Machine learning in trading: theory, models, practice and algo-trading - page 2044

 
Maxim Dmitrievsky:

You all believe )))

 
Aleksey Vyazmikin:

If I understood correctly from the video, there is a function/library, which searches for features in the convolutional network, i.e. ready-made patterns by which patterns/predictors should be found - I wonder what is expected to be found there, how this mask was made - what is the logic, do you know by any chance?

it picks up the mask from the weights by itself when training for a given pattern length

It just passes a window on the feature vector, the window is smaller than the number of features. A convolution takes place.

 
mytarmailS:

You all believe ))))

our motto is invincible

 
Aleksey Vyazmikin:

I'm not sure that our inputs are suitable for this network - everything looks smooth in the pictures.

Any data, any time series

This is the technology of the present and the future of the type
 
Maxim Dmitrievsky:

it itself picks up a mask from the weights during training for a given pattern length

It just passes a window on the feature vector, the window is smaller than the number of features. The convolution takes place.

It seemed that the speaker talked about some ready-made mask solutions, hmm, maybe I got it wrong.

 
Aleksey Vyazmikin:

It seemed that the speaker was talking about some kind of ready-made mask solutions, hmm, I must have misunderstood.

a mask is a sliding window

 
Maxim Dmitrievsky:

any data, any time series

These are present and future technologies of the type

Maybe a vector will show movements, but aren't fluctuations more important to us? Maybe we need to look for points with potential strong outliers?

 
Aleksey Vyazmikin:

Maybe the vector will show movements, but aren't fluctuations more important to us? Maybe we should be looking for points with potential strong outliers?

What?

 
mytarmailS:

You all believe )))

20 kliks.

see what's going on.

Epoch 820 train err: 0.4057188332080841 tst err: 0.4114921987056732

you have to write a tester to see

 

I ran in the tester, the left column expectation. Clearly there is a dependence on the day of the month, and catbust put 0.

Outputs are more diverse.

Files: