Source code for sas.dataloader.manipulations

"""
Data manipulations for 2D data sets.
Using the meta data information, various types of averaging
are performed in Q-space
"""
#####################################################################
#This software was developed by the University of Tennessee as part of the
#Distributed Data Analysis of Neutron Scattering Experiments (DANSE)
#project funded by the US National Science Foundation.
#See the license text in license.txt
#copyright 2008, University of Tennessee
######################################################################

#TODO: copy the meta data from the 2D object to the resulting 1D object
import math
import numpy

#from data_info import plottable_2D
from data_info import Data1D


[docs]def get_q(dx, dy, det_dist, wavelength): """ :param dx: x-distance from beam center [mm] :param dy: y-distance from beam center [mm] :return: q-value at the given position """ # Distance from beam center in the plane of detector plane_dist = math.sqrt(dx*dx + dy*dy) # Half of the scattering angle theta = 0.5 * math.atan(plane_dist/det_dist) return (4.0 * math.pi/wavelength) * math.sin(theta)
[docs]def get_q_compo(dx, dy, det_dist, wavelength, compo=None): """ This reduces tiny error at very large q. Implementation of this func is not started yet.<--ToDo """ if dy == 0: if dx >= 0: angle_xy = 0 else: angle_xy = math.pi else: angle_xy = math.atan(dx/dy) if compo == "x": out = get_q(dx, dy, det_dist, wavelength) * math.cos(angle_xy) elif compo == "y": out = get_q(dx, dy, det_dist, wavelength) * math.sin(angle_xy) else: out = get_q(dx, dy, det_dist, wavelength) return out
[docs]def flip_phi(phi): """ Correct phi to within the 0 <= to <= 2pi range :return: phi in >=0 and <=2Pi """ Pi = math.pi if phi < 0: phi_out = phi + (2 * Pi) elif phi > (2 * Pi): phi_out = phi - (2 * Pi) else: phi_out = phi return phi_out
[docs]def reader2D_converter(data2d=None): """ convert old 2d format opened by IhorReader or danse_reader to new Data2D format :param data2d: 2d array of Data2D object :return: 1d arrays of Data2D object """ if data2d.data == None or data2d.x_bins == None or data2d.y_bins == None: raise ValueError, "Can't convert this data: data=None..." from sas.dataloader.data_info import Data2D new_x = numpy.tile(data2d.x_bins, (len(data2d.y_bins), 1)) new_y = numpy.tile(data2d.y_bins, (len(data2d.x_bins), 1)) new_y = new_y.swapaxes(0, 1) new_data = data2d.data.flatten() qx_data = new_x.flatten() qy_data = new_y.flatten() q_data = numpy.sqrt(qx_data*qx_data + qy_data*qy_data) if data2d.err_data == None or numpy.any(data2d.err_data <= 0): new_err_data = numpy.sqrt(numpy.abs(new_data)) else: new_err_data = data2d.err_data.flatten() mask = numpy.ones(len(new_data), dtype=bool) #TODO: make sense of the following two lines... output = Data2D() output = data2d output.data = new_data output.err_data = new_err_data output.qx_data = qx_data output.qy_data = qy_data output.q_data = q_data output.mask = mask return output
class _Slab(object): """ Compute average I(Q) for a region of interest """ def __init__(self, x_min=0.0, x_max=0.0, y_min=0.0, y_max=0.0, bin_width=0.001): # Minimum Qx value [A-1] self.x_min = x_min # Maximum Qx value [A-1] self.x_max = x_max # Minimum Qy value [A-1] self.y_min = y_min # Maximum Qy value [A-1] self.y_max = y_max # Bin width (step size) [A-1] self.bin_width = bin_width # If True, I(|Q|) will be return, otherwise, # negative q-values are allowed self.fold = False def __call__(self, data2D): return NotImplemented def _avg(self, data2D, maj): """ Compute average I(Q_maj) for a region of interest. The major axis is defined as the axis of Q_maj. The minor axis is the axis that we average over. :param