python 生成随机稀疏矩阵

生成随机矩阵

import numpy as np
import scipy.sparse as ss

# 生成随机稀疏矩阵
num_col = 20
num_row = 10
num_ele = 40
a = [np.random.randint(0,num_row) for _ in range(num_ele)]
b = [np.random.randint(0,num_col) for _ in range(num_ele-num_col)] + [i for i in range(num_col)]  # 保证每一列都有值，不会出现全零列
c = [np.random.rand() for _ in range(num_ele)]
rows, cols, v = np.array(a), np.array(b), np.array(c)
sparseX = ss.coo_matrix((v,(rows,cols)))
X = sparseX.todense()

fig = plt.figure()
ax = fig.add_subplot(111)
ax.matshow(X, vmin=0, vmax=1, cmap='magma')
ax.set_xticks([])
ax.set_yticks([])
plt.show()

常用 colormap

生成稀疏编码信号

from sklearn.datasets import make_sparse_coded_signal

n_samples, n_components, n_features = 5, 10, 8
n_nonzero_coefs = 3

# Y = DX
# |x_i|_0 = n_nonzero_coefs

Y, D, X = make_sparse_coded_signal(n_samples=n_samples,
                                   n_components=n_components,
                                   n_features=n_features,
                                   n_nonzero_coefs=n_nonzero_coefs,
                                   random_state=42)

这是字典学习中处理的问题：
$$Y=DX$$
$X$ 为原始信号 $Y$ 的稀疏编码。