Machine learning in trading: theory, models, practice and algo-trading - page 1356
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 tried a set of my predictors on EURRUB and SBRF, the eur pours unpleasantly
Training:
Validation:
Test:
But Sber has pleased me - there is a lot to choose from
Training:
Validation:
Test:
Event intervals 2014-2019 test for 2018 - minutes.
Of course, there are not many deals, and probably most of the predictors can be thrown out, but still an interesting result, in my opinion.
I should add, the error balance of EURRUB looks quite good - probably, the profit does not cover the losses - the pair is too nervous...
Oleg, cut it out. Comrade is too smart, and has been trying to get the point across to us for a week now, that until Dad fell down the stairs, we can't say anything about the spectrum of his statements on the subject, because they don't exist.
Yes, it's useless.
Oleg, cut it out. Comrade is too smart, and has been trying to get the point across to us for a week, that until Dad fell down the stairs, we can't say anything about the spectrum of his statements on this subject, because they do not exist.
Radio amateurs avoid thinking either when calculating or even when petrolling. Statistically, falling down the stairs is often the cause of death, which makes the very existence of the statements of the faller unpredictable.
Radio amateurs also avoid thinking when calculating and even when petrolling. Statistically, falling down the stairs is often the cause of death, which makes the very existence of the fallen person's statements unpredictable.
The radio amateur has worked all his life in a dangerous radioactive industry, guarding warheads, hence his vivid surrealism in his speeches.
By the way, Maxim, I did a sample balancing for different situations (significant for the strategy) for Si, indeed learning on the test and validation samples improved - accuracy increased on the validation sample by about 5-7%, but according to preliminary data on the test sample results slightly worsened in monetary terms. I think this method is very correct for stationary models, but in cases of a changing market model it may not be quite right, but it's not the final choice yet - I'll syd a couple thousand models and I'll be able to make a final decision.
The radio amateur has worked all his life in a dangerous radioactive production, hence the vivid surrealism in his speeches
It makes you doubt the reality of nuclear parity)
By the way, Maxim, I did a sample balancing for different situations (significant for the strategy) for Si, indeed learning on the test and validation samples improved - accuracy increased on the validation sample by about 5-7%, but according to preliminary data on the test sample results slightly worsened in monetary terms. I think this method is very correct for stationary models, but in cases of changing market model it may not be quite right, but so far it's not the final choice - I'll do a couple of thousand models and then we can make the final decision.
Next we need to think about how to balance everything else. I don't think i'll make a final decision if i'll sit and sid and then i'll have to think about the rest.
then we need to think how to balance everything else. In any case, if the train and validation are not balanced, then the model does not learn anything, much less catbust, which evaluates everything by validation.
By validation comes to a halt, which means that if further sampling is more like validation than training, we're not learning something that won't happen again. Another thing is that there are often intermediate junk trees between the improvement of validation readings and the training itself - we should weed them out...
The main question here is whether the past is repeating soon, how the present is changing, at what rate or jumps - the answers to these questions would give a lot of information on how best to build a sample.
Main problem is lack of data, I have less than 10k lines to train and validate.
Well, the main problem is the lack of data, I have less than 10k lines for training and validation.
This is very little, especially if you do not know what you're looking for. Yes, and you know, not the fact that there is.
General words, of course, but you have to look for something that is everywhere and always and regularly repeated. Otherwise you can not teach the MO.