Skip to content

resampling

systematic_resample(weights)

Systematic resampling with a single random offset.

Generates N equally spaced probe points in [0, 1] with a single random offset between 0 and 1/N. Indices are derived by mapping these points through the inverse CDF of the categorical distribution defined by the weights. Each index i is resampled exactly floor(N * w_i) or ceil(N * w_i) times.

Adapted from FilterPy (R. Labbe): https://filterpy.readthedocs.io/en/latest/monte_carlo/resampling.html

Parameters:

Name Type Description Default
weights array - like

Normalized probability weights summing to 1.

 required

Returns:

Type Description
ndarray

Integer array of ancestor indices.
Source code in genlm/control/sampler/resampling.py 
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
def systematic_resample(weights):
    """Systematic resampling with a single random offset.

    Generates N equally spaced probe points in [0, 1] with a single
    random offset between 0 and 1/N. Indices are derived by mapping
    these points through the inverse CDF of the categorical distribution
    defined by the weights. Each index i is resampled exactly
    floor(N * w_i) or ceil(N * w_i) times.

    Adapted from FilterPy (R. Labbe):
    https://filterpy.readthedocs.io/en/latest/monte_carlo/resampling.html

    Args:
        weights (array-like): Normalized probability weights summing to 1.

    Returns:
        (np.ndarray): Integer array of ancestor indices.
    """
    N = len(weights)
    positions = (np.random.random() + np.arange(N)) / N

    indexes = np.zeros(N, "i")
    cumulative_sum = np.cumsum(weights)
    cumulative_sum[-1] = 1.0  # avoid round-off errors
    i, j = 0, 0
    while i < N:
        if positions[i] < cumulative_sum[j]:
            indexes[i] = j
            i += 1
        else:
            j += 1
    return indexes

stratified_resample(weights)

Stratified resampling with one random draw per stratum.

Divides [0, 1] into N equal strata and draws one uniform point independently within each, so that consecutive points are between 0 and 2/N apart. Indices are derived by mapping these points through the inverse CDF of the categorical distribution defined by the weights.

Adapted from FilterPy (R. Labbe): https://filterpy.readthedocs.io/en/latest/monte_carlo/resampling.html

Parameters:

Name Type Description Default
weights array - like

Normalized probability weights summing to 1.

 required

Returns:

Type Description
ndarray

Integer array of ancestor indices.
Source code in genlm/control/sampler/resampling.py 
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
def stratified_resample(weights):
    """Stratified resampling with one random draw per stratum.

    Divides [0, 1] into N equal strata and draws one uniform point
    independently within each, so that consecutive points are between
    0 and 2/N apart. Indices are derived by mapping these points
    through the inverse CDF of the categorical distribution defined
    by the weights.

    Adapted from FilterPy (R. Labbe):
    https://filterpy.readthedocs.io/en/latest/monte_carlo/resampling.html

    Args:
        weights (array-like): Normalized probability weights summing to 1.

    Returns:
        (np.ndarray): Integer array of ancestor indices.
    """
    N = len(weights)
    positions = (np.random.random(N) + np.arange(N)) / N

    indexes = np.zeros(N, "i")
    cumulative_sum = np.cumsum(weights)
    cumulative_sum[-1] = 1.0
    i, j = 0, 0
    while i < N:
        if positions[i] < cumulative_sum[j]:
            indexes[i] = j
            i += 1
        else:
            j += 1
    return indexes

residual_resample(weights)

Residual resampling: deterministic floor copies + multinomial remainder.

Takes floor(N * w_i) copies of each particle deterministically, then resamples the remaining slots from the fractional residuals using multinomial resampling.

Adapted from FilterPy (R. Labbe): https://filterpy.readthedocs.io/en/latest/monte_carlo/resampling.html

Parameters:

Name Type Description Default
weights array - like

Normalized probability weights summing to 1.

 required

Returns:

Type Description
ndarray

Integer array of ancestor indices.
Source code in genlm/control/sampler/resampling.py 
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
def residual_resample(weights):
    """Residual resampling: deterministic floor copies + multinomial remainder.

    Takes floor(N * w_i) copies of each particle deterministically,
    then resamples the remaining slots from the fractional residuals
    using multinomial resampling.

    Adapted from FilterPy (R. Labbe):
    https://filterpy.readthedocs.io/en/latest/monte_carlo/resampling.html

    Args:
        weights (array-like): Normalized probability weights summing to 1.

    Returns:
        (np.ndarray): Integer array of ancestor indices.
    """
    N = len(weights)
    weights = np.asarray(weights, dtype=float)
    indexes = np.zeros(N, "i")

    # Deterministic copies
    num_copies = np.floor(N * weights).astype(int)
    k = 0
    for i in range(N):
        for _ in range(num_copies[i]):
            indexes[k] = i
            k += 1

    # Multinomial resample on the residual
    if k < N:
        residual = weights * N - num_copies
        residual /= residual.sum()
        cumulative_sum = np.cumsum(residual)
        cumulative_sum[-1] = 1.0
        indexes[k:N] = np.searchsorted(cumulative_sum, np.random.random(N - k))

    return indexes

multinomial_resample(weights)

Multinomial resampling: independent categorical draws.

Each of the N ancestor indices is drawn independently from the categorical distribution defined by the weights.

Parameters:

Name Type Description Default
weights array - like

Normalized probability weights summing to 1.

 required

Returns:

Type Description
ndarray

Integer array of ancestor indices.
Source code in genlm/control/sampler/resampling.py 
113
114
115
116
117
118
119
120
121
122
123
124
125
126
def multinomial_resample(weights):
    """Multinomial resampling: independent categorical draws.

    Each of the N ancestor indices is drawn independently from the
    categorical distribution defined by the weights.

    Args:
        weights (array-like): Normalized probability weights summing to 1.

    Returns:
        (np.ndarray): Integer array of ancestor indices.
    """
    N = len(weights)
    return np.random.choice(N, size=N, replace=True, p=weights)

get_resampling_fn(method)

Get a resampling function by name.

Parameters:

Name Type Description Default
method str

One of 'systematic', 'stratified', 'residual', 'multinomial'.

 required

Returns:

Type Description
callable

Resampling function that takes weights and returns indices.

Raises:

Type Description
ValueError

If method is not recognized.
Source code in genlm/control/sampler/resampling.py 
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
def get_resampling_fn(method):
    """Get a resampling function by name.

    Args:
        method (str): One of 'systematic', 'stratified', 'residual', 'multinomial'.

    Returns:
        (callable): Resampling function that takes weights and returns indices.

    Raises:
        ValueError: If method is not recognized.
    """
    if method not in RESAMPLING_METHODS:
        raise ValueError(
            f"Unknown resampling method '{method}'. "
            f"Must be one of: {', '.join(RESAMPLING_METHODS.keys())}"
        )
    return RESAMPLING_METHODS[method]