Machine learning in trading: theory, models, practice and algo-trading - page 3690
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
For those who want to meditate on strategy. Find flaws/advise improvements :)
Any ideas are welcome, here or in the comments of the article.
Still believe in collective intelligence, though increasingly in the form of AI :)
https://www.mql5.com/ru/articles/16457
Another article on the topic. From the same authors, but more compact.
One of the authors is the same, the others are different.
In short, clustering is dead, long live biclustering ).I've written something similar, I haven't got to the tests yet. Curiously, my markup (from the last article) allows to find patterns through biclustering. Whereas on random partitioning, a la by increments or zigzag, the accuracy is only slightly beyond 60, at best. I would never have realised the trick and threw it away if I hadn't compared it for different samplers.
I've written something similar, I haven't got to the tests yet. Curiously, my markup (from the last article) allows to find patterns through biclustering. Whereas on random partitioning, a la by increments or zigzag, the accuracy is only slightly beyond 60, at best. I would never have realised the trick and threw it away if I hadn't compared it for different samplers.
И
Точность: 84.7%It's not very clear what the precision and distribution are. You get that accuracy = distribution of sales.
I thought accuracy should be equal to the number of correctly guessed sales, not the percentage of sales out of the total number of transactions (purchases + sales).
Maybe a typo in the code that shows sales instead of accuracy (in all 3 examples)?
И
It's not very clear what the accuracy and distribution are. You get that accuracy = distribution of sales.
I thought that accuracy should be equal to the number of correctly guessed sales, not the percentage of sales out of the total number of transactions (purchases + sales).
Maybe a typo in the code that shows sales instead of accuracy (in all 3 examples)?
In the description the main signal (direction) and its accuracy. In how many cases the found rule works correctly.
see the original article. This is also called support in data mining.In the description the main signal (direction) and its accuracy. In how many cases the found rule works correctly.
see the original article. This is also called support in data mining.. For all 3 examples = 100%
Accuracies are supposed to be random numbers and 3 times their sum = 100%. Strange.
But okay, you can't see all the details from the outside. Maybe I'll do a repeat of your article when I'm in the mood.
Suspiciously exact match of sums (accuracy of purchases + accuracy of sales)
For all 3 examples = 100%
Accuracies are supposed to be random numbers and 3 times their sum = 100%. Strange.
But ok, you can't see all the details from the outside. Might do a re-run of your article, once I get in the mood.
In that place could be any word, for example your advert )
I needed a way to test whether MO (classification, regression) is bad for finding patterns (of any kind) in highly noisy series. It turns out that it is very bad, and should only be used for predictions on already trained data.
All the horror of data mining in one bottle
Added visualisation of returns for clusters. What would happen if you buy or sell on this cluster. Each box with a moustache is the next bar after the signal. Above zero - positive return, below - negative.
Even though the average for selling for all the examples fluctuates around zero return, it's better than for buying. And there are fewer tails.
So far all this is not very clear, we will have to test it some more.