官术网_书友最值得收藏!

Singular value decomposition

When we decompose an integer into its prime factors, we can understand useful properties about the integer. Similarly, when we decompose a matrix, we can understand many functional properties that are not directly evident. There are two types of decomposition, namely eigenvalue decomposition and singular value decomposition.

All real matrices have singular value decomposition, but the same is not true for Eigenvalue decomposition. For example, if a matrix is not square, the Eigen decomposition is not defined and we must use singular value decomposition instead.

Singular Value Decomposition (SVD) in mathematical form is the product of three matrices U, S, and V, where U is m*r, S is r*r and V is r*n:

The following example shows SVD using a TensorFlow svd operation on textual data:

import numpy as np
import tensorflow as tf
import matplotlib.pyplot as plts

path = "/neuralnetwork-programming/ch01/plots"

text = ["I", "like", "enjoy",
"deep", "learning", "NLP", "flying", "."]
xMatrix = np.array([[0,2,1,0,0,0,0,0],
[2,0,0,1,0,1,0,0],
[1,0,0,0,0,0,1,0],
[0,1,0,0,1,0,0,0],
[0,0,0,1,0,0,0,1],
[0,1,0,0,0,0,0,1],
[0,0,1,0,0,0,0,1],
[0,0,0,0,1,1,1,0]], dtype=np.float32)

X_tensor = tf.convert_to_tensor(xMatrix, dtype=tf.float32)

# tensorflow svd
with tf.Session() as sess:
s, U, Vh = sess.run(tf.svd(X_tensor, full_matrices=False))

for i in range(len(text)):
plts.text(U[i,0], U[i,1], text[i])

plts.ylim(-0.8,0.8)
plts.xlim(-0.8,2.0)
plts.savefig(path + '/svd_tf.png')

# numpy svd
la = np.linalg
U, s, Vh = la.svd(xMatrix, full_matrices=False)

print(U)
print(s)
print(Vh)

# write matrices to file (understand concepts)
file = open(path + "/matx.txt", 'w')
file.write(str(U))
file.write("\n")
file.write("=============")
file.write("\n")
file.write(str(s))
file.close()

for i in range(len(text)):
plts.text(U[i,0], U[i,1], text[i])

plts.ylim(-0.8,0.8)
plts.xlim(-0.8,2.0)
plts.savefig(path + '/svd_np.png')

The output of this is shown as follows:

[[ -5.24124920e-01  -5.72859168e-01   9.54463035e-02   3.83228481e-01   -1.76963374e-01  -1.76092178e-01  -4.19185609e-01  -5.57702743e-02]
[ -5.94438076e-01 6.30120635e-01 -1.70207784e-01 3.10038358e-0
1.84062332e-01 -2.34777853e-01 1.29535481e-01 1.36813134e-01]
[ -2.56274015e-01 2.74017543e-01 1.59810841e-01 3.73903001e-16
-5.78984618e-01 6.36550903e-01 -3.32297325e-16 -3.05414885e-01]
[ -2.85637408e-01 -2.47912124e-01 3.54610324e-01 -7.31901303e-02
4.45784479e-01 8.36141407e-02 5.48721075e-01 -4.68012422e-01]
[ -1.93139315e-01 3.38495038e-02 -5.00790417e-01 -4.28462476e-01
3.47110212e-01 1.55483231e-01 -4.68663752e-01 -4.03576553e-01]
[ -3.05134684e-01 -2.93989003e-01 -2.23433599e-01 -1.91614240e-01
1.27460942e-01 4.91219401e-01 2.09592804e-01 6.57535374e-01]
[ -1.82489842e-01 -1.61027774e-01 -3.97842437e-01 -3.83228481e-01
-5.12923241e-01 -4.27574426e-01 4.19185609e-01 -1.18313827e-01]
[ -2.46898428e-01 1.57254755e-01 5.92991650e-01 -6.20076716e-01
-3.21868137e-02 -2.31065080e-01 -2.59070963e-01 2.37976909e-01]]
[ 2.75726271 2.67824793 1.89221275 1.61803401 1.19154561 0.94833982
0.61803401 0.56999218]
[[ -5.24124920e-01 -5.94438076e-01 -2.56274015e-01 -2.85637408e-01
-1.93139315e-01 -3.05134684e-01 -1.82489842e-01 -2.46898428e-01]
[ 5.72859168e-01 -6.30120635e-01 -2.74017543e-01 2.47912124e-01
-3.38495038e-02 2.93989003e-01 1.61027774e-01 -1.57254755e-01]
[ -9.54463035e-02 1.70207784e-01 -1.59810841e-01 -3.54610324e-01
5.00790417e-01 2.23433599e-01 3.97842437e-01 -5.92991650e-01]
[ 3.83228481e-01 3.10038358e-01 -2.22044605e-16 -7.31901303e-02
-4.28462476e-01 -1.91614240e-01 -3.83228481e-01 -6.20076716e-01]
[ -1.76963374e-01 1.84062332e-01 -5.78984618e-01 4.45784479e-01
3.47110212e-01 1.27460942e-01 -5.12923241e-01 -3.21868137e-02]
[ 1.76092178e-01 2.34777853e-01 -6.36550903e-01 -8.36141407e-02
-1.55483231e-01 -4.91219401e-01 4.27574426e-01 2.31065080e-01]
[ 4.19185609e-01 -1.29535481e-01 -3.33066907e-16 -5.48721075e-01
4.68663752e-01 -2.09592804e-01 -4.19185609e-01 2.59070963e-01]
[ -5.57702743e-02 1.36813134e-01 -3.05414885e-01 -4.68012422e-01
-4.03576553e-01 6.57535374e-01 -1.18313827e-01 2.37976909e-01]]

Here is the plot for the SVD of the preceding dataset:

主站蜘蛛池模板: 泰顺县| 罗平县| 沙坪坝区| 泾源县| 全州县| 浑源县| 平安县| 吉安市| 昌黎县| 安龙县| 福安市| 广宗县| 济源市| 海林市| 林周县| 霍林郭勒市| 孙吴县| 常德市| 玛曲县| 昌平区| 文山县| 屯门区| 闻喜县| 无极县| 营口市| 瑞安市| 瓦房店市| 房产| 呼和浩特市| 嫩江县| 定州市| 辛集市| 林周县| 双峰县| 汉川市| 吐鲁番市| 积石山| 彰化县| 都江堰市| 巴塘县| 淮阳县|