Machine learning in trading: theory, models, practice and algo-trading - page 2281
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
Why a million and not 100 or 100 million?
Why should it be earned and not unearned?
why so much gibberish in just 7 words? )
Yes, I should have put the word Perhaps in front.
The figure is from today. Today the MO gives out from 100 to 1000 rules If That. Well, at least with Max. And it works for a period of a year, two on one symbol.
For it to work long and on different characters of the rules probably need 1000 times more rules.
A figure from today. Today, the MoD gives out between 100 and 1,000 rules If Tho.
you see... everything is relative...
If you cram in 10k shitty features, you'll have 2k rules...
if you squeeze all 10k signs into 2 components of some algorithm, you'll have 10 rules no worse than those 2k
so... everything is relative...
for it to work long and on different characters of the rules you probably need 1000 times more rules.
you need quality, not quantity ... and vice versa the fewer rules, the better it will work in the future, read "curse of dimensionality" wikipedia
Nikolai Kamynin, for example, has about 3k rules and 300-500 attributes.
But I'm sure even if he told us what he was doing it would be 50k rules and 100k attributes and it wouldn't work...
You know... it's all relative...
If you cram in 10k shitty features, you'll have 2k rules...
if you squeeze all 10k attributes into 2 components of some algorithm, you'll have 10 rules no worse than those 2k
so... everything is relative...
you need quality, not quantity ... on the contrary, the fewer rules, the better it will work in the future, read "curse of dimensionality" wikipedia
Nikolai Kamynin, for example, has about 3k rules and 300-500 attributes.
But I'm sure even if he told us what he was doing, it would be 50k rules and 100k attributes and it wouldn't work...
Of course it's not linear) I'm based on what I think is a fairly close-to-reality estimate, and today's capacity. Today 1000k rules and 3k - 10k signs is unrealistic. just by capacities. In about 2 to 5 years, I may be wrong about the timing.
Today, just look for successful algorithms to get traits and rules.
It seems that any approach to forward price forecasting based solely on recent price data (and MA's, among others) would amount to "buy when it's going up". )
You may make your forecast using neurons or a fortune-teller.
I mean if you don't select/find predictors for a model of real importance - success is either by chance or fitting (while adaptive selection of a MA point period is a completely inborn idea)
And predictors may be in absolutely different planes and not directly related to the price data. What do you think?
What do you mean by predictors? News, politics, volume/interest, indicators?
In my opinion, the price takes into account everything. The question is how to pull this information out with algorithms.
Of course not linear everything) I proceed from, in my opinion, close enough to the reality of the estimate, and today's capacities. Today 1000k rules and 3k - 10k signs is unrealistic. just by capacities. In 2 to 5 years, I may be wrong about the timing.
It's easy, even a million or 10, the question is quality.
Today we just have to look for good algorithms to get the features and rules.
You are all mixed up, it's just for the algorithms that the power is not mature enough.
What do you understand by predictors? News, politics, volume/interest, indicators?
I think price takes everything into account, the question is how to pull that information out with algorithms.
A good predictor can be any chip on which the model can find a stable relationship on the predicted indicator.
That is, it's a subject of search and experimentation, and not just in the chips themselves, but in their normalization and interpretation.
The price takes into account everything - it reminds me a little of the Fatum theory )
I think that even if you don't argue with this statement, there is still a lag, and not in the sense of a time delay between news and the reaction to it, but in a more global sense
- If there is a factor that has a trend effect, its effect may be long, and it is not necessary to anticipate it beforehand, it is enough to "see" it as soon as possible in fact.
a good predictor - it can be any chip, from which the model can find a stable dependence on the predicted indicator.
That is, it is a subject of search and experimentation, and not only in the chips themselves, but also in their normalization and interpretation.
The price takes into account everything - it reminds me a little of the Fatum theory )
I think that even if you don't argue with this statement, there is still a lag, and not in the sense of a time delay between news and the reaction to it, but in a more global sense
- If there is a factor that has a trend effect, its effect may be long and it is not necessary to anticipate it in advance, it is enough to "see" it as soon as possible in fact.
Markets live the future for 3-6 months. The price chart can tell you a lot, including volumes and news. You can make money if you are the fastest or the smartest or the richest.
It's easy, even a million or 10, the question is quality.
You got it all mixed up, that's just for algorithms the power has not yet matured.
No, that's right. Exactly because even for today's algorithms the capacity hasn't matured enough, we need to look for more efficient algorithms for today's small capacities.
Can you share the information:
Make R connectors with R connectors.