NormExpan has incorrect shapes with breaks with recent scipy.
This is a slightly different solution, than #1397 by @piannucci
(I guess it avoids args introspection in scipy)
plus some basic unittests for NormExpan.
I didn't find a Gram-Charlier expansion distribution in R seek. (either it doesn't exist or it has a different name)
plus small cleanup: move examples out of module
test coverage for the expansion is still low
I tried a few more things to check whether the results are roughly correct, limit on RMSE, check for fat tails if kurtosis > 0, check for asymmetry if skew > 0, but haven't added them. They would be very coarse checks that qualitatively the results look right.
related: Evgeni has Edgeworth Expansion in PR #1325 with more complete, explicit implementation of other distribution methods.
python 3 has an unrelated test failure
in families test_link
Arrays are not almost equal to 8 decimals
sounds strange (that would require 18 significant digits) ?
ready to merge
There is still a lot to clean up and to increase test coverage, but I want to work on other things.
TST: NormExpan sandbox.distributions: add basic unittests
BUG: NormExpan sandbox.distributions, no shape parameter see #1397
REF/DOC: sandbox.distribution.extras move examples to script file
DOC/REF: make example script runnable (fix imports)
rebased and force pushed
tested with scipy 0.13.3, no failures or errors