pyaibox.ml package

Submodules

pyaibox.ml.dataset_visualization module

pyaibox.ml.dataset_visualization.visds(x, y=None, labelmap=None, axisn=0, mode='bar(nms)', **kwargs)

pyaibox.ml.reduction_pca module

pyaibox.ml.reduction_pca.pca(x, axisn=0, ncmpnts='auto99%', algo='svd')

Principal Component Analysis (pca) on raw data

Parameters

• x (array) – the input data

• axisn (int, optional) – the axis of number of samples, by default 0

• ncmpnts (int or str, optional) – the number of components, by default 'auto99%'

• algo (str, optional) – the kind of algorithms, by default 'svd'

Returns

U, S, K (if ncmpnts is integer)

Return type

array

Examples

The results shown in the above figure can be obtained by the following codes.

import numpy as np
import pyaibox as pb

rootdir, dataset = '/mnt/d/DataSets/oi/dgi/mnist/official/', 'test'
x, _ = pb.read_mnist(rootdir=rootdir, dataset=dataset, fmt='ubyte')
print(x.shape)
N, M2, _ = x.shape

u, s, k = pb.pca(x, axisn=0, ncmpnts='auto90%', algo='svd')
print(u.shape, s.shape, k)
u = u[:, :k]
y = x.reshape(N, -1) @ u  # N-k
z = y @ u.T.conj()
z = z.reshape(N, M2, M2)
print(pb.nmse(x, z, axis=(1, 2)))
xp = np.pad(x[:35], ((0, 0), (1, 1), (1, 1)), 'constant', constant_values=(255, 255))
zp = np.pad(z[:35], ((0, 0), (1, 1), (1, 1)), 'constant', constant_values=(255, 255))
plt = pb.imshow(pb.patch2tensor(xp, (5*(M2+2), 7*(M2+2)), axis=(1, 2)), titles=['Orignal'])
plt = pb.imshow(pb.patch2tensor(zp, (5*(M2+2), 7*(M2+2)), axis=(1, 2)), titles=['Reconstructed' + '(90%)'])

u, s, k = pb.pca(x, axisn=0, ncmpnts='auto0.7', algo='svd')
print(u.shape, s.shape, k)
u = u[:, :k]
y = x.reshape(N, -1) @ u  # N-k
z = y @ u.T.conj()
z = z.reshape(N, M2, M2)
print(pb.nmse(x, z, axis=(1, 2)))
zp = np.pad(z[:35], ((0, 0), (1, 1), (1, 1)), 'constant', constant_values=(255, 255))
plt = pb.imshow(pb.patch2tensor(zp, (5*(M2+2), 7*(M2+2)), axis=(1, 2)), titles=['Reconstructed' + '(70%)'])
plt.show()

u, s = pb.pca(x, axisn=0, ncmpnts=2, algo='svd')
print(u.shape, s.shape)
y = x.reshape(N, -1) @ u  # N-k
z = y @ u.T.conj()
z = z.reshape(N, M2, M2)
print(pb.nmse(x, z, axis=(1, 2)))

Module contents