[Numpy-discussion] Efficient orthogonalisation with scipy/numpy
Tue Jan 19 16:02:20 CST 2010
2010/1/19 Gael Varoquaux <email@example.com>:
> For the google-completness of this thread, to get a speed gain, one needs
> to use the 'econ=True' flag to qr.
Be warned that in some installations (in particular some using ATLAS),
supplying econ=True can cause a segfault under certain conditions (I
think only when the arrays are misaligned, e.g. coming from
unpickling, and even then only with certain data); the bugfix for this
may or may not be complete, and involves copying misaligned arrays.
More information about the NumPy-Discussion