advanced use of azimuth_range parameter for integration1d/2d methode

[gbenecke]

The azimuth_range parameter currently only allows to select a range from small to big angles in the azimuthal range. With this definition some cases of integration ranges are not possible (see image). I suggest to redefine the use of the azimuth_range parameter:

• azimuth_range is a tuple of two floats (chi0, chi1) in degree or radians (currents is radians, doc string says degree)
• the used range for integration starts at chi0 and goes counter clock wise until chi1

Attached is a script implementing this suggested method.

import pyFAI
import pylab
import numpy

def mask(chi, c0, c1):
    chi = numpy.copy(chi)
    if c0 >= c1:
        m = (chi <= c0) & (chi >= c1)
    else:
        m = (chi <= c0) | (chi >= c1)

    chi[~m] = 0
    return chi

ai = pyFAI.AzimuthalIntegrator(poni1=250, poni2=250, pixel1=1, pixel2=1)
chi = numpy.rad2deg(ai.chiArray((500,500)))

fig = pylab.gcf()

pylab.subplot(2,3,1)
pylab.title("c0=-45, c1=-135")
pylab.imshow(mask(chi, -45, -135), vmin=-180, vmax=180)
pylab.colorbar()

pylab.subplot(2,3,2)
pylab.title("c0=-135, c1=-45")
pylab.imshow(mask(chi, -135, -45), vmin=-180, vmax=180)
pylab.colorbar()

pylab.subplot(2,3,3)
pylab.title("c0=-135, c1=135")
pylab.imshow(mask(chi, -135, 135), vmin=-180, vmax=180)
pylab.colorbar()

pylab.subplot(2,3,4)
pylab.title("c0=135, c1=45")
pylab.imshow(mask(chi, 135, 45), vmin=-180, vmax=180)
pylab.colorbar()

pylab.subplot(2,3,5)
pylab.title("c0=45, c1=135")
pylab.imshow(mask(chi, 45, 135), vmin=-180, vmax=180)
pylab.colorbar()

pylab.subplot(2,3,6)
pylab.title("c0=45, c1=-45")
pylab.imshow(mask(chi, 45, -45), vmin=-180, vmax=180)
pylab.colorbar()

pylab.show()

[kif]

Hi Gunthard,

Could you explain a bit more what is the problem ?

• Maybe there is an issue with units deg vs radians. PyFAI tries to use always radians internally but should use the best unit "externally", for the azimuthal angle, degrees is sensible. If you spotted such inconsistency, please tell me.
• PyFAI is using origin="lower" for matplotlib image representation. I agree there is room for improvement here as it changes the sign of chi ! The Ideal situation would be to have the detector defined as a camera (facing the sample) pixels starting from top left and store the orientation (one of the 8) and finally provide a tool based on numpy.flipud/fliplr ... to have everything agzin correctly.
• There are two configuration for azimuthal angle: from -180 -> +180° (default disc@pi) and from 0->360° (disc@0). While those cover most configuration, they cannot integrate +100 -> 450° for example (unless one perform post-processing of the data).
• Keep in mind generating large mask and applying them is quite resources consuming: In [7]: %timeit mask(chi,10,60) 100 loops, best of 3: 3.95 ms per loop Whereas: In [11]: %timeit ai.integrate1d(r,500,unit="r_mm",method="lut") 100 loops, best of 3: 2.45 ms per loop

[gbenecke]

There are two configuration for azimuthal angle: from -180 -> +180° (default disc@pi) and from 0->360° (disc@0). While those cover most configuration, they cannot integrate +100 -> 450° for example (unless one perform post-processing of the data).

Sorry for not being precise enough in the first post, but this is exactly what I meant. Integration over the border of the azimuthal angle (-180/180 or 0/360) is not possible in the current implementation. I noticed that masking big parts is quite expensive, but due to the problem of beam stops, shadows and other distortion it is quite often necessary to only cake integrate. So this is only a suggestion for a future release.

[kif]

On Thu, 24 Oct 2013 04:23:20 -0700 gbenecke notifications@github.com wrote:

There are two configuration for azimuthal angle: from -180 -> +180° (default disc@pi) and from 0->360° (disc@0). While those cover most configuration, they cannot integrate +100 -> 450° for example (unless one perform post-processing of the data).

Sorry for not being precise enough in the first post, but this is exactly what I meant. Integration over the border of the azimuthal angle (-180/180 or 0/360) is not possible in the current implementation. I noticed that masking big parts is quite expensive, but due to the problem of beam stops, shadows and other distortion it is quite often necessary to only cake integrate. So this is only a suggestion for a future release.

Do you see a solution for offering providing this discontinuity at any position ?

Jérôme Kieffer On-Line Data analysis / Software Group ISDD / ESRF tel +33 476 882 445