Machine learning in trading: theory, models, practice and algo-trading - page 2090
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
https://pyts.readthedocs.io/en/latest/auto_examples/transformation/plot_rocket.html
At first I thought this thing could fight overtraining
But then it became clear that it was just a trait transformer... a good transformer. But I want less overtraining.
Maybe I should take a longer sampling period for learning?
Is it possible to output the obtained attributes as a clear code in MQL, or to get a file with the result?
I ran in the tester, the left column expectation. Clearly there is a dependence on the day of the month, and catbust put 0.
Outputs are more diverse.
If my article comes out, there in the model the importance of time (hours) in the first place - perhaps a lot depends on the target.
By time you can filter volatility, which means that stops and takes should depend on time - maybe this should be taught - dynamic market exit points?
If anyone makes a sample with news - ready to run in CatBoost with different parameters - this topic is interesting to me.
I tried to change type of trawl depending on expected news - the effect was good, but somehow sample collection was not automated.
And yes, according to my research waiting for news better than its coming out - smoother that way...
Forum on trading, automated trading systems & strategy testing
How to use machine learning in trading: theory, practice, trading and etc.
Valeriy Yastremskiy, 2020.11.05 10:41
I don't know how to use it so I don't know what to do with it. I myself did not yuzal. The other fascinated.
The archive from here.
All the rest only parse from sites on their own. and pre-grading is also not.
It is interesting, why there are 4 columns with indicators and what is in the last one - the figures are not clear at all.
I understand that it's better to look at the USD index, and if there is a movement detected there, then it is already translated to a specific pair...
If I want to understand how to set the time in Moscow :)
It's interesting, why there are 4 columns with indicators and what's in the last one - the figures are not clear at all.
I understand that it's better to look at the USD index, and if there is a movement detected there, then translate it to a specific pair...
If I want to understand how to convert it to Moscow time :)
Forum on trading, automated trading systems and trading strategies testing
Computer learning in trading: theory, practice, trading and etc.
Aleksey Nikolayev, 2020.11.05 16:08
There are transitions there - among news marked with N
As it's parsing FF calendar, first three numbers - value, forecast, previous revision; the fourth one - I don't know, may be the previous revision before it; the last integer - from chart address.
I thought it was Greenwich Mean Time. In Russia you need to remember when and where the clock was turned and whether it was already turned by that time - I do not remember....
I do not understand how to set it in seconds, so I will not bother.
Exactly right - short for Low, Medium and High Impact Expected. There is also N from Non-Economic - holidays and time translation.
Simplification is probably possible, but you need more research - for example, it may turn out that Medium on the dollar is stronger than High on the other currency.
Another problem - in one day may be a lot of news for one country - different in time, strength and direction.Somehow we have to carefully "bring it all to a common denominator.)There are a lot of problems here.
For example, what do you want to find?
What should happen? (or not to happen, so to speak the null hypothesis ....)) )
and in general, such task, imho, faster to look in database, like to unload ZigZag? with dates in one table, and in the second parsing news with categories of importance
But then again... what do we want to take as the null hypothesis?
1) The price has such a spectrum, if you take Fourier from the price the maximum amplitude will always be the slow fluctuations.
2)Meditating))
By eye it is not clear what to do with it, maybe MO will find
1) and why is it bad?
2) Are you the kind of guild that believes there is some kind of magical frequency that drives the market? :)
I see fourier in a different capacity altogether, as a tool for pattern recognition...
You can recognize images in Euclidean geometry, just take a piece of price --- normalize --- look for similar ones
But that's a rather crude method of comparison...
What's good about Fourier, here's what I see the advantages of
1) it has absolutely clear and unambiguous parameters amplitude/frequency/phase, so there are no free wills, and this is good
2) You can delete/add/change (filter) some harmonics from the pattern, thus it is possible to achieve more accurate tuning in the search of similarity, and that is good.
and the cherry on the cake
If we normalize the amplitudes of harmonics relative to their neighbors, we get the invariance to the pattern size, the network will see and understand the same pattern from ticks or from a monthly timeframe; it should work even better than convolution nets, and it's cool ))
I remember you were looking for where to open orders, what went wrong? )
And I thought it was Greenwich Mean Time. In Russia you need to remember when and where the clock was turned and whether it had been turned by that time - I don't remember....
And how in seconds, so as not to bother - I do not understand.
It says, the time of New York, it should be brought to the terminal time of the tester.
(Sorry, you can not, it's in seconds since '70 in the MCL.) And if you translate it just subtract, and if the date converted to a clock time, must be wiggle and date changes take into account).