data2D: Data2D object :param maj_min: min value on the major axis :return: Data1D object """ if len(data2D.detector) != 1: msg = "_Slab._avg: invalid number of " msg += " detectors: %g" % len(data2D.detector) raise RuntimeError, msg # Get data data = data2D.data[numpy.isfinite(data2D.data)] q_data = data2D.q_data[numpy.isfinite(data2D.data)] err_data = data2D.err_data[numpy.isfinite(data2D.data)] qx_data = data2D.qx_data[numpy.isfinite(data2D.data)] qy_data = data2D.qy_data[numpy.isfinite(data2D.data)] # Build array of Q intervals if maj == 'x': if self.fold: x_min = 0 else: x_min = self.x_min nbins = int(math.ceil((self.x_max - x_min) / self.bin_width)) qbins = self.bin_width * numpy.arange(nbins) + x_min elif maj == 'y': if self.fold: y_min = 0 else: y_min = self.y_min nbins = int(math.ceil((self.y_max - y_min)/self.bin_width)) qbins = self.bin_width * numpy.arange(nbins) + y_min else: raise RuntimeError, "_Slab._avg: unrecognized axis %s" % str(maj) x = numpy.zeros(nbins) y = numpy.zeros(nbins) err_y = numpy.zeros(nbins) y_counts = numpy.zeros(nbins) # Average pixelsize in q space for npts in range(len(data)): # default frac frac_x = 0 frac_y = 0 # get ROI if self.x_min <= qx_data[npts] and self.x_max > qx_data[npts]: frac_x = 1 if self.y_min <= qy_data[npts] and self.y_max > qy_data[npts]: frac_y = 1 frac = frac_x * frac_y if frac == 0: continue # binning: find axis of q if maj == 'x': q_value = qx_data[npts] min = x_min if maj == 'y': q_value = qy_data[npts] min = y_min if self.fold and q_value < 0: q_value = -q_value # bin i_q = int(math.ceil((q_value - min) / self.bin_width)) - 1 # skip outside of max bins if i_q < 0 or i_q >= nbins: continue #TODO: find better definition of x[i_q] based on q_data x[i_q] += frac * q_value # min + (i_q + 1) * self.bin_width / 2.0 y[i_q] += frac * data[npts] if err_data == None or err_data[npts] == 0.0: if data[npts] < 0: data[npts] = -data[npts] err_y[i_q] += frac * frac * data[npts] else: err_y[i_q] += frac * frac * err_data[npts] * err_data[npts] y_counts[i_q] += frac # Average the sums for n in range(nbins): err_y[n] = math.sqrt(err_y[n]) err_y = err_y / y_counts y = y / y_counts x = x / y_counts idx = (numpy.isfinite(y) & numpy.isfinite(x)) if not idx.any(): msg = "Average Error: No points inside ROI to average..." raise ValueError, msg #elif len(y[idx])!= nbins: # msg = "empty bin(s) due to tight binning..." # print "resulted",nbins- len(y[idx]), msg return Data1D(x=x[idx], y=y[idx], dy=err_y[idx])
[docs]class SlabY(_Slab): """ Compute average I(Qy) for a region of interest """ def __call__(self, data2D): """ Compute average I(Qy) for a region of interest :param data2D: Data2D object :return: Data1D object """ return self._avg(data2D, 'y')
[docs]class SlabX(_Slab): """ Compute average I(Qx) for a region of interest """ def __call__(self, data2D): """ Compute average I(Qx) for a region of interest :param data2D: Data2D object :return: Data1D object """ return self._avg(data2D, 'x')
[docs]class Boxsum(object): """ Perform the sum of counts in a 2D region of interest. """ def __init__(self, x_min=0.0, x_max=0.0, y_min=0.0, y_max=0.0): # Minimum Qx value [A-1] self.x_min = x_min # Maximum Qx value [A-1] self.x_max = x_max # Minimum Qy value [A-1] self.y_min = y_min # Maximum Qy value [A-1] self.y_max = y_max def __call__(self, data2D): """ Perform the sum in the region of interest :param data2D: Data2D object :return: number of counts, error on number of counts, number of points summed """ y, err_y, y_counts = self._sum(data2D) # Average the sums counts = 0 if y_counts == 0 else y error = 0 if y_counts == 0 else math.sqrt(err_y) # Added y_counts to return, SMK & PDB, 04/03/2013 return counts, error, y_counts def _sum(self, data2D): """ Perform the sum in the region of interest :param data2D: Data2D object :return: number of counts, error on number of counts, number of entries summed """ if len(data2D.detector) != 1: msg = "Circular averaging: invalid number " msg += "of detectors: %g" % len(data2D.detector) raise RuntimeError, msg # Get data data = data2D.data[numpy.isfinite(data2D.data)] q_data = data2D.q_data[numpy.isfinite(data2D.data)] err_data = data2D.err_data[numpy.isfinite(data2D.data)] qx_data = data2D.qx_data[numpy.isfinite(data2D.data)] qy_data = data2D.qy_data[numpy.isfinite(data2D.data)] y = 0.0 err_y = 0.0 y_counts = 0.0 # Average pixelsize in q space for npts in range(len(data)): # default frac frac_x = 0 frac_y = 0 # get min and max at each points qx = qx_data[npts] qy = qy_data[npts] # get the ROI if self.x_min <= qx and self.x_max > qx: frac_x = 1 if self.y_min <= qy and self.y_max > qy: frac_y = 1 #Find the fraction along each directions frac = frac_x * frac_y if frac == 0: continue y += frac * data[npts] if err_data == None or err_data[npts] == 0.0: if data[npts] < 0: data[npts] = -data[npts] err_y += frac * frac * data[npts] else: err_y += frac * frac * err_data[npts] * err_data[npts] y_counts += frac return y, err_y, y_counts
[docs]class Boxavg(Boxsum): """ Perform the average of counts in a 2D region of interest. """ def __init__(self, x_min=0.0, x_max=0.0, y_min=0.0, y_max=0.0): super(Boxavg, self).__init__(x_min=x_min, x_max=x_max, y_min=y_min, y_max=y_max) def __call__(self, data2D): """ Perform the sum in the region of interest :param data2D: Data2D object :return: average counts, error on average counts """ y, err_y, y_counts = self._sum(data2D) # Average the sums counts = 0 if y_counts == 0 else y / y_counts error = 0 if y_counts == 0 else math.sqrt(err_y) / y_counts return counts, error
[docs]def get_pixel_fraction_square(x, xmin, xmax): """ Return the fraction of the length from xmin to x.:: A B +-----------+---------+ xmin x xmax :param x: x-value :param xmin: minimum x for the length considered :param xmax: minimum x for the length considered :return: (x-xmin)/(xmax-xmin) when xmin < x < xmax """ if x <= xmin: return 0.0 if x > xmin and x < xmax: return (x - xmin) / (xmax - xmin) else: return 1.0
[docs]class CircularAverage(object): """ Perform circular averaging on 2D data The data returned is the distribution of counts as a function of Q """ def __init__(self, r_min=0.0, r_max=0.0, bin_width=0.0005): # Minimum radius included in the average [A-1] self.r_min = r_min # Maximum radius included in the average [A-1] self.r_max = r_max # Bin width (step size) [A-1] self.bin_width = bin_width def __call__(self, data2D, ismask=False): """ Perform circular averaging on the data :param data2D: Data2D object :return: Data1D object """ # Get data W/ finite values data = data2D.data[numpy.isfinite(data2D.data)] q_data = data2D.q_data[numpy.isfinite(data2D.data)] qx_data = data2D.qx_data[numpy.isfinite(data2D.data)] err_data = data2D.err_data[numpy.isfinite(data2D.data)] mask_data = data2D.mask[numpy.isfinite(data2D.data)] dq_data = None # Get the dq for resolution averaging if data2D.dqx_data != None and data2D.dqy_data != None: # The pinholes and det. pix contribution present # in both direction of the 2D which must be subtracted when # converting to 1D: dq_overlap should calculated ideally at # q = 0. Note This method works on only pinhole geometry. # Extrapolate dqx(r) and dqy(phi) at q = 0, and take an average. z_max = max(data2D.q_data) z_min = min(data2D.q_data) x_max = data2D.dqx_data[data2D.q_data[z_max]] x_min = data2D.dqx_data[data2D.q_data[z_min]] y_max = data2D.dqy_data[data2D.q_data[z_max]] y_min = data2D.dqy_data[data2D.q_data[z_min]] # Find