Another way to present multi-dimention data

Dear Group,

Just want to share another way to present multi-dimention data. After the optimization process, it happens to me that sometime it is necessary to explore the iterations rather than taking the "optimized" solution. How to present the multi-dimention (one gene add one dimention) data to explore it visually? Other than presentation used by Octopus and Lmnts (http://lmnts.lmnarchitects.com/parametrics/processing-and-optimizat...), here is another attempt. Inspired by http://disi.unitn.it/~brunato/webmining/2010/assignment1/Enrico_Sar....

Using X and Y for the two most critical genes and RGB for the other three optional genes. Data values are grouped into grids and the highest value is presented as ratio to the grid area.

Attachments:

5DPlotGridRGB.gh, 41 KB

Replies to This Discussion

Permalink Reply by Thian Siong, Choo on June 20, 2013 at 9:51am

Hey Xiaoming,
Interesting proposition on visualizing multidimensional datasets. Just curious if it is possible to identify the Pareto front from your diagram?
I might sound old school but I think the Pareto front is an important element in analyzing data generated from multi-objective optimisation. It allows one to quickly pick out designs with top ranking performances with different trade- offs.
Best Regards
Thian Siong

Permalink Reply by Xiaoming Yang on June 22, 2013 at 2:50am

Hi Thian Siong,

I have to admit I don't know much about Pareto front. I just checked online. There is a difference I noticed between multi-objective optimization and what I presented above. For multi-objective optimization, for example you have f(x,y)=ax^2+by and g(x,y)=cx^2+dy.(a,b,c,d are constants) What you want to achieve is maximize f and minimize g. That is where you can use Pareto front to study the trade-offs.

I presented above is a different problem f(x,y,z,p,q) = ax^2+by^3+cz+dp+eq^2(a,b,c,d,e are constants). There is only one objective, f(x,y,z,p,q) which is presented as ratio to the grid.

This is my understanding, correct me if I am wrong.

Xiaoming

RSS