Hi,
I have an idea for a machine learning project and I need some feedback.

In the article [0] https://arxiv.org/pdf/1711.01519.pdf  A multinomial  
regression is used to find the optimal chunk size and prefetching  
distance. However, a multinonial regression is actually a  
classification algorithm so it can only choose values from a finite  
set. This allows to get a pretty good value for the output variables  
but it isn't the 'optimal' values. I propose using a regression  
algorithm instead that could output continuous values.The first one  
that comes to my mind is Nearest Neighbor Regression but I'm sure  
there are other ones that could be used. This would allow the  
regression to output the 'optimal' value of chunk size and prefetching  
distance instead of choosing the best one from a finite set of values.

Then, I would like to implement this in HPX as an alternative to the  
multinomial regression and compare the performance of both algorithms.  
This will allow me to see if such a precision is really needed on  
chunk size and prefetching distance or if a classification algorithm  
is good enough. To be able to compare both algorithm, I assume I will  
have to use the same data that was used in [0].

What are your thought?
Thank you.
Gabriel




_______________________________________________
hpx-users mailing list
[email protected]
https://mail.cct.lsu.edu/mailman/listinfo/hpx-users

Reply via email to