Identify outliers

[karthik]

So read all numeric columns and return rows that seem like they contain outliers.
We can use the row index to make a whisker page (@sckott could help here).
@hilaryparker Want to take this one?

[sckott]

@hilaryparker assign yourself if you want to do, otherwise i can take a whack at it

[emhart]

@hilaryparker in issue #1 we talked about visualizing this. I wrote a little gist to tackle it using an iqr cutoff. May or may not be useful. https://gist.github.com/emhart/9025719

[davharris]

Possibly relevant CRAN packages:

• http://cran.r-project.org/web/packages/outliers/index.html
• http://cran.r-project.org/web/packages/mvoutlier/index.html

The first one does simple one-dimensional stuff, possibly assuming everything is Gaussian. Probably not useful. The second one looks much more interesting. It looks at the multivariate distribution and claims to use more robust methods.

I've never used either package or checked them for correctness.

[hilaryparker]

Just to close the loop on this, the function for identifying outliers right now finds outliers outside +/-1.96_SD for now. It's also flexible so that you can change the outlier detection method to 1.5_IQR, etc. if you so choose.