qdx at q = 0 dq_overlap_x = (x_min * z_max - x_max * z_min) / (z_max - z_min) # when extrapolation goes wrong if dq_overlap_x > min(data2D.dqx_data): dq_overlap_x = min(data2D.dqx_data) dq_overlap_x *= dq_overlap_x # Find qdx at q = 0 dq_overlap_y = (y_min * z_max - y_max * z_min) / (z_max - z_min) # when extrapolation goes wrong if dq_overlap_y > min(data2D.dqy_data): dq_overlap_y = min(data2D.dqy_data) # get dq at q=0. dq_overlap_y *= dq_overlap_y dq_overlap = numpy.sqrt((dq_overlap_x + dq_overlap_y) / 2.0) # Final protection of dq if dq_overlap < 0: dq_overlap = y_min dqx_data = data2D.dqx_data[numpy.isfinite(data2D.data)] dqy_data = data2D.dqy_data[numpy.isfinite(data2D.data)] - dq_overlap # def; dqx_data = dq_r dqy_data = dq_phi # Convert dq 2D to 1D here dqx = dqx_data * dqx_data dqy = dqy_data * dqy_data dq_data = numpy.add(dqx, dqy) dq_data = numpy.sqrt(dq_data) #q_data_max = numpy.max(q_data) if len(data2D.q_data) == None: msg = "Circular averaging: invalid q_data: %g" % data2D.q_data raise RuntimeError, msg # Build array of Q intervals nbins = int(math.ceil((self.r_max - self.r_min) / self.bin_width)) qbins = self.bin_width * numpy.arange(nbins) + self.r_min x = numpy.zeros(nbins) y = numpy.zeros(nbins) err_y = numpy.zeros(nbins) err_x = numpy.zeros(nbins) y_counts = numpy.zeros(nbins) for npt in range(len(data)): if ismask and not mask_data[npt]: continue frac = 0 # q-value at the pixel (j,i) q_value = q_data[npt] data_n = data[npt] ## No need to calculate the frac when all data are within range if self.r_min >= self.r_max: raise ValueError, "Limit Error: min > max" if self.r_min <= q_value and q_value <= self.r_max: frac = 1 if frac == 0: continue i_q = int(math.floor((q_value - self.r_min) / self.bin_width)) # Take care of the edge case at phi = 2pi. if i_q == nbins: i_q = nbins - 1 y[i_q] += frac * data_n # Take dqs from data to get the q_average x[i_q] += frac * q_value if err_data == None or err_data[npt] == 0.0: if data_n < 0: data_n = -data_n err_y[i_q] += frac * frac * data_n else: err_y[i_q] += frac * frac * err_data[npt] * err_data[npt] if dq_data != None: # To be consistent with dq calculation in 1d reduction, # we need just the averages (not quadratures) because # it should not depend on the number of the q points # in the qr bins. err_x[i_q] += frac * dq_data[npt] else: err_x = None y_counts[i_q] += frac # Average the sums for n in range(nbins): if err_y[n] < 0: err_y[n] = -err_y[n] err_y[n] = math.sqrt(err_y[n]) #if err_x != None: # err_x[n] = math.sqrt(err_x[n]) err_y = err_y / y_counts err_y[err_y == 0] = numpy.average(err_y) y = y / y_counts x = x / y_counts idx = (numpy.isfinite(y)) & (numpy.isfinite(x)) if err_x != None: d_x = err_x[idx] / y_counts[idx] else: d_x = None if not idx.any(): msg = "Average Error: No points inside ROI to average..." raise ValueError, msg return Data1D(x=x[idx], y=y[idx], dy=err_y[idx], dx=d_x)
[docs]class Ring(object): """ Defines a ring on a 2D data set. The ring is defined by r_min, r_max, and the position of the center of the ring. The data returned is the distribution of counts around the ring as a function of phi. Phi_min and phi_max should be defined between 0 and 2*pi in anti-clockwise starting from the x- axis on the left-hand side """ #Todo: remove center. def __init__(self, r_min=0, r_max=0, center_x=0, center_y=0, nbins=20): # Minimum radius self.r_min = r_min # Maximum radius self.r_max = r_max # Center of the ring in x self.center_x = center_x # Center of the ring in y self.center_y = center_y # Number of angular bins self.nbins_phi = nbins def __call__(self, data2D): """ Apply the ring to the data set. Returns the angular distribution for a given q range :param data2D: Data2D object :return: Data1D object """ if data2D.