BF: Normalization of GQI2 gqi_vector.

[arokem]

A long overdue fix, , with thanks to @qytian for fixing this and reminding me to
put in this PR on his behalf.

Follow-up of this conversation: http://mail.scipy.org/pipermail/nipy-devel/2013-November/009574.html

[Garyfallidis]

Hey thanks @qytian and @arokem for looking at this. Can you show as the improved ODFs in a before after figure for a 60 degrees crossing for example?

[arokem]

On master:

One this branch:

[arokem]

http://nbviewer.ipython.org/gist/arokem/f92967b718ead8fa494b

[Garyfallidis]

Can you generate the same 2 figures but with sampling_length = 3?

[arokem]

[Garyfallidis]

Aha!

[Garyfallidis]

Okay thank you. I asked for this so we can have a reference on the change of sampling_length after the new pi correction.

Thank you both :)

[qytian]

The abstract "Deconvolution enhanced Generalized Q-Sampling 2 and DSI deconvolution" recommends an integration length lamda between 2 and 3. I am not sure whether here includes the "pi" constant, i.e. should be [2, 3] / pi.

[Garyfallidis]

Yes, because it was using the old code (without your fix). @arokem and @qytian would you mind adding this information to the docstring of the function? I mean say that range of the sampling length has changed? We need also to write this in the api_changes.rst file. Thank you again.

[arokem]

Yeah - no problem. @qytian and I will work on another PR for that.

On Thu, Feb 26, 2015 at 10:43 AM, Eleftherios Garyfallidis <
notifications@github.com> wrote:

Yes, because it was using the old code (without your fix). @arokem
https://github.com/arokem and @qytian https://github.com/qytian would
you mind adding this information to the docstring of the function? I mean
say that range of the sampling length has changed? We need also to write
this in the api_changes.rst file. Thank you again.

—
Reply to this email directly or view it on GitHub
#581 (comment).

[samuelstjean]

Regarding the second screenshot, is that supposed to happen? You seem to have reversed the orientation of the crossings to 90 degrees on that stick figure, while it was more like 60 degrees before.

[Garyfallidis]

Yeah, it's correct. It is the same as sampling from outside the PDF grid in DSI. Your values will not be correct or very small (in the case of GQI2) and then you may start seeing effects of the sampling sphere in the results.