cartesian_wiener_filter.py 4.96 KB
Newer Older
Theo Steininger's avatar
Theo Steininger committed
1
import numpy as np
Martin Reinecke's avatar
Martin Reinecke committed
2
import nifty4 as ift
Theo Steininger's avatar
Theo Steininger committed
3
4

if __name__ == "__main__":
Martin Reinecke's avatar
Martin Reinecke committed
5
    signal_to_noise = 0.5  # The signal to noise ratio
Theo Steininger's avatar
Theo Steininger committed
6

Martin Reinecke's avatar
Martin Reinecke committed
7
    # Setting up parameters
Martin Reinecke's avatar
Martin Reinecke committed
8
9
10
11
12
13
14
15
16
17
18
19
    L_1 = 2.                   # Total side-length of the domain
    N_pixels_1 = 512           # Grid resolution (pixels per axis)
    L_2 = 2.                   # Total side-length of the domain
    N_pixels_2 = 512           # Grid resolution (pixels per axis)
    correlation_length_1 = 1.
    field_variance_1 = 2.      # Variance of field in position space
    response_sigma_1 = 0.05    # Smoothing length of response
    correlation_length_2 = 1.
    field_variance_2 = 2.      # Variance of field in position space
    response_sigma_2 = 0.01    # Smoothing length of response

    def power_spectrum_1(k):   # note: field_variance**2 = a*k_0/4.
Theo Steininger's avatar
Theo Steininger committed
20
21
22
23
        a = 4 * correlation_length_1 * field_variance_1**2
        return a / (1 + k * correlation_length_1) ** 4.

    signal_space_1 = ift.RGSpace([N_pixels_1], distances=L_1/N_pixels_1)
Martin Reinecke's avatar
Martin Reinecke committed
24
    harmonic_space_1 = signal_space_1.get_default_codomain()
25
26
27
28
29
    signal_space_2 = ift.RGSpace([N_pixels_2], distances=L_2/N_pixels_2)
    harmonic_space_2 = signal_space_2.get_default_codomain()

    signal_domain = ift.DomainTuple.make((signal_space_1, signal_space_2))
    mid_domain = ift.DomainTuple.make((signal_space_1, harmonic_space_2))
Martin Reinecke's avatar
Martin Reinecke committed
30
31
    harmonic_domain = ift.DomainTuple.make((harmonic_space_1,
                                            harmonic_space_2))
32

Martin Reinecke's avatar
Martin Reinecke committed
33
    ht_1 = ift.HarmonicTransformOperator(harmonic_domain, space=0)
Martin Reinecke's avatar
Martin Reinecke committed
34
    power_space_1 = ift.PowerSpace(harmonic_space_1)
Theo Steininger's avatar
Theo Steininger committed
35

Martin Reinecke's avatar
Martin Reinecke committed
36
    mock_power_1 = ift.PS_field(power_space_1, power_spectrum_1)
Theo Steininger's avatar
Theo Steininger committed
37

Martin Reinecke's avatar
Martin Reinecke committed
38
    def power_spectrum_2(k):  # note: field_variance**2 = a*k_0/4.
Theo Steininger's avatar
Theo Steininger committed
39
40
41
        a = 4 * correlation_length_2 * field_variance_2**2
        return a / (1 + k * correlation_length_2) ** 2.5

Martin Reinecke's avatar
Martin Reinecke committed
42
    ht_2 = ift.HarmonicTransformOperator(mid_domain, space=1)
Martin Reinecke's avatar
Martin Reinecke committed
43
    power_space_2 = ift.PowerSpace(harmonic_space_2)
Theo Steininger's avatar
Theo Steininger committed
44

Martin Reinecke's avatar
Martin Reinecke committed
45
    mock_power_2 = ift.PS_field(power_space_2, power_spectrum_2)
Theo Steininger's avatar
Theo Steininger committed
46

Martin Reinecke's avatar
Martin Reinecke committed
47
    ht = ht_2*ht_1
Theo Steininger's avatar
Theo Steininger committed
48

Martin Reinecke's avatar
Martin Reinecke committed
49
50
51
52
53
    mock_power = ift.Field(
        (power_space_1, power_space_2),
        val=ift.dobj.from_global_data(
            np.outer(ift.dobj.to_global_data(mock_power_1.val),
                     ift.dobj.to_global_data(mock_power_2.val))))
Theo Steininger's avatar
Theo Steininger committed
54

Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
55
    diagonal = ift.power_synthesize_nonrandom(mock_power, spaces=(0, 1))**2
Theo Steininger's avatar
Theo Steininger committed
56