__class__.__name__ not in ["Data2D", "plottable_2D"]: raise RuntimeError, "Ring averaging only take plottable_2D objects" Pi = math.pi # Get data data = data2D.data[numpy.isfinite(data2D.data)] q_data = data2D.q_data[numpy.isfinite(data2D.data)] err_data = data2D.err_data[numpy.isfinite(data2D.data)] qx_data = data2D.qx_data[numpy.isfinite(data2D.data)] qy_data = data2D.qy_data[numpy.isfinite(data2D.data)] q_data_max = numpy.max(q_data) # Set space for 1d outputs phi_bins = numpy.zeros(self.nbins_phi) phi_counts = numpy.zeros(self.nbins_phi) phi_values = numpy.zeros(self.nbins_phi) phi_err = numpy.zeros(self.nbins_phi) for npt in range(len(data)): frac = 0 # q-value at the point (npt) q_value = q_data[npt] data_n = data[npt] # phi-value at the point (npt) phi_value = math.atan2(qy_data[npt], qx_data[npt]) + Pi if self.r_min <= q_value and q_value <= self.r_max: frac = 1 if frac == 0: continue # binning i_phi = int(math.floor((self.nbins_phi) * phi_value / (2 * Pi))) # Take care of the edge case at phi = 2pi. if i_phi == self.nbins_phi: i_phi = self.nbins_phi - 1 phi_bins[i_phi] += frac * data[npt] if err_data == None or err_data[npt] == 0.0: if data_n < 0: data_n = -data_n phi_err[i_phi] += frac * frac * math.fabs(data_n) else: phi_err[i_phi] += frac * frac * err_data[npt] * err_data[npt] phi_counts[i_phi] += frac for i in range(self.nbins_phi): phi_bins[i] = phi_bins[i] / phi_counts[i] phi_err[i] = math.sqrt(phi_err[i]) / phi_counts[i] phi_values[i] = 2.0 * math.pi / self.nbins_phi * (1.0 * i + 0.5) idx = (numpy.isfinite(phi_bins)) if not idx.any(): msg = "Average Error: No points inside ROI to average..." raise ValueError, msg #elif len(phi_bins[idx])!= self.nbins_phi: # print "resulted",self.nbins_phi- len(phi_bins[idx]) #,"empty bin(s) due to tight binning..." return Data1D(x=phi_values[idx], y=phi_bins[idx], dy=phi_err[idx])
[docs]def get_pixel_fraction(qmax, q_00, q_01, q_10, q_11): """ Returns the fraction of the pixel defined by the four corners (q_00, q_01, q_10, q_11) that has q < qmax.:: q_01 q_11 y=1 +--------------+ | | | | | | y=0 +--------------+ q_00 q_10 x=0 x=1 """ # y side for x = minx x_0 = get_intercept(qmax, q_00, q_01) # y side for x = maxx x_1 = get_intercept(qmax, q_10, q_11) # x side for y = miny y_0 = get_intercept(qmax, q_00, q_10) # x side for y = maxy y_1 = get_intercept(qmax, q_01, q_11) # surface fraction for a 1x1 pixel frac_max = 0 if x_0 and x_1: frac_max = (x_0 + x_1) / 2.0 elif y_0 and y_1: frac_max = (y_0 + y_1) / 2.0 elif x_0 and y_0: if q_00 < q_10: frac_max = x_0 * y_0 / 2.0 else: frac_max = 1.0 - x_0 * y_0 / 2.0 elif x_0 and y_1: if q_00 < q_10: frac_max = x_0 * y_1 / 2.0 else: frac_max = 1.0 - x_0 * y_1 / 2.0 elif x_1 and y_0: if q_00 > q_10: frac_max = x_1 * y_0 / 2.0 else: frac_max = 1.0 - x_1 * y_0 / 2.0 elif x_1 and y_1: if q_00 < q_10: frac_max = 1.0 - (1.0 - x_1) * (1.0 - y_1) / 2.0 else: frac_max = (1.0 - x_1) * (1.0 - y_1) / 2.0 # If we make it here, there is no intercept between # this pixel and the constant-q ring. We only need # to know if we have to include it or exclude it. elif (q_00 + q_01 + q_10 + q_11) / 4.0 < qmax: frac_max = 1.0 return frac_max
[docs]def get_intercept(q, q_0, q_1): """ Returns the fraction of the side at which the q-value intercept the pixel, None otherwise. The values returned is the fraction ON THE SIDE OF THE LOWEST Q. :: A B +-----------+--------+ <--- pixel size 0 1 Q_0 -------- Q ----- Q_1 <--- equivalent Q range if Q_1 > Q_0, A is returned if Q_1 < Q_0, B is returned if Q is outside the range of [Q_0, Q_1], None is returned """ if q_1 > q_0: if (q > q_0 and q <= q_1): return (q - q_0) / (q_1 - q_0) else: if (q > q_1 and q <= q_0): return (q - q_1) / (q_0 - q_1) return None
