Machine learning in trading: theory, models, practice and algo-trading - page 2384

 
Maxim Dmitrievsky

were you able to do manual markup? this technique should be more interesting

 
Evgeni Gavrilovi:

did you manage to do manual markup? this technique should be more interesting

I did not, this approach does not interest me

 
mytarmailS:

I found a way to generate such rules that work on both the tray and the test, finally MLEEP....

my broker, my trainee and the test



YOUR broker YOUR data and test in 9 years passes !!!

The grail is found!!!!!!!

The funny thing is that I will disappear from the forum tomorrow, something like a month, very unsuccessfully (I want to share but no time, when I can write off what and how to do.


For comfortable trading need to make 200-400 of these rules(or patrenov if you want)

To understand the scale, my feeble laptop can mine 5-8 rules per day

i see nothing wrong with mining rules, i wanted to make a visual miner (very fast) with sliders to search for patterns

but, as usual, haven't done it yet.

 
Maxim Dmitrievsky:

I don't see anything wrong with mining rules, I wanted to make a visual miner (very fast) with sliders, to search for patterns

But, as usual, I haven't done it yet.

What did you want to output and how?

 
Valeriy Yastremskiy:

What did you want to withdraw and how?

like boxplots

 
Maxim Dmitrievsky:

like boxplots

Training results or boxplot rules that have been tampered with?

 
Maxim Dmitrievsky:

I don't see anything wrong with mining the rules.

Of course you don't)), and even if you did, it wouldn't affect me in any way)

Maxim Dmitrievsky:

I wanted to make a visual miner (very fast) with sliders, to search for patterns

but, as usual, not done yet.

Try it, but the result will be at best like a forrest rendering, that is, none...


Forrest output is a sum of triggered rules, rules are not filtered out and rejected in any way, and rejected rules are about 100%)

Rules are tested neither for repeatability (there may be only one response), nor for adequacy (does it work); rules are simply stretched to the data (the model is fitted to the data)

The model approximates a training sample randomly, hoping that cross validation will help, but it fails due to objective reasons (there are too few important events in the market)


I tried a different approach, I don't fit the model to the data, but I form hypotheses and check them.

1) I form plausible(already filtered) hypotheses in the form of rules.

2) Hypotheses are tested on small data

3) Hypotheses that have been tested on small data are tested on large data

In fact, out of a million plausible rules, only one remains

It is difficult for an untrained reader to understand the difference between the two approaches, but the difference between them is an abyss.

 
mytarmailS:

Of course you can not see)), and even if I saw it would not affect me in any way))

Try it, but the result will at best be like a Random Forest, i.e. none...


The output of a Forest is the sum of triggered rules, the rules are not filtered out and rejected in any way, and rejected rules are about 100%)

Rules are tested neither for repeatability (there may be only one response), nor for adequacy (does it work); rules are simply stretched to the data (the model is fitted to the data)

The model approximates a training sample randomly, hoping that cross validation will help, but it fails due to objective reasons (there are too few important events in the market)


I tried a different approach, I don't fit the model to the data, I form hypotheses and check them.

1) I form plausible(already filtered) hypotheses in the form of rules.

2) Hypotheses are tested on small data

3) Hypotheses that have been tested on small data are tested on large data

In fact, only one out of a million plausible rules remains

It is difficult for the untrained reader to understand the difference between the two approaches, but the difference between them is a chasm

So show me an example of the TC results

 
Maxim Dmitrievsky:

So show me the results of TC

i can't yet, i need to mine at least 500 "dirty" rules of which the final check will pass 10%...

Just to give you an idea, I made 2 "dirty" rules last night


I am working on speeding up rule synthesis, today I figured out how to increase speed by 5 times and rewrote my code.

 
mytarmailS:

I can't yet, I need at least 500 "dirty" rules of which the final check will pass 10%...

Just to give you an idea, I made 2 "dirty" rules last night.

I think Alexei was offering processing power, he likes to do long calculations, maybe you can do a co-op :)

on R without vectorization will still be slow. You can use some fast database