Le 27/6/15 00:16, Pierre CHANSON a écrit :
Hola todos,

These days I am working on compete with R. I am taking the datasets and examples from http://cran.r-project.org/doc/contrib/Zhao_R_and_data_mining.pdf and make it with Roassal even better ! Or almost... ;)

I love this idea.
Practicing on real cases is the best.

First chapter first pages, the histograms... In Roassal they were not working so well.

So lately I restructured the RTDistribution package. This, by default make a classical frequency distribution, with an optimized number of classes. The method #strategyBlock: allows to change the strategy for the distribution.
Pierre it would be good that distribution objects - domain) are not bound to Roassal but should be packaged in SciSmalltalk.
Roassal should be the visualisation not the math and the domain.

I also worked on the RTHistogramSet and method #histogram of SequenceableCollection to improve a bit the looking. There  is still here some work to do  but at least the histogram works well and looks fine.

We worked with Alejandro on the algorithm of frequency distribution to make it efficient compared to the classic one (Only one iteration on datas during the algo itself).

Here are some examples, don't hesitate to try a bit and give me feedbacks if any bugs, so I can move on the next graphs ;)

#(5 3 8 6 5 4 2 9 1 2) histogram.

#(1 2 3) histogram.

#(555 1 1 2) histogram.

((1 to: 1000) collect: [:i | i atRandom ]) histogram.

Also, I found this page and thought it was really nice animations !! http://bost.ocks.org/mike/algorithms/



Moose-dev mailing list