* bug fix: force todense of sparse matrices in check_equal, as two sparse
matrices are never equal in scipy 0.6
* numeric_grad no longer cares that all parameters are of the same type
* new Reshape optimization (@todo: reshape-flatten optimization)