Skip to content

phepy.toys.haystack

HaystackToyExample 

HaystackToyExample(
    rng: np.random.Generator, N: int = 10000
)

Bases: ToyExample

Source code in phepy/toys/haystack.py 
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
38
39
40
41
42
43
44
45
46
47
48
49
50
51
def __init__(
    self, rng: np.random.Generator, N: int = 10_000
) -> HaystackToyExample:
    N *= 2

    n = 10
    a = 2

    # https://stats.stackexchange.com/a/124554
    A = np.matrix(
        [rng.normal(size=n) + rng.normal(size=1) * a for i in range(n)]
    )
    A = A * np.transpose(A)
    D_half = np.diag(np.diag(A) ** (-0.5))
    covs = D_half * A * D_half

    X = rng.multivariate_normal(np.zeros(shape=n), covs, size=N)
    X[:, 3] = -0.42

    # Since X[:,3] is const, it does not matter what its multiplier is
    C = rng.choice([-2, -1, 0, 0, 1, 2], size=n)

    Y = np.dot(X, C)

    I_train = rng.choice(N, size=N // 2, replace=False)
    I_valid = np.ones(N)
    I_valid[I_train] = 0
    (I_valid,) = np.nonzero(I_valid)

    self.__X_train = X[I_train]
    self.__Y_train = Y[I_train]
    self.__X_valid = X[I_valid]
    self.__Y_valid = Y[I_valid]

    XT = rng.multivariate_normal(np.zeros(shape=n), covs, size=N // 2)
    XT[:, 3] = (rng.random(size=N // 2) - 0.5) - 0.42
    # Provide some definitely ID points
    XT[np.random.choice(N // 2, size=N // 200), 3] = -0.42

    self.__X_test = XT

X_test property 

X_test: np.ndarray

X_train property 

X_train: np.ndarray

X_valid property 

X_valid: np.ndarray

Y_train property 

Y_train: np.ndarray

Y_valid property 

Y_valid: np.ndarray

aspect_ratio property 

aspect_ratio: float

is_in_distribution 

is_in_distribution(X: np.ndarray) -> np.ndarray
Source code in phepy/toys/haystack.py 
79
80
81
def is_in_distribution(self, X: np.ndarray) -> np.ndarray:
    # Can only be ID if the constant feature value matches
    return X[:, 3] == -0.42

plot 

plot(
    conf: np.ndarray,
    ax: mpl.axes.Axes,
    cmap: Union[str, mpl.colors.Colormap],
    with_scatter: bool = True,
)
Source code in phepy/toys/haystack.py 
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
def plot(
    self,
    conf: np.ndarray,
    ax: mpl.axes.Axes,
    cmap: Union[str, mpl.colors.Colormap],
    with_scatter: bool = True,
):
    ax.spines["top"].set_visible(False)
    ax.spines["right"].set_visible(False)
    ax.spines["bottom"].set_visible(False)
    ax.spines["left"].set_visible(False)

    ax.set_xticks([-0.42 - 0.4, -0.42, -0.42 + 0.4])
    ax.set_xticklabels([r"$-0.4 \sigma$", "const", r"$+0.4 \sigma$"])
    ax.get_yaxis().set_visible(False)

    ax.axvline(-0.42, c="white", lw=7, zorder=-1)

    # with_scatter is ignored
    ax.scatter(self.X_test[:, 3], conf, c="white", s=6, rasterized=True)
    ax.scatter(self.X_test[:, 3], conf, c="black", s=1, rasterized=True)

    ax.imshow(
        [[x / 100, x / 100] for x in range(100, -1, -1)],
        cmap=cmap,
        interpolation="bicubic",
        vmin=0.0,
        vmax=1.0,
        extent=[-0.93, 0.09, -0.01, 1.01],
        zorder=-2,
    )

reconstruct 

reconstruct(X: np.ndarray) -> np.ndarray
Source code in phepy/toys/haystack.py 
73
74
75
76
77
def reconstruct(self, X: np.ndarray) -> np.ndarray:
    # Project X onto the line
    return X * np.array(
        [[1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]]
    ) - np.array([[0.0, 0.0, 0.0, 0.42, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]])