class _Sector: """ Defines a sector region on a 2D data set. The sector is defined by r_min, r_max, phi_min, phi_max, and the position of the center of the ring where phi_min and phi_max are defined by the right and left lines wrt central line and phi_max could be less than phi_min. Phi is defined between 0 and 2*pi in anti-clockwise starting from the x- axis on the left-hand side """ def __init__(self, r_min, r_max, phi_min=0, phi_max=2*math.pi, nbins=20): self.r_min = r_min self.r_max = r_max self.phi_min = phi_min self.phi_max = phi_max self.nbins = nbins def _agv(self, data2D, run='phi'): """ Perform sector averaging. :param data2D: Data2D object :param run: define the varying parameter ('phi' , 'q' , or 'q2') :return: Data1D object """ if data2D.__class__.__name__ not in ["Data2D", "plottable_2D"]: raise RuntimeError, "Ring averaging only take plottable_2D objects" Pi = math.pi # Get the all data & info data = data2D.data[numpy.isfinite(data2D.data)] q_data = data2D.q_data[numpy.isfinite(data2D.data)] err_data = data2D.err_data[numpy.isfinite(data2D.data)] qx_data = data2D.qx_data[numpy.isfinite(data2D.data)] qy_data = data2D.qy_data[numpy.isfinite(data2D.data)] dq_data = None # Get the dq for resolution averaging if data2D.dqx_data != None and data2D.dqy_data != None: # The pinholes and det. pix contribution present # in both direction of the 2D which must be subtracted when # converting to 1D: dq_overlap should calculated ideally at # q = 0. # Extrapolate dqy(perp) at q = 0 z_max = max(data2D.q_data) z_min = min(data2D.q_data) x_max = data2D.dqx_data[data2D.q_data[z_max]] x_min = data2D.dqx_data[data2D.q_data[z_min]] y_max = data2D.dqy_data[data2D.q_data[z_max]] y_min = data2D.dqy_data[data2D.q_data[z_min]] # Find qdx at q = 0 dq_overlap_x = (x_min * z_max - x_max * z_min) / (z_max - z_min) # when extrapolation goes wrong if dq_overlap_x > min(data2D.dqx_data): dq_overlap_x = min(data2D.dqx_data) dq_overlap_x *= dq_overlap_x # Find qdx at q = 0 dq_overlap_y = (y_min * z_max - y_max * z_min) / (z_max - z_min) # when extrapolation goes wrong if dq_overlap_y > min(data2D.dqy_data): dq_overlap_y = min(data2D.dqy_data) # get dq at q=0. dq_overlap_y *= dq_overlap_y dq_overlap = numpy.sqrt((dq_overlap_x + dq_overlap_y) / 2.0) if dq_overlap < 0: dq_overlap = y_min dqx_data = data2D.dqx_data[numpy.isfinite(data2D.data)] dqy_data = data2D.dqy_data[numpy.isfinite(data2D.data)] - dq_overlap # def; dqx_data = dq_r dqy_data = dq_phi # Convert dq 2D to 1D here dqx = dqx_data * dqx_data dqy = dqy_data * dqy_data dq_data = numpy.add(dqx, dqy) dq_data = numpy.sqrt(dq_data) #set space for 1d outputs x = numpy.zeros(self.nbins) y = numpy.zeros(self.nbins) y_err = numpy.zeros(self.nbins) x_err = numpy.zeros(self.nbins) y_counts = numpy.zeros(self.nbins) # Get the min and max into the region: 0 <= phi < 2Pi phi_min = flip_phi(self.phi_min) phi_max = flip_phi(self.phi_max) q_data_max = numpy.max(q_data) for n in range(len(data)): frac = 0 # q-value at the pixel (j,i) q_value = q_data[n] data_n = data[n] # Is pixel within range? is_in = False # phi-value of the pixel (j,i) phi_value = math.atan2(qy_data[n], qx_data[n]) + Pi ## No need to calculate the frac when all data are within range if self.r_min <= q_value and q_value <= self.r_max: frac = 1 if frac == 0: continue #In case of two ROIs (symmetric major and minor regions)(for 'q2') if run.lower() == 'q2': ## For minor sector wing # Calculate the minor wing phis phi_min_minor = flip_phi(phi_min - Pi) phi_max_minor = flip_phi(phi_max - Pi) # Check if phis of the minor ring is within 0 to 2pi if phi_min_minor > phi_max_minor: is_in = (phi_value > phi_min_minor or \ phi_value < phi_max_minor) else: