Machine learning in trading: theory, models, practice and algo-trading - page 1548
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
A random forest is a fit to a story with no room for adjustment. I squeezed all the options out of SL a year ago.
The gradient boosting seems to be pre-learned from previous errors.
Linear regression has a much better chance to produce a profit. When training, you should not use real prices but relative ones.
How should I classify it, by slope, or what?
I would also like to ask about MLP in Skit-Learn. It takes a very long time to count and the result is not particularly impressive, is it worth bothering with it?
SDV reference vectors are in question, again the long calculation doesn't fully explore them?
Hello all. I have decided to use freelance services for the first time. I want to switch to Forts for good. Let's see how the tools will cope with my tasks. As for the vector reference method. Yes it is a bit heavy in the calculation, but in the current realities of machine hours is not worth the regret, if you're not in search, but already found. :-)
did you understand that reshetov and svm are the same thing? do it yourself, you will save nerves)
did you understand that reshetov and svm are the same thing? do it yourself, you will save nerves)
Maximka of course I figured it out long time ago as well as long time ago I optimize it on powerful computers with a bunch of cores and in this case the results are more stable.
Wow, that's some news.
it's time to run after a lot of cores tooThat's some news.
it's time to go get a bunch of coresSo where to go, you cardboard fool. I told you that I rent on Mayle at awesome conditions. Consider a server with 32 cores, for a month costs 1000 rubles. Tell me the class? !!!!. But the most important discovery for me was something else. It turns out there is a well-known fact that we all remember at the last moment "price models" which is expressed in machine hours spent on its acquisition. Translate M/H into money and voila. Price my models in real money (measured in fact) 150-200 rubles, and that with the reserve. That is, 1 grand is enough for just the same 5 construction or runs in each run of 5-6 adequate models get to keep. We build once a week and get 3-4 trusted models, one of which we put on the real. I have already mastered so that for me to start the server is associated with mining for time and it's not because I'm greedy for money and started here suddenly save. No, the point here is different. Price model is just another metric for its quality. But it can not be used directly. It's more of a level indicator. And no matter how you spin it, the more expensive the model, the better it can be.
Either I do it on a home computer for 5 rubles apiece, or on a powerful 200 rubles apiece. But there is a fucking difference, take my word for it. Anyway, there is some effect, which I can not explain. I can imagine what kind of data can be optimized at 2000 rubles apiece. First of all, the sample will be larger, and the model will be better overall. So... more thoughts out loud....
So where to run to, you cardboard idiot. I told you that I rent from Mail under awesome conditions. Consider a server with 32 cores, for a month costs 1000 rubles. Tell me the class? !!!!. But the most important discovery for me was something else. It turns out there is a well-known fact that we all remember at the last moment "price models" which is expressed in machine hours spent on its acquisition. Translate M/H into money and voila. Price my models in real money (measured in fact) 150-200 rubles, and that with the reserve. That is, 1 grand is enough for just the same 5 construction or runs in each run of 5-6 adequate models get to keep. We build once a week and get 3-4 trusted models, one of which we put on the real. I have already mastered so that for me to start the server is associated with mining for time and it's not because I'm greedy for money and started here suddenly save. No, the point here is different. Price model is just another metric for its quality. But it can not be used directly. It's more of a level indicator. And no matter how you spin it, the more expensive the model, the better it can be.
Either I do it on a home computer for 5 rubles apiece, or on a powerful 200 rubles apiece. But there is a fucking difference, take my word for it. Anyway, there is some effect, which I can not explain. I can imagine what kind of data can be optimized at 2000 rubles apiece. First of all, the sample will be larger, and the model will be better overall. So... another thought out loud....
The usual effect, when you don't wait half a day and get a bunch of hot cakes in a minute
The usual effect when you don't wait half a day and get a bunch of hot pies in a minute
It's just that the taste of these pies is very different. That's what I wanted to say.
You promised a video. Will there ever be one?
Let me start.
You promised a video. Will there ever be one?
let me start