But, I'm going to try again. Some of the highlights for what I've been working on (and to try out to link to the Picasa plots).

**Density estimation with orthonormal polynomials**

for example with Chebychev, but there are also a few others (Hermite, Fourier, Legendre, ...).

I haven't gotten around to doing multivariate versions yet.

**Removing a trend with orthogonal polynomials**

**Diagnostic Regression Plot**

to see whether there is something "wrong" with the estimated model.

I mostly just followed the description in NIST's dataplot, for example ccpr. NIST also has some background information, for example How can I tell if a model fits my data?

This was written before Chris Jordan Squire contributed lowess for statsmodel which would be nice to add to these graphs.

**Fun with correlation: Visual illustration of correlation of stock returns**

this one uses some robust correlation estimators from scikit-learn (sklearn) (pca is mine)

Because of the density of the plots I suppressed the actual firm labels.

**Some scatterplots with normal distribution ellipses**

and you can find various other plots in my gallery at picasaweb mostly for my statsmodels related work, some multivariate distributions, some splines, some mixed effects models, partial cleanup of generalized additive models and some (pseudo-)Bayesian linear models with shrinkage to an informative prior (which has a classical econometrics interpretation similar to Ridge Regression).

I hope I will get around to actually writing about some of the details for this and for the code that doesn't have nice plots in the gallery.

Josef