is_in = (phi_value > phi_min_minor and \ phi_value < phi_max_minor) #For all cases(i.e.,for 'q', 'q2', and 'phi') #Find pixels within ROI if phi_min > phi_max: is_in = is_in or (phi_value > phi_min or \ phi_value < phi_max) else: is_in = is_in or (phi_value >= phi_min and \ phi_value < phi_max) if not is_in: frac = 0 if frac == 0: continue # Check which type of averaging we need if run.lower() == 'phi': temp_x = (self.nbins) * (phi_value - self.phi_min) temp_y = (self.phi_max - self.phi_min) i_bin = int(math.floor(temp_x / temp_y)) else: temp_x = (self.nbins) * (q_value - self.r_min) temp_y = (self.r_max - self.r_min) i_bin = int(math.floor(temp_x / temp_y)) # Take care of the edge case at phi = 2pi. if i_bin == self.nbins: i_bin = self.nbins - 1 ## Get the total y y[i_bin] += frac * data_n x[i_bin] += frac * q_value if err_data[n] == None or err_data[n] == 0.0: if data_n < 0: data_n = -data_n y_err[i_bin] += frac * frac * data_n else: y_err[i_bin] += frac * frac * err_data[n] * err_data[n] if dq_data != None: # To be consistent with dq calculation in 1d reduction, # we need just the averages (not quadratures) because # it should not depend on the number of the q points # in the qr bins. x_err[i_bin] += frac * dq_data[n] else: x_err = None y_counts[i_bin] += frac # Organize the results for i in range(self.nbins): y[i] = y[i] / y_counts[i] y_err[i] = math.sqrt(y_err[i]) / y_counts[i] # The type of averaging: phi,q2, or q # Calculate x[i]should be at the center of the bin if run.lower() == 'phi': x[i] = (self.phi_max - self.phi_min) / self.nbins * \ (1.0 * i + 0.5) + self.phi_min else: # We take the center of ring area, not radius. # This is more accurate than taking the radial center of ring. #delta_r = (self.r_max - self.r_min) / self.nbins #r_inner = self.r_min + delta_r * i #r_outer = r_inner + delta_r #x[i] = math.sqrt((r_inner * r_inner + r_outer * r_outer) / 2) x[i] = x[i] / y_counts[i] y_err[y_err == 0] = numpy.average(y_err) idx = (numpy.isfinite(y) & numpy.isfinite(y_err)) if x_err != None: d_x = x_err[idx] / y_counts[idx] else: d_x = None if not idx.any(): msg = "Average Error: No points inside sector of ROI to average..." raise ValueError, msg #elif len(y[idx])!= self.nbins: # print "resulted",self.nbins- len(y[idx]), #"empty bin(s) due to tight binning..." return Data1D(x=x[idx], y=y[idx], dy=y_err[idx], dx=d_x)
[docs]class SectorPhi(_Sector): """ Sector average as a function of phi. I(phi) is return and the data is averaged over Q. A sector is defined by r_min, r_max, phi_min, phi_max. The number of bin in phi also has to be defined. """ def __call__(self, data2D): """ Perform sector average and return I(phi). :param data2D: Data2D object :return: Data1D object """ return self._agv(data2D, 'phi')
[docs]class SectorQ(_Sector): """ Sector average as a function of Q for both symatric wings. I(Q) is return and the data is averaged over phi. A sector is defined by r_min, r_max, phi_min, phi_max. r_min, r_max, phi_min, phi_max >0. The number of bin in Q also has to be defined. """ def __call__(self, data2D): """ Perform sector average and return I(Q). :param data2D: Data2D object :return: Data1D object """ return self._agv(data2D, 'q2')
