Machine learning in trading: theory, models, practice and algo-trading - page 1873
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
Anybody dealing with nets, how do you adjust the ratio of epochs, batting size and lenght?
Exactly the same way as the period of stochastic and with exactly the same result
Does anyone do any networking, how do you balance the number of epochs, batting size and laning rate?
What an interesting question you asked! Check out how I do it. Let me remind you that Reshetov's optimizer has a method of reference vectors, which defines the direction of the vector to the hyperplane that divides two classes with each other. And as we know this method is the most labor-intensive one, it's not the method of gradient descent for searching of the error back propagation. It is a bit different, more likely it is a search of the perpendicular plane between two different signs at the same time to take into account the maximum distance from these two classes that this plane is perpendicular to them both. In general, on each vector, there is so much horror. You understand that when you change the next vector, the previous plane is displaced, and during one epoch, there is such a search for all planes that divide the different classes lying next to each other, that it's a nightmare. So I have the number of epochs changing dynamically. Not so great is possible to open. And the logic of the change depends on the number of inputs to the polynomial. With each new input the search increases geometrically. As a result, with 11 inputs I have about 800 epochs, with 5 I would like to write about 3000 but here I get it. I have not downloaded java for a long time, and in light of recent events, when I was frankly fucked because I did not know what RSME and saw those cover it, yes mytarmailS yes ???? So I'm not an asshole. You should think about what I work at first, what would make you say such accusations :-) Just kidding, I'm nice when I'm nice :-)))))))))))))))))))!!!!!!!!!
So here's the exact data <7=1600 epochs, >=12 epochs 30 would be. I understand that I'm fiercely losing power right now, but the Achilles' heel of the method. You need not only the number of cores, but also their frequency. Isn't that adequate compared to other methods. I just don't get it.... To find a method that is easy for a computer is to deliberately limit yourself at the very least... Exclusively IMHO!!!!! And the data I cited is for 32 cores with maximum frequency..... I'm renting it here on the occasion in Mile. Highly recommended.... Handy stuff, not advertising purely fascinated :-)
And yes, mytarmails, I didn't mention you in this post for nothing, because I figured out the assessment you suggested and found it extremely intersssive. I asked myself what would be this estimate of the models I received, provided I don't use it in the optimization process, but just see what it is. There's a feeling that the best model is the one that has the same optimization result (according to training conditions) but has the worst RMSE score. What do you think? Understand or repeat???? Here is not a pi... You've been wanting to tell you for a long time. I didn't have an opportunity, but now I have it :-)))))))
You just need to draw a graph for it to be clear, but it's a pain. Don't bother....
I can't catch the bug with missing quotes )
Let's say, there is a dateframe with dateframe indexes, five-minute indexes. It is grouped by 2 hours. The 2nd hour follows the 1st hour and then the 2nd hour of the next day follows the 1st hour of the next day.
You need to determine in the loop where there was a skip in a 5-minute or a whole hour, and drop a couple of hours if there is a skip in at least one of them. Like dropna or fillna (for nearest values). But there is no NA
You've got problems with your head.)
It seems to me that the profit is not in the clusters but in representation of the information, I think that it is necessary to present the MO information in the form of candlesticks for a larger timeframe, for example an hour should be represented by 5 minutes from the hour opening till its closing and to predict the next hour and it seems to me that it will be no worse or even better than to deal with clusters, but it is only a hypothesis.
I think that the profit in general is not in the clusters, but in the representation of the information, there is a suspicion that the need to represent the MO information in the form of candles of a larger TF, those for example an hour represent a 5-minute from the opening hour to its closing, and predict the next hour, and I think that will not be worse, or even better than to deal with clusters, but it is so, just a hypothesis .
I don't care how, I don't care.
But the hours or minutes are missing somewhere, I'm trying to find out where. Basically the clock junction doesn't match.
Can you just build a bunch of curves on my file like you did before and see? You can do it without clusters.
Take the 5 and 6 clocks.
I get this nonsense, when going from 5 to 6 o'clock sharp jumps sometimes, about the same size. Hardly any real transitions, more likely a skip in the hour and a shift in the day somewhere. The clustering may work crookedly because of this.
I don't care how I do it.
But the hours or minutes are missing somewhere, and I'm tired of looking for it. Basically the clock junction doesn't match.
Can you just build a bunch of curves on my file like you did before and see? You can do it without clusters.
Take the 5th and 6th clocks.
I get this nonsense, when going from 5 to 6 o'clock sharp jumps sometimes, about the same size. Hardly any real transitions, more likely a skip in the hour and a shift in the day somewhere. The clustering may be working crookedly because of this.
I'll try
-------------
What the heck is this?
Is it supposed to be like this?
You got a problem with your head.)
Nah, I just don't like people throwing poo at me....
who wants to practice? Upper chart entry, third exit. 2 and 4 are test charts. Some explanations, 1 graph increment, 3 graph signal to buy some "classic" system. By the way, question for a puzzle, is ns able to emulate ligic and with memory indicators, something like a zigzag and maximum/minimum indicator?
I'll try
-------------
What the hell is this???
Is that supposed to be???
there is a comma separator