Machine learning in trading: theory, models, practice and algo-trading - page 1395

 
Maxim Dmitrievsky:

that's not how it works, there's a 45g slant line, not horizontal

don't count the deviations... even suggested that you don't know what))
The horizontal line is zero in real terms, just to be clearer.
On the left side all below it (50 trades) is in profit, all above it (10 trades) is in loss. On the right side it is vice versa.
 
elibrarius:
the horizontal line is the zero of the actual result, just to be more visible.
On the left side, all below it (50 sticks) are in profit, all above it (10 sticks) are in loss. On the right side is the opposite.

is the ratio of current values to predicted values, the line is drawn at 45 degrees through the cloud, just sampling at zero centered. The variance there is huge.

You could have just given the average error of the model

Kindergarten
 
Maxim Dmitrievsky:

is the ratio of current values to predicted values, the line is drawn at 45 degrees through the cloud, just sampling at zero centered. The variance there is huge.

Kindergarten is shorter.
On the axes I see Real and Predict. Not the ratio.
 
elibrarius:
I see Real and Predict on the axis captions. Not the ratio.

Don't piss me off ) connect the predicate values and the real values with a grid, there is a diagonal

or google scatter plot

 
Maxim Dmitrievsky:

Don't piss me off ) connect the prediction values and the real values with a grid, there is a diagonal

or google scatter plot

Calm)
I do not understand you. Draw your vision of where the predictions in this picture are correct and where they are not.
Or let's wait for Yuri to explain what's in his picture.
 
elibrarius:
Calm)
I don't understand you. Draw your vision of where the correct predictions in this picture are and where they are not.
Or wait for Jura - let him explain what's in his picture.

correct predictions lie on the line, everything else is an error

Calculate the square of the deviations from the line and you get the NS error

the smaller the error, the smaller the spread around the line

 
Maxim Dmitrievsky:

the correct predictions lie on the line, all the rest is error

I don't strive for the absolute correctness of a prediction. For me all the trades that made a profit are correct.
Examples:


1) predicted -10 got -8 - that's an excellent profit, not an error at all
2) they predicted -4.8 and got -13 - much more profit than predicted.
3) they predicted -3.5 and got +5, there will be a loss - this is a mistake. Just like all points to the left and above 0. Only they will bring losses and trading on them is an error.

 
elibrarius:

I don't strive for the absolute correctness of the prediction. To me, the correct ones are all those that make a profit.
Examples:


1) predicted -10 got -8 is an excellent profit, not an error at all
2) predicted -4,8 got -13 - much more profit than predicted, it's certainly not an error.
3) they predicted -3.5 and got +5, there will be a loss - this is a mistake. Like all points on the left and above 0.

mnde...

sausages on skewers are better to cook, it will be more useful

A robust model looks something like this, for example (first google image)


 
elibrarius:

Let's better teach the perseprtron from the alglib to pre-learn, eh?

every time a new tray runs, the weights are randomized, remove the randomization and try to pre-learn, as you can do in all normal packages

Can you imagine how many interesting things you can do with it?

 
Maxim Dmitrievsky:

Let's better teach the perseprtron from the alglib to pre-learn, eh?

every time a new tray runs, the weights are randomized, remove randomization and try to pre-learn, as you can do in all normal packages

Imagine how many interesting things you can do with it

I'm into forests. If I go back to the NS, it won't be for a long time. I've already spent a year on them.