python-几种实现SVD的方式和区别

from sklearn.utils.extmath import randomized_svd
from scipy.sparse.linalg import svds
import numpy as np

# data.shape(6, 4)
data = np.array([[np.nan, 3, np.nan, 4],
                     [3, np.nan, 4, np.nan],
                     [np.nan, np.nan, 2, np.nan],
                     [5, np.nan, 3, 4],
                     [np.nan, np.nan, 4, np.nan],
                     [np.nan, 3, 3, np.nan]])
data[np.isnan(data)] = 0

u1,s1,v1=randomized_svd(data,4)    #  u1.shape(6, 4), s1.shape(4,),  v1.shape(4, 4)
# s1:array([9.16751646, 5.46594686, 4.61133842, 1.67798238])

u2,s2,v2=svds(data,3)    # 第二个参数k must be between 1 and min(data.shape)
# u2.shape(6, 3), s2.shape(3,),  v2.shape(3, 4)
# s2:array([4.61133842, 5.46594686, 9.16751646])

u3,s3,v3=np.linalg.svd(data)    # u3.shape(6, 6), s3.shape(4,),  v3.shape(4, 4)
# s3:array([9.16751646, 5.46594686, 4.61133842, 1.67798238])

除了以上三种方式，还有sklearn.decomposition下的TruncatedSVD，具体内容自行查看手册。如下为手册中的例子。使用方式与以上几种方法稍有不同。

from sklearn.decomposition import TruncatedSVD
from scipy.sparse import csr_matrix
import numpy as np
np.random.seed(0)
X_dense = np.random.rand(100, 100)
X_dense[:, 2 * np.arange(50)] = 0
X = csr_matrix(X_dense)
svd = TruncatedSVD(n_components=5, n_iter=7, random_state=42)
svd.fit(X)

print(svd.explained_variance_ratio_)

print(svd.explained_variance_ratio_.sum())

print(svd.singular_values_)