[docs]class Ringcut(object): """ Defines a ring on a 2D data set. The ring is defined by r_min, r_max, and the position of the center of the ring. The data returned is the region inside the ring Phi_min and phi_max should be defined between 0 and 2*pi in anti-clockwise starting from the x- axis on the left-hand side """ def __init__(self, r_min=0, r_max=0, center_x=0, center_y=0): # Minimum radius self.r_min = r_min # Maximum radius self.r_max = r_max # Center of the ring in x self.center_x = center_x # Center of the ring in y self.center_y = center_y def __call__(self, data2D): """ Apply the ring to the data set. Returns the angular distribution for a given q range :param data2D: Data2D object :return: index array in the range """ if data2D.__class__.__name__ not in ["Data2D", "plottable_2D"]: raise RuntimeError, "Ring cut only take plottable_2D objects" # Get data qx_data = data2D.qx_data qy_data = data2D.qy_data mask = data2D.mask q_data = numpy.sqrt(qx_data * qx_data + qy_data * qy_data) # check whether or not the data point is inside ROI out = (self.r_min <= q_data) & (self.r_max >= q_data) return (out)
[docs]class Boxcut(object): """ Find a rectangular 2D region of interest. """ def __init__(self, x_min=0.0, x_max=0.0, y_min=0.0, y_max=0.0): # Minimum Qx value [A-1] self.x_min = x_min # Maximum Qx value [A-1] self.x_max = x_max # Minimum Qy value [A-1] self.y_min = y_min # Maximum Qy value [A-1] self.y_max = y_max def __call__(self, data2D): """ Find a rectangular 2D region of interest. :param data2D: Data2D object :return: mask, 1d array (len = len(data)) with Trues where the data points are inside ROI, otherwise False """ mask = self._find(data2D) return mask def _find(self, data2D): """ Find a rectangular 2D region of interest. :param data2D: Data2D object :return: out, 1d array (length = len(data)) with Trues where the data points are inside ROI, otherwise Falses """ if data2D.__class__.__name__ not in ["Data2D", "plottable_2D"]: raise RuntimeError, "Boxcut take only plottable_2D objects" # Get qx_ and qy_data qx_data = data2D.qx_data qy_data = data2D.qy_data mask = data2D.mask # check whether or not the data point is inside ROI outx = (self.x_min <= qx_data) & (self.x_max > qx_data) outy = (self.y_min <= qy_data) & (self.y_max > qy_data) return (outx & outy)
[docs]class Sectorcut(object): """ Defines a sector (major + minor) region on a 2D data set. The sector is defined by phi_min, phi_max, where phi_min and phi_max are defined by the right and left lines wrt central line. Phi_min and phi_max are given in units of radian and (phi_max-phi_min) should not be larger than pi """ def __init__(self, phi_min=0, phi_max=math.pi): self.phi_min = phi_min self.phi_max = phi_max def __call__(self, data2D): """ Find a rectangular 2D region of interest. :param data2D: Data2D object :return: mask, 1d array (len = len(data)) with Trues where the data points are inside ROI, otherwise False """ mask = self._find(data2D) return mask def _find(self, data2D): """ Find a rectangular 2D region of interest. :param data2D: Data2D object :return: out, 1d array (length = len(data)) with Trues where the data points are inside ROI, otherwise Falses """ if data2D.__class__.__name__ not in ["Data2D", "plottable_2D"]: raise RuntimeError, "Sectorcut take only plottable_2D objects" Pi = math.pi # Get data qx_data = data2D.qx_data qy_data = data2D.qy_data phi_data = numpy.zeros(len(qx_data)) # get phi from data phi_data = numpy.arctan2(qy_data, qx_data) # Get the min and max into the region: -pi <= phi < Pi phi_min_major = flip_phi(self.phi_min + Pi) - Pi phi_max_major = flip_phi(self.phi_max + Pi) - Pi # check for major sector if phi_min_major > phi_max_major: out_major = (phi_min_major <= phi_data) + (phi_max_major > phi_data) else: out_major = (phi_min_major <= phi_data) & (phi_max_major > phi_data) # minor sector # Get the min and max into the region: -pi <= phi < Pi phi_min_minor = flip_phi(self.phi_min) - Pi phi_max_minor = flip_phi(self.phi_max) - Pi # check for minor sector if phi_min_minor > phi_max_minor: out_minor = (phi_min_minor <= phi_data) + \ (phi_max_minor >= phi_data) else: out_minor = (phi_min_minor <= phi_data) & \ (phi_max_minor >= phi_data) out = out_major + out_minor return out