.
Then you are in art, not science, because scientific knowledge is hierarchical, from the simple to the complex and there is continuity.
and good programmers and mathematicians have problems with imagination, according to you.
Fantasy is something naive. Abstract thinking is more correct. The height of abstractionism is mathematical formulas and abstract concepts, not mush in the mind of some particular fantasist
so I wouldn't position myself so that they're like mathematicians and I'm a dreamer and I'm full of ideas.
I do not seek scientific justification for my words and methods - this is easily done by you, constantly referring to the fact that so already did and invented and give different scientific names to this.
And fantasy and abstract thinking are different processes - fantasy is a process of creation, and abstraction is a process(way) of presenting information.
You're missing the point - humans have strengths in their constitution and it's the development of those that will yield greater efficiencies for the person and the cause they're engaged in.
I'm not saying that mathematics is not necessary for success, on the contrary, I'm saying that you need a person well versed in it who can help you understand the nuances of ideas contained in formulas!
+++
No offense to Alexei - strike me dead if I understand a word of what he writes. Neither the goals nor the methods of achieving them are clear or substantiated. The spirit of the Teacher, who spent 15 years on neural networks, and now works at a car wash, so hovers over him.
Hurry up and find a grounding, or thunder will strike :) The problem is that you look for confirmations of my ideas (about which of ideas - a herbarium?) in scientific works, and apparently you do not find them there that leads to the conclusion of their lack of validity not by the author of ideas, but by an authoritative scientific opinion.
It's not even a year since I started doing MO, and, as you can see, somehow I get working models, which is not so bad. It took me three years to make my forex advisor work. I use return to the average as a result of trend decay - there are sets of Expert Advisors that have been working for years - look for the signal. There are some sets of Expert Advisors that are more than a year old - have a look at the signal. Where is your signal to assess the achievement?
Urgently find grounding, or thunder will strike :) The problem is that you look for confirmations of my ideas (about which of ideas - a herbarium?) in scientific works, and apparently you do not find them there that leads to the conclusion of their lack of validity not by the author of ideas, but by an authoritative scientific opinion.
It's not even a year since I started doing MO, and, as you can see, somehow I get working models, which is not so bad. It took me three years to make my forex advisor work. I use return to the average as a result of trend decay - there are sets of Expert Advisors that have been working for years - look for the signal. And where can I find your signal to assess its performance?
My signal is on the contest. It is about 30-40% per month. Moreover, according to the method that we once discussed in the PM, only finalized. What the hell are you doing in the NS? This is a mystery...
Why would a mathematician help a non-mathematician by explaining formulas to him when he obviously has a stronger hand and a much better understanding of what is going on?
he might just condescend to the dreamer.
You just don't know what you're writing about. You took a ready-made strong solution in the form of catbust and started diluting it with your weak fantasies, that's what you're doing. Of course, by the law of large numbers eventually you may stumble upon a good solution to a problem by accident, but only by accident
Do you really have such power of abstraction that you are able to model my knowledge and thoughts on the subject? I doubt it.
There are many mathematicians(and others with knowledge) who are constrained by knowledge, and afraid to contradict it - this is the psychotype of such a person. Otherwise, every person who graduated from the institute would continue to work on scientific works, and not exist at the expense of remuneration from the employer.
My fantasies I realize, check and improve - I create system on production and selection of models, I see dynamics, and it is too early to speak about erroneous vector of movement.
If no one is interested in my findings, I can stop publishing them.
My signal is on the competition. Something like 30-40% per month. Moreover, according to the method that we once discussed in the PM, only finalized. What the hell are you doing in the NS? This is a mystery...
This is why I started using MO, because I killed a year before I created a trend EA, which worked perfectly on historical data, and in 2018 it started to fail on fresh data and I improved it, but it kept on failing again. I decided with help of MO to find optimal filter settings, which were already in my Expert Advisor, and started making predictors... In general, MO opened my eyes to fitting any strategy to the history and, since it is very difficult and time consuming to do it manually, I decided to pass my trading experience to predictors and it turned out that I could trade using them without making up cunning dependencies between the ideas (predictors). In general the MO is a tool for combining observations into a solution, I check and select these solutions, and they form what is called a herbarium. That is, unlike Maksim, I have a basic strategy, which is improved at the expense of MO.
I anticipate all your moves ahead of time, because I went through this stage at an accelerated pace (yes, yes, the right track is only one)
and know what is missing to make something work, I recommend literature which has the missing elements
I do not pretend to be a teacher or mentor, just give my opinion. If I start to explain something you will not understand, so the book.
So your next move is to properly balance samples and get rid of a bunch of garbage (predictors).My immediate moves recorded three hours ago on paper - is the analysis of catbust leaves and analysis of the responsiveness of the models on the sample in order to further combine them. With samples there is no answer, for the reason that we have no stationarity, and hence no completeness of observations - I have an idea of uniform distribution of different trading situations on samples, but so far I have not got to the realization. And to remove predictors is not possible yet - combining groups into one - yes it is interesting to realize, but it is not clear how. I am for combinations of predictors, both forced and random.
It will shoot out and is described in the literature as one of the main methods of MO, the rest are garbage ideas
I have some ideas, but I need to test different variants and I don't know yet how to implement the distribution in MQL.
The rest of the ideas I need to understand the process, monitoring, selecting models and their combinations, ie. they are not aimed at improving the models separately, but rather at their evaluation.
Well, good, now I need to understand how to better classify these areas - there are a number of ideas, but I will need to test different options, and I do not know how to implement the distribution itself in MQL so far.
The rest of the ideas I need to understand the process, monitoring, the selection of models and their combinations, that is, they are not aimed separately at improving the models, but rather to evaluate them.
Honestly, I'm here in hope, that your next post will be "here are the first results....", all your researches were applied in practice at least, if not then maybe all your works at the moment is a road to nowhere?
For those who are not friendly with python and R - in the appendix is a batcher generator with basic settings, go through the Seed
The code is closed, unfortunately, for the reason of application not my class for work with tables.
The output will be 7 files:
_01_Train_All.txt //Starting training.
_02_Rezultat_Exam.txt//Applies the model on the test sample and saves the results into a file
_02_Rezultat_Test.txt//Applies the model on the validation sample, saves the results to a file
_02_Rezultat_Train.txt//Applies the model on the training sample, stores the results in a file
_03_Metrik_Exam.txt//Calculates metrics of the model on the test sample
_03_Metrik_Test.txt//Calculates metrics of the model on the validation sample
_03_Metrik_Train.txt//Calculates metrics of a model on the training sample
The files should be renamed to bat. The last 6 batnixes can be run in parallel to speed up the process, but only after the first batnix is finished - so that the models have been created by this time.
In a directory with batniks there should be the CatBoost and 3 samples.
The name of files of samples
train.csv //Train.
exam.csv// Test
test.csv//Validation (used for a stop of training).
The samples must have a header.
TheLabel and theAuxiliary columns of the samples should be specified in a separate text file without an extension (not .txt).
The columns will be numbered from zero.
The files will be placed in the Setup directory, which will be created in the project directory (specified in the script).
Models will be created in a subdirectory of the project called "Rezultat", there will be subdirectories for each model with the name of the setup file with the target and model number.
I will develop script for myself, if anyone interested, ready to share a compiled instance (I can also give the source code, but without a class can not compile it).
Download exe file of CatBoost for working with command line can be downloaded from this link correctly indicate the release version in the script settings.