Machine learning in trading: theory, models, practice and algo-trading - page 2972
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 used becky mailer until the last time, and not many people knew about it, emeditor instead of notepad is not the most frequent choice, but these are tools, and language is a tool, what difference does it make to make a craft? The cuts are the main thing)))) Looks like my shovel is yellow, and it's better than your green one))))))
Although maybe not quite right, the size and configuration of shovels are different)))))
This is a longstanding squabble, back from the middle of the thread. Impossible to stop )
I came up with another "brilliant" idea to transform data for AMO, so that all the information about the price can be fed directly.
I got such a good dataset 20k observations 12k attributes, the target is ordinary ZZ, charged XGboost, got on the test akurasi 62% ))))
Something is clearly wrong with this world, something is clearly wrong....
Something is clearly wrong with this world, something is clearly wrong....
So what's the point of the conversion?
So what's the point of the conversion?
Oh, it's a long story. What's the point if it doesn't work?
Oh, it's a long story. What's the point if it doesn't work?
Thoughts are sometimes right, but the idea is applied in the wrong place....
Thoughts are sometimes right, but the idea is applied out of place....
No, there's no one here to help me.
No, there's no one here to help me.
It's sad when no one understands you.
It's sad when no one understands you.
And what is the point of announcing an accurasi or classification error with unknown financial result, maybe it's time in this thread to give results translated into profits and losses?
I.e. sum/ratio of losses and gains in points/percentages.
And of course on new data. I have even stopped displaying trains in printouts, I don't look at them anyway, and SSD is a pity.
It's sad that machine learning with targeting doesn't work and one-shot learning doesn't work....
I don't know what there is not to learn - for me, something, yes, to learn. For example, I have Accuracy of 70%, but it is still not an objective indicator.
In general, the problem is not in the possibility to get a model that will continue to work, but in the following:
1. Obtaining a stable segment of the initial predictor (for example, based on an indicator), which will retain its statistical characteristics in the future.
2. Selecting a model from a set of models that are more likely to be effective in the future on new data.
I have shown all this in the corresponding thread, and I think it is necessary to solve the problem starting from the first point. Which I am doing, but I need ideas from non-standard approaches of descriptive statistics.
My idea is to get a model, which will select stable quantum segments by a number of statistical features. Anyone interested is welcome to join this project.