57
    S = ift.DiagonalOperator(diagonal)
Theo Steininger's avatar
Theo Steininger committed
58
59

    np.random.seed(10)
Martin Reinecke's avatar
Martin Reinecke committed
60
    mock_signal = ift.power_synthesize(mock_power, real_signal=True)
Theo Steininger's avatar
Theo Steininger committed
61
62
63

    # Setting up a exemplary response
    N1_10 = int(N_pixels_1/10)
Martin Reinecke's avatar
Martin Reinecke committed
64
65
66
    mask_1 = np.ones(signal_space_1.shape)
    mask_1[N1_10*7:N1_10*9] = 0.
    mask_1 = ift.Field(signal_space_1, ift.dobj.from_global_data(mask_1))
Theo Steininger's avatar
Theo Steininger committed
67
68

    N2_10 = int(N_pixels_2/10)
Martin Reinecke's avatar
Martin Reinecke committed
69
70
71
    mask_2 = np.ones(signal_space_2.shape)
    mask_2[N2_10*7:N2_10*9] = 0.
    mask_2 = ift.Field(signal_space_2, ift.dobj.from_global_data(mask_2))
Theo Steininger's avatar
Theo Steininger committed
72

Martin Reinecke's avatar
Martin Reinecke committed
73
74
75
76
77
78
79
80
    R = ift.GeometryRemover(signal_domain)
    R = R*ift.DiagonalOperator(mask_1, signal_domain,spaces=0)
    R = R*ift.DiagonalOperator(mask_2, signal_domain,spaces=1)
    R = R*ht
    R = R * ift.create_harmonic_smoothing_operator(harmonic_domain, 0,
                                                   response_sigma_1)
    R = R * ift.create_harmonic_smoothing_operator(harmonic_domain, 1,
                                                   response_sigma_2)
Theo Steininger's avatar
Theo Steininger committed
81
82
    data_domain = R.target

Martin Reinecke's avatar
Martin Reinecke committed
83
84
    noiseless_data = R(mock_signal)
    noise_amplitude = noiseless_data.std()/signal_to_noise
Theo Steininger's avatar
Theo Steininger committed
85
    # Setting up the noise covariance and drawing a random noise realization
Martin Reinecke's avatar
Martin Reinecke committed
86
87
    ndiag = ift.Field.full(data_domain, noise_amplitude**2)
    N = ift.DiagonalOperator(ndiag)
Martin Reinecke's avatar
Martin Reinecke committed
88
89
    noise = ift.Field.from_random(
        domain=data_domain, random_type='normal',
Martin Reinecke's avatar
Martin Reinecke committed
90
91
        std=noise_amplitude, mean=0)
    data = noiseless_data + noise
Theo Steininger's avatar
Theo Steininger committed
92
93

    # Wiener filter
Martin Reinecke's avatar
Martin Reinecke committed
94
    j = R.adjoint_times(N.inverse_times(data))
95
    ctrl = ift.GradientNormController(name="inverter", tol_abs_gradnorm=0.1)
96
    inverter = ift.ConjugateGradient(controller=ctrl)
97
    wiener_curvature = ift.library.WienerFilterCurvature(
Martin Reinecke's avatar
Martin Reinecke committed
98
        S=S, N=N, R=R, inverter=inverter)
Theo Steininger's avatar
Theo Steininger committed
99

Martin Reinecke's avatar
Martin Reinecke committed
100
    m_k = wiener_curvature.inverse_times(j)
Martin Reinecke's avatar
Martin Reinecke committed
101
    m = ht(m_k)
Theo Steininger's avatar
Theo Steininger committed
102

Martin Reinecke's avatar
Martin Reinecke committed
103
104
    plotdict = {"xlabel": "Pixel index", "ylabel": "Pixel index",
                "colormap": "Planck-like"}
Martin Reinecke's avatar
Martin Reinecke committed
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
    plot_space = ift.RGSpace((N_pixels_1, N_pixels_2))
    ift.plot(ift.Field(plot_space,val=ht(mock_signal).val), name='mock_signal.png',
             **plotdict)
    ift.plot(ift.Field(plot_space,val=data.val), name='data.png', **plotdict)
    ift.plot(ift.Field(plot_space,val=m.val), name='map.png', **plotdict)
    # sampling the uncertainty map
    sample_variance = ift.Field.zeros(signal_domain)
    sample_mean = ift.Field.zeros(signal_domain)
    n_samples = 10
    for i in range(n_samples):
        sample = ht(wiener_curvature.generate_posterior_sample()) + m
        sample_variance += sample**2
        sample_mean += sample
    variance = sample_variance/n_samples - (sample_mean/n_samples)**2
    ift.plot(ift.Field(plot_space, val=ift.sqrt(variance).val), name="uncertainty.png", **plotdict)