p = ProgressPlot()
errors = []
for i in range(60):
print(i)
W, Pjlk, timing = M(W, data, Nl=Nl, ml=ml, beta=fudge,
nproc=24,
roll_center=CENTER)
[print(k, '%.3f'%v) for k, v in timing.items()]
W, error = C(W, envelope)
errors.append(error)
p.update(np.log10(W), Pjlk, errors, vmax=1)
# expand the resolution now and then
if i and (Nj<Nj_max) and (i % increase_every) == 0:
W = np.pad(W, ((2,2),(0,0),(0,0)))
Nj = W.shape[0]
Q3 = dq3 * Nj
envelope = generate_envelope(Nj, data.shape[-1], Q=(Q3, Q12, Q12), Dmax=(Dmax, 1, 1), theta=theta)
print('increased Nj to %u and Q3 to %.2e'%(Nj, Q3))
if i and (fudge < fudge_max) and (i % increase_every) == 0:
fudge *= 2**(1/2)
print('increased fudge to %e'%(fudge))
# assuming that we now know the q-range, we can interpolate to qx, qy, qz
# the outermost q3 slices contain everything that couldn't be assigned well.
oldshape = W.shape[0]
Q3 = W.shape[0] / oldshape * Q3
W_ortho, Qnew = rectify_sample(W, (Q3, Q12, Q12), theta)
np.savez('assembled_%u.npz'%subset, W=W, W_ortho=W_ortho, Pjlk=Pjlk, rolls=rolls, Q_ortho=Qnew, Q=(Q3, Q12, Q12))
resolutions = (2, 5, 8, 10, 15, 12, 20, 30, 50)
a2.set_xticks([2 * np.pi / dr for dr in resolutions])
a2.set_xticklabels(resolutions)
a2.set_xlim(a1.get_xlim())
np.savez('validation.npz',
q=x,
fsc=fsc,
nri=nri,
fsc_cut=fsc_cut,
fsc_mix=fsc_mix)
#a1.legend()
# rectify the full reconstruction, save and plot
p, mask = load(folder + 'modes_10.h5', N)
#p = np.flip(p, axis=0)
p, psize = rectify_sample(p, (dr3, dr1, dr2), 15.0, find_order=True) # x z y
np.savez('rectified.npz', data=p, psize=psize)
# plot along all three axes to understand the aspect ratio
fig, ax = plt.subplots(ncols=3)
ext = np.array((-psize / 2 * N, psize / 2 * N, -psize / 2 * N,
psize / 2 * N)) * 1e9 # valid after the operations below
# from the front
ax[0].imshow(np.abs(p).sum(axis=0), extent=ext)
plt.setp(ax[0],
xlim=[-50, 50],
ylim=[-50, 50],
title='front view',
xlabel='y',
ylabel='z')
# from the top
av = np.mean(np.angle(data[com[0], com[1], com[2]]))
data[:] *= np.exp(-1j * av)
av = np.mean(
np.angle(data[com[0] - 2:com[0] + 2, com[1] - 2:com[1] + 2,
com[2] - 2:com[2] + 2, ]))
data[:] *= np.exp(-1j * av)
# manually shift the particle if needed
if WHAT == OUTPUT:
data = np.roll(data, axis=(1, 2), shift=(0, -1))
data *= np.exp(1j * .1)
elif WHAT == REGULAR:
data = np.roll(data, axis=(0, 1, 2), shift=(-1, -1, -1))
# now resample on the orthogonal grid
data, psize = rectify_sample(data, dr, theta, interp=1, find_order=False)
# cut out the good part
N = 18
a, b, c = data.shape
data = data[a // 2 - N // 2:a // 2 + N // 2,
b // 2 - N // 2:b // 2 + N // 2,
c // 2 - N // 2:c // 2 + N // 2, ]
### silx 3D plot
if False:
# Create a SceneWindow widget in an app
app = qt.QApplication([])
window = SceneWindow()
# Get the SceneWidget contained in the window and set its colors
