[Neuroimaging] reg. load nifti image

[Matthew Brett]

Hi,

On Thu, Jan 21, 2016 at 1:11 AM, Raghav Mehta
<raghav.mehta at research.iiit.ac.in> wrote:
> Hi,
>
> Pre-defined thresold is -6.6e-14, while value of w2 for my data is -6.401211e-07. So when I change the thresold is works perfectly?? Can you please explain how this may effect processing of data. as I had to multiply quaternion thresold with 1e9 to get proper thresold value
>

Oh - whoops - I missed off a negative sign, maybe that explains the confusion.

The default threshold is:

In [11]: -np.finfo(np.float32).eps * 3

Out[11]: -3.5762786865234375e-07

Your value appears to be a little below that, at  -6.401211e-07.  So,
with a threshold of:

In [13]: -np.finfo(np.float32).eps * 10

Out[13]: -1.1920928955078125e-06

you should be fairly safe, with a reasonable threshold.

>>> import numpy as np
>>> import nibabel as nib
>>> nib.Nifti1Header.quaternion_threshold = -np.finfo(np.float32).eps * 10  # notice new minus sign
>>> img = nib.load('bad_image.nii')

Does that work?

Matthew