method missing wedge percentile

[EdgarGF93]

A simple alternative to optimize the missing wedge mask by using the percentile on result.count array

[EdgarGF93]

import time
import numpy
import fabio
import matplotlib.pyplot as plt
from matplotlib.pyplot import subplots
from pyFAI.test.utilstest import UtilsTest
from pyFAI.gui.jupyter import plot1d, plot2d
from pyFAI import load, version as pyFAI_version
from pyFAI.integrator.fiber import get_missing_wedge_mask_by_percentile


y6 = fabio.open(UtilsTest.getimage("Y6.edf")).data
air = fabio.open(UtilsTest.getimage("air.edf")).data
data = y6 - 0.03 * air
def update_style(ax):
    for img in ax.get_images():
        img.set_cmap("viridis")
        img.set_clim(10,500)
t0 = time.perf_counter()
print("pyFAI version:", pyFAI_version) 


poni_filename = UtilsTest.getimage("LaB6_3.poni")
fi = load(filename=poni_filename, type_="pyFAI.integrator.fiber.FiberIntegrator")
params = {
    "sample_orientation" : 6,
    "npt_ip" : 1000,
    "ip_range" : [-5,5],
    "npt_oop" : 1000,
    "oop_range" : [5,7],
    "vertical_integration" : False,
}

res2d_default = fi.integrate2d_grazing_incidence(data=data, method=("full", "csr", "cython"), **params)


fig, axes = subplots(ncols=10, nrows=2, figsize=(15,5))
percentiles = numpy.linspace(0.05, 0.1, 20)
for ax, p in zip(axes.ravel(), percentiles):
    ax.imshow(get_missing_wedge_mask_by_percentile(result=res2d_default, percentile=p), origin="lower")
    ax.set_title(f"{p:.3f}")

plt.show()

[EdgarGF93]

if percentile=0.058 looks the best one for this map:

res1d_no = fi.integrate1d_grazing_incidence(data=data, method=("no", "csr", "cython"),  **params

                                               )

percentile = 0.058

res2d_full = fi.integrate2d_grazing_incidence(data=data, method=("full", "csr", "cython"), 
                                               use_missing_wedge=True,
                                               missing_wedge_percentile=percentile,
                                               **params
                                               )



res1d_full = fi.integrate1d_grazing_incidence(data=data, method=("full", "csr", "cython"), 
                                               use_missing_wedge=True,
                                               missing_wedge_percentile=percentile,
                                               **params
                                               )



fig, axes = subplots(ncols=3, figsize=(10,5))

ax=axes[0]
plot2d(res2d_full, ax=ax)
update_style(ax)


ax=axes[1]
plot1d(res1d_no, ax=ax, label="no splitting")
update_style(ax)

ax=axes[2]
plot1d(res1d_full, ax=ax, label="full")
update_style(ax)


plt.show()

We can now oversample the integration:

[EdgarGF93]

An example with the whole map:
We estimate that the percentage of bins falling into the empty spaces is 20-30%, and then a simple loop gives a good number

fig, axes = subplots(ncols=10, nrows=4, figsize=(15,15))
percentiles = numpy.linspace(0.2, 0.3, 40)
for ax, p in zip(axes.ravel(), percentiles):
    ax.imshow(get_missing_wedge_mask_by_percentile(result=res2d_default, percentile=p), origin="lower")
    ax.set_title(f"{p:.3f}")
plt.show()

[kif]

LGTM, would have been nice to have a unit-test.