Machine learning in trading: theory, models, practice and algo-trading - page 1552
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
I'm basically looking for parts for the bot, writing videos for myself. anything interesting I see - I write it down.
to tell the truth, there hasn't been a single article that I've borrowed anything directly very useful from, just a little bit from everywhereYes, it is clear that not always there is something to borrow...
But about CatBoost - apparently it's not very suitable for time series, as well as other models - all of them don't take into account repeatability of pattern (sheet) on sampling history, its distribution over sampling - it's very important.
Yes, it is clear that there is not always something to borrow...
But about CatBoost - it seems not very suitable for time series, as well as other models - all of them do not take into account repeatability of pattern (sheet) on sampling history, its distribution over sampling - it is very important.
In what way does it not take into account? under-sampling of some patterns?
You can accumulate them.
In general, in itself it is very cool and handy, + improving all the time. They say it is ahead of xgboost in almost everything. But it's still an open question - what is better for time series?
In what way does it not take into account? Under-sampling of some patterns?
you can accumulate them
As far as I understood, CatBoost generally takes a random window in the sampling algorithm to calculate the split to 64 values (just don't know if this applies to categorical predictors or to all).
The point is that most algorithms don't care if in 1/10 of a sample the activation of the leaf occurred or if it was distributed all over the sample - I believe that the distribution should be all over the sample (let's say every 1/5 of at least 10-15%), and you should consider the economic indicators in this case besides statistical - this is what I do, checking the sheets separately.
In general, itself, it is very cool and convenient, + constantly improving. They say it has overtaken xgboost in almost everything. But it's still an open question - what is better for time series NS or boosting?
As I said before, according to developers' opinion if predictors are similar to each other, in the same units of measurement, then NS is better, but I don't have returnees, but you should try NS.
As I said before, according to the developers, if predictors are similar to each other, in the same units, then NS is better, but I do not have recurrences simply, and you would try NS.
for increments exactly recurrence fit well and their modifications, then I'll try
Recurrence ones and their modifications are good for increments, I'll try them later
Anything is possible, it is necessary to try.
On the last video - I don't agree that the code will not change the interest - if a person is just trying his forces in python and MO, it will be more interesting to watch, and the questions may be on the merits. Although, the audience can be difficult to understand, and yes, not everything at once.
On chips - wouldn't it be better to try different linear formulas for selecting increments rather than just random? Maybe we should have three returns with an offset from 1 to 10 and thirty with an offset from 10 to 50.
A little criticism) Maxim I do not ***understand.....
Tip, to understand what you're conjuring there, you need to see the code. So post the code for each video on the blog or somewhere, just open source, without any files.
Put comments in the code. Then you will be able to use pieces of code in practice and make some arguments. But for now it's just video clips with musings aloud)
P.S. What's the library, which regulates the scale of the graph, works only in the browser?
Not everything can be done in code.
For example, placing pending orders(some of their variants)
A little criticism) Maxim I do not ***understand.....
Tip, to understand what you're conjuring there, you need to see the code. So post the code for each video on the blog or somewhere, just open source, without any files.
Put comments in the code. Then you will be able to use pieces of code in practice and make some arguments. But for now it's just video clips with musings aloud)
P.S. What is the library, which adjusts the scale of the graph, works only in the browser?
I'm crazy answering all the code, time is not enough
I'll send you some intermediate version later.
https://kernc.github.io/backtesting.py/