Join GitHub today
DTI memory: use the same step in prediction as you use in fitting. #857
I can also merge that commit into this branch, if that's OK with you.
On Tue, Feb 2, 2016 at 4:01 PM, Bago Amirbekian email@example.com
This is memory profiling with bdb2030: https://gist.github.com/arokem/36c41c91e8a4aa1b0bd3
changed the title from
WIP: DTI memory: use the same step in prediction as you use in fitting.
DTI memory: use the same step in prediction as you use in fitting.
Feb 3, 2016
It actually works now
In : %memit data_p = tenfit.predict(gtab, S0) /home/samuel/python/dipy/reconst/dti.py:1759: RuntimeWarning: divide by zero encountered in log D[..., 6] = -np.log(b0) peak memory: 6689.07 MiB, increment: 5655.26 MiB
Might want to check that log(b0), just in case it threw off the profiler. A quick guess is that it does log(S0)? At least it does not blow up like before, so that's an improvement.
You know, it ran fairly quickly as opposed to not at all, so I was really
2016-02-04 16:29 GMT+01:00 Ariel Rokem firstname.lastname@example.org:
"fairly quickly as opposed to not at all" should be our motto :-)
If you don't mask these voxels yourself, it will try to predict them for
On Thu, Feb 4, 2016 at 8:25 AM, Samuel St-Jean email@example.com