Market etiquette or good manners in a minefield - page 64

New comment
 

There, it worked.


 

What does K equal for you?

If it's 1, it's unclear why you have such a bad learning curve... You plot from 0 to 1.5 for certainty. Why aren't you outputting error variance values?

 

This picture at K=2, d=9+1

Here's the same for d=24+1:

And the learning rate is bad for clarity... there is a learning rate of 0.1

 

Construct for K=1, N=100 and d=5 in the range of zero to half and derive the statistics.

 

Did you mean N=100 is the number of eras?


 
Aha! As far as I understand, you save the calculated error over n epochs, over N experiments, in a two-dimensional matrix, and then process it with a statistics subroutine. I'll do it now.
 

That's right.

I have an L-epoch index, an n-statistics index.

 

Something about the deviations seems out of place to me:


 

You mixed up the sign in one of the outliers. You need to subtract and you have added to the average.

And find out why you don't get 1 on the first readout (the graph shows 0.8). This indicates a possible confusion in the indexing of the vectors.

 
Done.
1...575859606162636465666768697071...104
New comment
Reason: