def get_approx_digit_distribution(n_pca_comp=10,
n_mixtures=5,
covariance_type="spherical") -> Distribution:
"""
Returns an GMM approximation to MNIST.
Args:
n_pca_comp: The number of dimensions to reduce the image size to.
n_mixtures: The number of mixtures to use for each digit.
Returns:
Distribution: An approximate model of mnist.
"""
X, y = load_digits(return_X_y=True)
X = X.astype(np.float) / X.max()
y = np.array([int(v) for v in y])
distributions = []
for i in range(10):
print(f"training model on digit {i}")
dist = train_gmm_pca_model(
X[y == i, :],
n_mixtures=n_mixtures,
n_pca_comp=n_pca_comp,
covariance_type=covariance_type,
)
distributions.append(dist)
mnist_dist = Distribution.mixture(distributions)
mnist_dist.visualize = plot_image_samples([8, 8], False)
mnist_dist.rvs = get_samples(X)
return mnist_dist
def get_mnist_distribution() -> Sampler:
"""
Returns a MNIST sampler.
Returns:
Sampler: Sampling mnist digits
"""
X, y = fetch_openml("mnist_784", version=1, return_X_y=True)
X = 2 * X.astype(np.float) / X.max() - 1
X = X.reshape([-1, 1, 28, 28])
mnist_dist = Sampler.from_samples(
X, noise=lambda shape: 2 * np.random.rand(*shape) / 255)
mnist_dist.visualize = plot_image_samples([28, 28], False)
return mnist_dist
def get_digit_distribution() -> Sampler:
"""
Returns a Digits Sampler.
Returns:
Sampler: Sampling digits
"""
X, y = load_digits(return_X_y=True)
X = 2 * X.astype(np.float) / X.max() - 1
n = X.shape[0]
X = X.reshape([n, 1, 8, 8])
digit_dist = Sampler.from_samples(
X, noise=lambda shape: 2 * np.random.rand(*shape) / 255)
digit_dist.visualize = plot_image_samples([8, 8], False)
return digit_dist
def get_pattern_distribution(patterns: Tuple[str, ...] = (
"checkerboard", )) -> "Sampler":
"""
Returns a pattern sampler.
Args:
patterns: A tuple of patterns to use.
Returns:
Sampler: Sampling from the patterns
"""
pattern_arrays = [PATTERNS[pattern].flatten() for pattern in patterns]
X = np.stack(pattern_arrays, axis=0)
pattern_dist = Sampler.from_samples(X)
pattern_dist.visualize = plot_image_samples([2, 2], False)
return pattern_dist
def get_sm_digit_distribution() -> Sampler:
"""
Returns a Digits Sampler.
Returns:
Sampler: Sampling digits
"""
X, y = load_digits(return_X_y=True)
X = 2 * X.astype(np.float) / X.max() - 1
n = X.shape[0]
X_torch = torch.tensor(X.reshape([n, 1, 8, 8]))
X = torch.nn.AvgPool2d(kernel_size=(2, 2))(X_torch).numpy()
digit_dist = Sampler.from_samples(
X, noise=lambda shape: 2 * np.random.rand(*shape) / 255
)
digit_dist.visualize = plot_image_samples([4, 4], False)
return digit_dist